Hybrid RSM–SF-AHP–fuzzy MARCOS approach for multi-response optimization of WAAM-fabricated Ni–SS bimetallics in face milling Authors: S. P. Sundar Singh Sivam, Stalin Kesavan, A. Johnson Santhosh Published: 2026 ABSTRACT This study develops Wire Arc Additively Manufactured (WAAM) Nickel–Stainless Steel bimetallic samples with silicon enhancement and investigates their face-milling machinability using a hybrid multi-criteria optimization framework. Seventeen machining trials were designed and conducted using a Response Surface Methodology (RSM)-based Central Composite Design (CCD) to evaluate the effects of cutting speed (8000–9500 rpm), feed rate (0.1–0.25 mm/tooth), depth of cut (0.5–1.5 mm), and tool flute count (two and four flutes) on surface roughness (Ra), material removal rate (MRR), power consumption (Pc), and cutting force (CF). A three-stage hybrid methodology was implemented, in which RSM modelled machinability behaviour and identified significant process parameters, Spherical Fuzzy AHP (SF-AHP) assigned criteria weights under uncertainty, and Fuzzy MARCOS ranked the machining alternatives. Run 1 (9500 rpm, 0.1 mm/tooth, 1.5 mm, four flutes) was identified as the optimal setting, exhibiting lower surface roughness (~ 0.6 μm) and higher MRR (~ 35.6 mm³/min), whereas Run 6 (8750 rpm, 0.175 mm/tooth, 1 mm, two flutes) showed comparatively poorer performance with higher Ra (~ 0.8 μm), reduced MRR (~ 34 mm³/min), and less favorable power consumption and cutting force, indicating weaker overall machinability. Cutting speed and depth of cut strongly influenced Ra and MRR, while feed rate and flute configuration primarily affected Pc and CF. Validation through confirmation experiments, criteria-weight verification, and ranking consistency demonstrated strong agreement between predicted and measured performance. Although limited to a specific silicon-enhanced Ni–SS bimetallic system and finite experimental range, the proposed framework provides a practical and robust strategy for multi-response machinability optimization of WAAM components, offering industrial benefits such as improved surface finish, reduced power consumption, enhanced productivity, and better machining stability. Introduction The growing demand for high-performance materials in aerospace, automotive, marine, and energy-related industries has intensified the search for manufacturing routes capable of producing structurally reliable components with high material efficiency and reduced production cost1. In this context, wire arc additive manufacturing (WAAM) has emerged as a promising metal additive manufacturing technology due to its high deposition rate, low material wastage, and suitability for fabricating large-scale near-net-shape components2. Compared with powder-based additive manufacturing processes, WAAM offers significant advantages in terms of productivity, scalability, and economic feasibility, particularly for medium- to large-sized metallic parts3. Among the materials considered in the present study, nickel alloy and stainless steel (SS) are widely used parent materials in advanced engineering applications. Nickel alloys are extensively employed in aerospace engines, turbines, marine systems, and energy equipment because of their excellent high-temperature strength, oxidation resistance, and corrosion resistance, although they are associated with high cost and machining difficulty. Stainless steel is broadly used in structural, automotive, marine, and process industries owing to its good mechanical properties, corrosion resistance, formability, and comparatively lower cost; however, it is susceptible to work hardening and offers relatively lower high-temperature capability than nickel alloys. Owing to these machining characteristics, both materials are of considerable relevance for the present investigation. A substantial proportion of existing WAAM research has focused on process development, heat transfer behavior, deposition stability, microstructural evolution, and bead geometry control4,5. Several studies have examined the influence of heat input, arc characteristics, and deposition strategy on the structural quality of WAAM-fabricated parts6. Other investigations have reported that deposition path design, process monitoring, and bead overlapping play an important role in improving dimensional consistency and reducing process-related defects7. Additional work has explored hybrid or modified WAAM approaches to enhance structural integrity, reduce distortion, and improve mechanical performance8. These contributions have significantly advanced the understanding of WAAM as a fabrication route. However, they have largely concentrated on deposition-stage optimization rather than the post-processing behavior of the fabricated components. Post-processing is a critical stage in the industrial implementation of WAAM because near-net-shape deposition alone is rarely sufficient to achieve the dimensional precision, surface integrity, and functional tolerances required in engineering applications9. Machining is therefore often necessary to produce the final geometry and surface condition of WAAM-fabricated parts10. In practice, machinability directly affects tool wear, surface roughness, force generation, specific energy demand, and manufacturing efficiency11. Consequently, a material system that performs well during deposition may still encounter practical limitations if its post-processing machinability is poor. Despite this important industrial consideration, the machining behavior of WAAM-fabricated materials remains less thoroughly studied than their deposition characteristics. This limitation becomes more significant when bimetallic materials are considered. Bimetallic structures are attractive because they enable the integration of complementary properties from two dissimilar alloys within a single component12. Such materials can be designed to combine corrosion resistance, high-temperature capability, toughness, strength, and wear resistance according to service requirements13. Nickel-based alloys and stainless steels are particularly relevant in this regard because both are widely used in harsh operating environments. Nickel-based alloys are valued for their superior thermal stability, oxidation resistance, and mechanical performance at elevated temperatures14. Stainless steels are widely applied because of their corrosion resistance, structural reliability, and broad industrial usability15. A Nickel–Stainless Steel bimetallic system therefore offers strong potential for multi-functional applications where both durability and environmental resistance are required. However, the same material combination that provides attractive service properties may also create difficulties during machining. Nickel-based alloys are often associated with work hardening, high cutting resistance, and elevated thermal load during machining16. Stainless steels can also present machining challenges due to strain hardening, adhesion, and variability in chip formation under unfavorable cutting conditions17. When these two materials are integrated into a bimetallic structure, the resulting interface and compositional gradient may generate local variations in hardness, thermal conductivity, microstructure, and deformation behavior18. As a result, the cutting tool may experience changing contact conditions as it moves across different regions of the material. This can affect force generation, power consumption, material removal stability, and final surface quality. Therefore, the machinability of WAAM-fabricated Nickel–Stainless Steel bimetallics cannot be assumed from the behavior of the constituent metals alone and requires direct investigation. The role of silicon addition further increases the complexity of this material system. Silicon is known to influence solidification behavior, microstructural evolution, and phase characteristics in metallic alloys19. In bimetallic WAAM structures, such compositional modification may alter local hardness distribution, interfacial response, and resistance to material removal during machining20. These effects are especially relevant in face milling, where repeated interrupted cutting and multi-point tool engagement can amplify the influence of heterogeneous material behavior21. Even though silicon-containing Ni–SS bimetallic systems may offer useful performance benefits, their face-milling response remains insufficiently understood. This knowledge gap restricts the ability to select suitable cutting parameters for reliable and efficient finishing operations. Current literature also reveals an imbalance in research emphasis. Existing studies on WAAM materials predominantly address fabrication quality, metallurgical characterization, and mechanical behavior22. While these are essential aspects, fewer studies have focused on how WAAM-fabricated materials perform during secondary machining operations23. This shortage is even more evident for bimetallic WAAM materials, where material heterogeneity can substantially influence machinability24. Most available studies also concentrate on single-material deposits, meaning that the machining behavior of dissimilar WAAM bimetallics remains comparatively underexplored25. Accordingly, there is a clear need for studies that move beyond deposition optimization and directly examine the machining performance of complex WAAM-fabricated bimetallic systems. Another important limitation in the literature concerns the analytical framework used to evaluate machining performance. Machinability is inherently multi-response in nature. In most machining environments, surface roughness, material removal rate, cutting force, and power consumption do not improve simultaneously under the same cutting condition26. A parameter combination that increases productivity may adversely affect surface quality or energy demand. Likewise, a condition that minimizes force may not maximize removal rate. This creates a multi-criteria decision problem in which trade-offs must be managed carefully. Conventional one-factor or single-response approaches are not sufficient for selecting an overall best machining condition when several performance measures must be considered together27. Therefore, a more robust methodology is required to analyze the interaction between cutting parameters and to support rational decision-making among competing responses. Response Surface Methodology (RSM) is well established as a useful statistical tool for modeling and analyzing manufacturing processes involving multiple interacting variables28. It enables the development of predictive equations, quantification of factor significance, and evaluation of nonlinear effects and interaction behavior29. In machining studies, RSM is particularly valuable because it provides a structured experimental design framework while also enabling response prediction within the selected factor space30. However, predictive modeling alone does not fully resolve the optimization problem when several responses must be balanced simultaneously. A statistical model may identify influential parameters and response trends, but it does not automatically determine the most suitable experimental alternative when the criteria are conflicting. For this reason, fuzzy multi-criteria decision-making tools are increasingly relevant in advanced manufacturing analysis31. Fuzzy approaches are useful when decision problems involve uncertainty, ambiguity, and subjective weighting of performance criteria32. In practical machining optimization, it is often necessary to assign relative importance to responses such as surface roughness, productivity, cutting force, and power requirement33. Such weighting is difficult to perform using purely deterministic methods when expert judgment and uncertainty are involved. Spherical Fuzzy AHP (SF-AHP) is particularly suitable in this context because it extends classical pairwise comparison by providing greater flexibility in representing uncertainty and hesitation in human decision-making34. As a result, SF-AHP offers a stronger basis for criteria weighting in complex engineering optimization problems where multiple responses must be evaluated simultaneously35. Despite these methodological advances, the integration of predictive statistical modeling with uncertainty-aware fuzzy weighting remains limited in the machinability analysis of WAAM-fabricated bimetallic materials36. Many studies apply statistical tools without incorporating a rigorous multi-criteria decision framework37. Other studies employ ranking approaches without sufficiently modeling the nonlinear response behavior of the machining process38. This separation reduces the effectiveness of the analysis, particularly in heterogeneous materials where both process-response prediction and multi-criteria prioritization are important. For WAAM-fabricated Ni–SS bimetallic systems with silicon addition, such an integrated framework is especially necessary because machining performance depends not only on cutting conditions but also on material heterogeneity and the balance among competing response variables. Based on the foregoing discussion, three key research gaps can be identified. First, although WAAM has been extensively studied for single-metal deposits, the face-milling machinability of WAAM-fabricated Nickel–Stainless Steel bimetallic materials remains insufficiently investigated39. Second, the influence of silicon addition and bimetallic heterogeneity on key machining responses, including surface roughness, material removal rate, cutting force, and power consumption, has not been adequately clarified40. Third, previous studies have rarely combined response modeling and fuzzy multi-criteria weighting within a unified methodology for optimizing the machinability of such materials41. These limitations are important because insufficient understanding of machining behavior can reduce process reliability, increase finishing cost, and constrain the broader industrial adoption of WAAM-fabricated bimetallic components. This study introduces a novel sequential framework integrating Response Surface Methodology (RSM), Spherical Fuzzy Analytic Hierarchy Process (SF-AHP), and Fuzzy MARCOS for the multi-response machinability optimization of silicon-enhanced WAAM-fabricated Ni–SS bimetallics. The integration of RSM, SF-AHP, and Fuzzy MARCOS was selected based on the specific needs of this multi-response machining problem. RSM offers statistical modeling of nonlinear response behavior and identifies significant effects of cutting speed, feed rate, depth of cut, and flute number on responses such as surface roughness (Ra), material removal rate (MRR), power consumption (Pc), and cutting force (CF). However, RSM alone does not assign relative importance to conflicting responses. Therefore, SF-AHP was used to determine criteria weights under uncertainty, accounting for membership, non-membership, and hesitation in expert judgment. Fuzzy MARCOS was then applied to rank machining alternatives using utility degrees relative to ideal and anti-ideal solutions. Compared to fuzzy TOPSIS, which relies on geometric distance from ideal solutions, and VIKOR, which focuses on compromise ranking, Fuzzy MARCOS provides a more direct utility-based ranking structure. This hybrid framework is particularly suitable for the WAAM Ni–SS bimetallic machining problem, as it combines predictive modeling, uncertainty-aware criteria prioritization, and ranking under conflicting responses, thus advancing multi-response optimization in advanced manufacturing. Materials and methods Sample Fabrication Fabrication workflow for the bimetallic sample Fig. 1Fabrication workflow for the bimetallic sample: (a) control computer, (b) bimetallic build design (layer sequence), (c) 316LSi filler wire, (d) ERNiCrMo-3 filler wire, (e) CMT welding power source and wire-feeding unit, (f) bimetallic deposition setup/fixture, (g) robotic CMT deposition system, and (h) fabricated bimetallic sample. Fabrication workflow for the bimetallic sample: (a) control computer, (b) bimetallic build design (layer sequence), (c) 316LSi filler wire, (d) ERNiCrMo-3 filler wire, (e) CMT welding power source and wire-feeding unit, (f) bimetallic deposition setup/fixture, (g) robotic CMT deposition system, and (h) fabricated bimetallic sample. Figure 1(a) Control Computer, The control computer served as the central processing unit for deposition planning and parameter synchronization. It regulated welding current, travel speed, shielding gas flow, and motion coordination between the servo-driven XY stage and ABB robotic arm, ensuring precise path execution, stable arc behavior, and repeatable multilayer deposition. Figure 1(b) Bimetallic Build Design (Layer Sequence), The bimetallic build design followed a controlled alternating layer deposition strategy to achieve metallurgical integration between 316LSi stainless steel and ERNiCrMo-3 nickel alloy. The structure was fabricated using a sequential stacking pattern in which each stainless steel layer was followed by a nickel-based alloy layer, forming a periodic compositional architecture along the build height. This alternating sequence promoted gradual compositional transition, minimized abrupt thermal mismatch, and enhanced interfacial bonding. The deposition path was programmed to maintain parallel bead alignment across successive layers, ensuring geometric uniformity and consistent heat distribution. A total of 60 deposition passes were executed to construct a square-wall geometry (50 × 50 × 4 mm). The layer pattern was optimized to reduce residual stresses, suppress intermetallic formation, and maintain structural continuity. Inter-pass cooling of 1 min was applied to control thermal accumulation and stabilize microstructural evolution. This patterned flow ensured compositional homogeneity, controlled dilution at the interface, and robust metallurgical bonding suitable for mechanical and machinability evaluation. Figure 1(c) 316LSi Filler Wire, The 316LSi filler wire of 1.2 mm diameter was used as the stainless steel constituent in the bimetallic structure. Its nominal composition includes approximately 16–18% Cr, 10–14% Ni, 2–3% Mo, ≤ 0.08% C, 0.8–1.0% Si, with Fe as the balance. The silicon addition enhances melt fluidity and bead wetting characteristics, improving layer uniformity during WAAM deposition. The alloy provides excellent corrosion resistance, good weldability, and stable austenitic microstructure, making it suitable for integration with nickel-based alloys in bimetallic architectures. Figure 1(d) ERNiCrMo-3 Filler Wire, The ERNiCrMo-3 nickel-based filler wire of 1.2 mm diameter served as the high-performance alloy layer in the bimetallic system. Its typical composition includes Ni (balance), 20–23% Cr, 8–10% Mo, 3–4% Nb + Ta, ≤ 0.1% C, and minor Fe content. This composition provides superior corrosion resistance, high-temperature strength, and excellent metallurgical compatibility with stainless steel. The presence of molybdenum and chromium enhances pitting resistance and structural stability, while niobium contributes to solid-solution strengthening and crack resistance during thermal cycling. Figure 1(e) CMT Welding Power Source and Wire-Feeding Unit. The Cold Metal Transfer (CMT) welding system comprised a digitally controlled power source, wire feeder, roller assembly, and argon shielding unit. Welding current was varied between 100 and 140 A, travel speed between 4.5 and 7.5 mm/s, and shielding gas flow rate between 14 and 18 L/min. The CMT process ensured controlled heat input, reduced spatter, and stable arc characteristics. Real-time monitoring via WeldEye software maintained consistent deposition conditions and synchronized wire feed with torch motion, enabling high-quality bimetallic layer formation. Figure 1(f) Bimetallic Deposition Setup/Fixture, The deposition fixture securely mounted the stainless steel substrate plate (125 × 100 × 4 mm) during fabrication. The setup ensured geometric stability, vibration control, and precise torch-to-workpiece distance of 15 mm at a 90° orientation. The fixture was integrated with a servo motor–driven XY stage equipped with encoders for accurate linear positioning. This configuration maintained consistent bead geometry, minimized misalignment, and supported repeatable multilayer stacking under controlled thermal conditions. Figure 1(g) Robotic CMT Deposition System, The robotic deposition system utilized an ABB industrial robotic arm integrated with the CMT torch for multi-axis motion control. The robotic arm ensured uniform travel speed, accurate layer placement, and consistent heat distribution across the build surface. The system was synchronized with the control computer and encoder-based servo stage to achieve high positional accuracy and repeatability. This integration enabled precise alternating deposition of 316LSi and ERNiCrMo-3 layers while maintaining stable arc dynamics and minimizing thermal distortion. Figure 1(h) Fabricated Bimetallic Sample, The final WAAM-fabricated bimetallic structure exhibited dense multilayer stacking with uniform bead morphology and strong metallurgical bonding between nickel and stainless steel layers. A total of 60 deposition passes produced a 50 × 50 × 4 mm square-wall geometry. Controlled inter-pass cooling (1 min interval) and monitored heat input ensured minimized residual stresses and homogeneous microstructural evolution. The fabricated sample demonstrated structural integrity and dimensional stability, making it suitable for subsequent mechanical characterization and machinability analysis. Face milling operations for ranking alternatives Fig. 2Experimental setup of Face milling, (a) Fabricated sample from WAAM, (b) 2 Flute, (c) 4 Flute, (d) Tool Nomenclature, (e) Machine used for Milling, (f) Enlarged view of the sample, (g) Face milled sample. Experimental setup of Face milling, (a) Fabricated sample from WAAM, (b) 2 Flute, (c) 4 Flute, (d) Tool Nomenclature, (e) Machine used for Milling, (f) Enlarged view of the sample, (g) Face milled sample. Figure 2: Experimental setup for face milling. The figure presents various stages of the face milling process including the fabricated sample, milling tools, machine setup, and the final milled sample. Figure 2(a) Fabricated sample from WAAM: The sample shown was fabricated using Wire Arc Additive Manufacturing (WAAM), which is an additive manufacturing process that uses an electric arc to melt wire feedstock and build up the material layer by layer. This method is commonly employed for fabricating complex metallic structures, offering benefits such as reduced material waste, high deposition rates, and the ability to work with high-strength alloys. The sample has undergone preparation for milling, with its surface exhibiting distinct layers typical of WAAM, making it suitable for high-precision machining processes like face milling. Figure 2(b) 2-Flute: The 2-flute cutter shown is designed for efficient material removal in face milling operations. It features two cutting edges that provide a balance between cutting efficiency and tool life. This tool type is particularly suitable for less aggressive milling, offering a smooth finish on softer materials and providing good chip clearance. The two-flute cutter can maintain higher cutting speeds and is often used for light milling operations where fine finishes are required. Figure 2(c) 4-Flute: The 4-flute cutter offers greater strength and rigidity compared to the 2-flute cutter. This tool is ideal for more aggressive milling operations, providing increased material removal rates. The additional cutting edges reduce the load on each flute, which helps to maintain tool life under heavy cutting conditions. This type of cutter is especially effective for cutting harder materials and for achieving a smoother finish at higher cutting speeds. Figure 2(d) Tool Nomenclature: The tool nomenclature shown here is for a 4 mm face milling cutter with specific dimensions: L1 = 50 mm, L2 = 10 mm, D1 = 4 mm, and D2 = 6 mm. The nomenclature helps in identifying the various aspects of the cutter, where L1 represents the overall length of the tool, L2 is the length of the cutting edge, D1 is the diameter of the cutter, and D2 refers to the diameter of the shank. These dimensions are crucial for selecting the correct tool for specific milling operations, ensuring precision and compatibility with the machine setup. Figure 2(e) Machine used for Milling: The milling operation was carried out using the BFW Gaurav (BMV 35+), a compact and high-efficiency vertical machining center. This machine is designed for milling, drilling, and tapping operations, equipped with a 12,000 RPM spindle, 15 kW power, and BT-40 tooling. It offers precise machining with ± 0.01 mm accuracy, making it suitable for robust metal cutting. The BFW Gaurav features a strong ribbed box-type structure for improved thermal stability and rigidity. With automatic tool changers and a high-performance Fanuc control system, it is capable of handling diverse machining tasks efficiently. Figure 2(f) Enlarged view of the sample: An enlarged view of the sample is shown, providing a closer look at the machined surface after the face milling operation. The image reveals the fine details of the milled surface, showing the uniformity and quality of the finish achieved by the milling process. The surface texture is crucial in assessing the effectiveness of the milling cutter and the machine setup in achieving the desired specifications. The high precision of the machine ensures that the sample is milled to the exact required dimensions and surface quality. Figure 2 (g) Face milled sample: The finished face-milled sample demonstrates the outcome of the milling process, with clear evidence of the material removal achieved. The surface is smooth, and the edges are sharply defined, indicating that the milling operation has been carried out with high precision. The sample shows consistent material removal across the surface, ensuring uniformity in thickness and shape. The effectiveness of the 4 mm face milling cutter and the BFW Gaurav machine is evident in the high-quality finish, which meets the specific dimensional and surface roughness requirements for the application. Hybrid approach The proposed hybrid approach integrates Response Surface Methodology (RSM), Spherical Fuzzy Analytic Hierarchy Process (SF-AHP), and Fuzzy MARCOS to achieve a comprehensive and reliable optimization of the machining process. RSM is used to capture the nonlinear relationships between machining parameters and machinability responses accurately, while SF-AHP addresses the uncertainty and hesitation in expert judgment to determine the relative importance of quality, productivity, and energy-related criteria. These weighted criteria are then applied within the Fuzzy MARCOS framework to rank machining alternatives by considering both beneficial and non-beneficial responses. The adoption of this integrated approach is driven by the limitations of single-method techniques in addressing the multi-response nature of machining problems. By sequentially linking prediction, criteria weighting, and decision-making, the proposed methodology offers a holistic optimization strategy that enhances process understanding, improves decision accuracy, and identifies machining conditions that simultaneously improve surface quality, reduce power consumption, and increase tool and process efficiency in WAAM-fabricated bimetallic components. The selection of SF-AHP and Fuzzy MARCOS is based on the multi-response nature of the problem. The relative importance of Ra, MRR, Pc, and CF involves expert judgment and uncertainty; hence, SF-AHP is employed due to its ability to represent membership, non-membership, and hesitation in the weighting process, offering more flexibility than traditional crisp weighting methods. Fuzzy MARCOS was chosen for alternative ranking because it evaluates each machining condition relative to ideal and anti-ideal reference solutions, providing a utility-based ranking under conflicting criteria. Although methods like TOPSIS and VIKOR are effective for multi-criteria decision-making, TOPSIS focuses on geometric distance from ideal solutions, while VIKOR emphasizes compromise ranking. In contrast, the SF-AHP–Fuzzy MARCOS combination provides a structured framework that separately addresses uncertain criteria weighting and robust alternative ranking, making it particularly suitable for optimizing the machinability of WAAM-fabricated Ni–SS bimetallic materials. Experimental design Figure 3(a) illustrates the application of Response Surface Methodology (RSM) in modelling the machinability behavior of WAAM-fabricated bimetallic samples by examining the influence of process parameters—cutting speed, feed rate, depth of cut, and flute count—on key performance responses, including surface roughness (Ra), material removal rate (MRR), power consumption (Pc), and cutting force (CF). The procedure begins with the selection of process parameters, followed by the determination of the central composite design with an appropriate α value, identification of response variables, and execution of virtual experiments. Model validation is carried out through ANOVA, assessing adequacy via statistical indicators such as the F-test, lack-of-fit test, adjusted and predicted R² values, and signal-to-noise ratio (S/N > 4). Once validated, regression equations are developed, and predicted values are compared to experimental results. Two-dimensional and three-dimensional surface plots are then generated to visualize parameter interactions and response trends, facilitating optimization and experimental verification. The outputs from this stage feed into subsequent phases: RSM quantifies the effects of machining parameters on Ra, MRR, Pc, and CF; SF-AHP assigns relative weights to these responses based on their importance under uncertainty; and Fuzzy MARCOS uses these weighted criteria to rank machining alternatives and identify the optimal parameter combination. This sequential integration ensures that the final decision is backed by both rigorous statistical modelling and uncertainty-aware multi-criteria decision-making. The 17-run RSM-based experimental design was selected to strike a balance between statistical modelling capacity and experimental feasibility. Given the significant fabrication, machining, and measurement effort required for WAAM-fabricated bimetallic samples, a compact design was adopted to minimize experimental burden while still allowing the evaluation of dominant effects of cutting speed, feed rate, depth of cut, and flute number on Ra, MRR, Pc, and CF. The design is sufficient for identifying main effects, curvature behavior, and selected two-factor interactions within the parameter range studied. The adequacy of the fitted models was verified through residual diagnostics, actual-versus-predicted plots, ANOVA, non-significant lack-of-fit results, high coefficient of determination values, and adequate precision values. However, this design does not fully capture all higher-order or three-factor interactions. Future research may use expanded factorial or sequential RSM designs to explore these effects more comprehensively. The machining responses were modelled using a second-order polynomial equation in the Response Surface Methodology framework, expressed as:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\boldsymbol{Y}={\boldsymbol{\beta\:}}_{0}+\sum_{\boldsymbol{i}=1}^{\boldsymbol{k}}{\boldsymbol{\beta\:}}_{\boldsymbol{i}}{\boldsymbol{X}}_{\boldsymbol{i}}+\sum_{\boldsymbol{i}=1}^{\boldsymbol{k}}{\boldsymbol{\beta\:}}_{\boldsymbol{i}\boldsymbol{i}}{\boldsymbol{X}}_{\boldsymbol{i}}^{2}+\sum_{\boldsymbol{i}<\boldsymbol{j}}^{\boldsymbol{k}}{\boldsymbol{\beta\:}}_{\boldsymbol{i}\boldsymbol{j}}{\boldsymbol{X}}_{\boldsymbol{i}}{\boldsymbol{X}}_{\boldsymbol{j}}+\boldsymbol{\epsilon\:}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\boldsymbol{Y}$$\end{document}represents the predicted response (Ra, MRR, Pc, or CF), \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\beta\:}}_{0}$$\end{document}is the intercept term, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\beta\:}}_{\boldsymbol{i}}$$\end{document}are the linear coefficients, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\beta\:}}_{\boldsymbol{i}\boldsymbol{i}}$$\end{document}are the quadratic coefficients, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{\beta\:}}_{\boldsymbol{i}\boldsymbol{j}}$$\end{document}are the interaction coefficients. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{X}}_{\boldsymbol{i}}$$\end{document}and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{X}}_{\boldsymbol{j}}$$\end{document}represent the coded values of the process parameters, namely cutting speed (A), feed rate (B), depth of cut (C), and flute number (D), and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\boldsymbol{\epsilon\:}$$\end{document}represents the experimental error. The coefficients of the model were determined using regression analysis based on experimental data. The face milling experiments were designed using an RSM-based experimental plan to evaluate the influence of four process variables: cutting speed, feed rate, depth of cut, and flute configuration. Cutting speed was varied from 8000 to 9500 rpm, feed rate from 0.10 to 0.25 mm/tooth, and depth of cut from 0.5 to 1.5 mm, while two-flute and four-flute tools were considered as tool-configuration levels. The selected design enabled evaluation of the dominant main effects, curvature behavior, and selected two-factor interactions with a limited number of experimental trials51,52. To reduce systematic bias, the experimental runs were performed in randomized order. This helped minimize the influence of progressive tool heating, machine drift, and local material variation in the WAAM bimetallic sample. Replicated/confirmation trials were used to assess experimental variability and verify the predicted optimal response. The adequacy of the design for RSM modelling was further confirmed through ANOVA, residual diagnostics, actual-versus-predicted agreement, non-significant lack-of-fit, and goodness-of-fit indicators. However, the compact 17-run design is mainly suitable for dominant effects and selected two-factor interactions and may not fully resolve all possible higher-order interactions. “To verify the statistical adequacy of the RSM models, residual analysis, normality checks, and multicollinearity diagnostics were performed. The normal probability plots of externally studentized residuals showed that most residuals followed the reference line, indicating approximate normality. The residuals versus predicted values were randomly distributed around zero without clear curvature or funnel-shaped patterns, confirming constant variance and absence of major model misspecification. The residuals versus run-order plots showed no systematic trend, confirming independence and process stability during experimentation. In addition, multicollinearity was assessed using variance inflation factor (VIF) values. The obtained VIF values were within acceptable limits, indicating that severe multicollinearity was not present among the model terms. Therefore, the residual diagnostics, normality checks, VIF analysis, significant ANOVA results, non-significant lack-of-fit, and good agreement between predicted and experimental values confirm the validity of the developed RSM models for machining-response prediction and optimization. Fig. 3Hybrid Methodology, (a) RSM, (b) Spherical Fuzzy AHP, (c) Fuzzy MARCOS. Hybrid Methodology, (a) RSM, (b) Spherical Fuzzy AHP, (c) Fuzzy MARCOS. Spherical fuzzy AHP –criteria weights Spherical Fuzzy Analytic Hierarchy Process (SF-AHP) was selected for this study due to its ability to effectively handle uncertainty, hesitation, and subjective expert judgment in the weighting of machining responses. In machining optimization, experts must balance conflicting criteria such as surface roughness, material removal rate, power consumption, and cutting force, which often involve linguistic vagueness and ambiguity. While Type-1 Fuzzy AHP can represent such vagueness, it lacks the capacity to explicitly model hesitation. Interval-Valued Fuzzy AHP captures uncertainty through lower and upper bounds, but does not distinguish between support, opposition, and hesitation as flexibly as needed. Intuitionistic Fuzzy AHP accounts for membership and non-membership degrees but does not provide the level of flexibility required in this context. In contrast, SF-AHP extends this representation by allowing membership, non-membership, and hesitation degrees to be expressed independently within a spherical fuzzy framework, offering greater adaptability. This flexibility is particularly useful when expert judgments are influenced by the uncertain behavior of machining processes, heterogeneous material responses, and trade-offs between quality, productivity, energy demand, and cutting stability. Consequently, SF-AHP was adopted to provide more realistic, uncertainty-aware criteria weights for the WAAM Ni–SS bimetallic machining optimization. Figure 3(b) illustrates the SF-AHP process, where factors and their respective levels are determined, and initial comparison matrices are constructed. Geometric means are employed to calculate fuzzy weights for each comparison, and the consistency ratio (C.R.) is checked for validity (C.R. < 0.1). After calculating the fuzzy weights, they are integrated into a spherical fuzzy comparison matrix, providing a robust decision-making framework. This approach ensures that appropriate weight coefficients are assigned to various machining criteria based on their relative importance, thereby offering an accurate representation of process requirements in the optimization model. Fuzzy MARCOS - ranking alternatives Multi-response machinability optimization involves inherent conflicts among the selected responses. In this study, Ra, Pc, and CF were considered cost-type criteria, while MRR was considered a benefit-type criterion. A machining condition that increases MRR may also increase Pc and CF due to higher chip load and tool engagement. Similarly, conditions that reduce Ra may not always provide the highest productivity. Therefore, the optimal solution cannot be selected based on a single response. The proposed hybrid framework resolves these conflicts by combining response prediction, uncertainty-based criteria weighting, and fuzzy utility-based ranking. RSM identifies the effect of machining parameters on each response, SF-AHP assigns relative weights to the responses, and Fuzzy MARCOS ranks the alternatives by comparing their weighted performance with ideal and anti-ideal solutions. Thus, the final optimum represents a compromise solution that balances surface quality, productivity, energy demand, and cutting stability rather than optimizing one response at the expense of others. Figure 3 (c) Fuzzy MARCOS (Multi-criteria Analysis and Ranking of Alternatives): The final stage evaluates alternative machining parameters with respect to the process requirements. The aggregated decision and normalized matrix are calculated to establish the fuzzy weighted comparability sequence and Si values. This approach employs fuzzy exponentially weighted comparability to rank alternatives and assess the sensitivity of rankings based on varying inputs. The final evaluation generates fuzzy scores, providing a clear ranking of the machining alternatives, which helps to identify the optimal combination of parameters based on the fuzzy analysis. This hybrid methodology integrates RSM for response modeling, Spherical Fuzzy AHP for assigning weights, and Fuzzy MARCOS for ranking alternatives, creating a comprehensive decision-making tool for optimizing machining parameters in WAAM-fabricated bimetallic materials. Results and discussion Mechanical characterization Mechanical property and hardness Fig. 4(a) Fabricated Sample, (b) Hardness Values, (c) Tensile Value. (a) Fabricated Sample, (b) Hardness Values, (c) Tensile Value. Figure 4a presents the as-fabricated Ni-alloy/stainless steel bimetallic structure containing silicon as a minor alloying addition. The multilayer deposition exhibits uniform bead geometry and consistent layer stacking along the build direction. No visible macroscopic defects such as delamination, interlayer cracking, or excessive porosity are observed, indicating stable process parameters and adequate metallurgical bonding between the dissimilar alloys. The continuous interface suggests effective dilution control and thermal compatibility between the Ni-based alloy and stainless steel substrate. The addition of silicon is expected to enhance melt pool fluidity and reduce solidification cracking susceptibility by promoting improved wetting behavior and controlled micro segregation. The absence of abrupt structural discontinuities implies successful compositional gradient formation across the interface, which is essential for mitigating thermal residual stresses in bimetallic systems. The Vickers hardness profile measured at fourteen locations along the deposition height is shown in Fig. 4b. The hardness values range from approximately 202 HV to 220 HV, with an average value of ~ 211 HV. The relatively narrow variation (~ 18 HV) demonstrates a homogeneous microstructural evolution throughout the deposited region. The slight increase in hardness at specific locations (≈ 220 HV) can be attributed to localized grain refinement and possible solid-solution strengthening induced by silicon addition. Silicon is known to enhance lattice distortion and restrict dislocation mobility, contributing to moderate strengthening without promoting excessive brittleness. Conversely, regions exhibiting lower hardness (~ 202 HV) likely correspond to dilution zones dominated by austenitic stainless-steel phases. The absence of sharp hardness fluctuations across the interface indicates minimal formation of brittle intermetallic compounds. Such mechanical uniformity is particularly beneficial in multi-material systems, as abrupt hardness gradients often lead to stress concentration and premature failure. Therefore, the obtained hardness profile confirms effective metallurgical compatibility and controlled thermal cycling during fabrication. The engineering stress–strain response (Fig. 4c) reveals a typical ductile deformation behavior characterized by a smooth elastic–plastic transition, pronounced strain hardening, an ultimate tensile strength of approximately 450 MPa, and total elongation of 32–35%. The substantial plasticity indicates effective load transfer across the metallurgical interface and confirms that the austenitic matrix remains structurally dominant throughout deformation. The absence of premature fracture or abrupt stress drops prior to failure suggests robust interfacial bonding between the Ni-alloy and stainless steel regions, demonstrating successful metallurgical integration without the formation of mechanically weak interlayers. The synergistic combination of moderate strength and high ductility is primarily governed by solid-solution strengthening from Ni and Si, together with microstructural refinement induced by rapid solidification during deposition. Silicon contributes to lattice distortion–based strengthening while suppressing brittle phase formation, thereby preserving ductility and promoting stable strain hardening. Complementary hardness measurements (200–220 HV) further position the material within a favorable machinability regime. This moderate and uniform hardness distribution minimizes abrasive tool wear and prevents differential tool loading across the bimetallic interface, thereby reducing vibration and chatter during milling operations. The high ductility facilitates stable and continuous chip formation, although chip-breaking strategies may be necessary to control chip morphology. Additionally, silicon enhances surface integrity by mitigating built-up edge formation and improving solidification homogeneity. Collectively, the mechanical and hardness characteristics indicate that the fabricated bimetallic system achieves an optimal balance between structural performance and post-processing machinability, making it suitable for advanced manufacturing applications. Microstructure Fig. 5Microstructure: (a) Bimetallic Structure, (b) Top Layer, (c) Mid Layer, (d) Bottom Layer. Microstructure: (a) Bimetallic Structure, (b) Top Layer, (c) Mid Layer, (d) Bottom Layer. The Ni–SS interface plays a pivotal role in determining the machining response of the WAAM-fabricated bimetallic material. During face milling, the cutting edge does not engage with a homogeneous workpiece but instead interacts with Ni-rich regions, SS-rich regions, and transition zones created by compositional dilution and metallurgical bonding. These regions can differ in properties such as hardness, thermal conductivity, microstructure, and plastic deformation behavior, leading to varying contact conditions across the interface. These changes influence several aspects of the machining process, including force generation, power consumption, chip formation, material removal stability, and surface quality. The Ni-rich region may increase cutting resistance and heat generation, while the SS-rich region may contribute to adhesion and strain hardening. The combined effect of these mechanisms at the bimetallic interface can cause localized instability, particularly under interrupted multi-tooth engagement during face milling. Therefore, the observed variations in surface roughness (Ra), material removal rate (MRR), power consumption (Pc), and cutting force (CF) should be viewed not only as effects of cutting parameters but also as consequences of the heterogeneous tool–workpiece interaction across the Ni–SS interface. Figure 5 presents the macro- and microstructural characteristics of the fabricated Ni-alloy/stainless steel bimetallic structure along the build direction. The macroscopic view (Fig. 5a) shows a well-defined multilayer architecture with consistent bead geometry, free from visible macrocracks, delamination, or distortion, suggesting stable melt pool behavior, controlled dilution, and effective thermal management during deposition. Microstructural examination reveals a gradient evolution influenced by thermal conditions: the bottom region (Fig. 5d) exhibits coarser, well-developed dendritic structures from substrate-driven directional solidification and thermal cycling; the mid-layer (Fig. 5c) displays a mixed cellular–dendritic morphology with interconnected grain boundary networks, typical of epitaxial growth under moderated cooling and thermal accumulation; while the top layer (Fig. 5b) shows finer, more homogeneous equiaxed grains due to enhanced convective heat dissipation and rapid solidification near the free surface. Uniformly dispersed secondary phases, particularly in the upper region, suggest effective silicon distribution, promoting solid-solution strengthening while suppressing coarse intermetallic formation. The absence of microcracks, voids, or segregation across all regions confirms strong metallurgical bonding and successful integration between the Ni-alloy and stainless steel constituents. The gradual transition from coarse dendritic structures at the bottom to refined equiaxed grains at the top correlates with the measured hardness profile, where finer grains correspond to slightly higher hardness and coarser regions to marginally lower hardness, though the overall variation is minimal. This microstructural stability, achieved through controlled thermal gradients and silicon-assisted phase regulation, supports the balanced mechanical performance, structural integrity, and suitability of the bimetallic system for load-bearing and high-temperature applications. Predictive approach Experimental design Table 1Experimental response.AlternativesFactor 1Factor 2Factor 3Factor 4Response 1Response 2Response 3Response 4A: Cutting SpeedB: Feed rateC: Depth of CutD: No of FlutesRaMRRPcCFRpmmm/toothmmNo’sMicronsmm³/minW N 195000.11.5Four0.635.6320120280000.251Two0.928.4310115380000.1751.5Four0.7532.5315118480000.11.5Two0.8530305113595000.10.5Two0.738.2325122687500.1751Two0.834318119780000.250.5Four0.9527.6310112887500.251.5Four0.736.3322121987500.251.5Four0.736.53231221080000.11Four0.8530.43121141195000.10.5Four0.6538.73271241280000.10.5Two0.929.93081101387500.1750.5Four0.833.83191171495000.251.5Two0.75373241201580000.1751.5Four0.833.53171181695000.251Four0.640.53301251795000.250.5Two0.739335127 Table 1Experimental response.AlternativesFactor 1Factor 2Factor 3Factor 4Response 1Response 2Response 3Response 4A: Cutting SpeedB: Feed rateC: Depth of CutD: No of FlutesRaMRRPcCFRpmmm/toothmmNo’sMicronsmm³/minW N 195000.11.5Four0.635.6320120280000.251Two0.928.4310115380000.1751.5Four0.7532.5315118480000.11.5Two0.8530305113595000.10.5Two0.738.2325122687500.1751Two0.834318119780000.250.5Four0.9527.6310112887500.251.5Four0.736.3322121987500.251.5Four0.736.53231221080000.11Four0.8530.43121141195000.10.5Four0.6538.73271241280000.10.5Two0.929.93081101387500.1750.5Four0.833.83191171495000.251.5Two0.75373241201580000.1751.5Four0.833.53171181695000.251Four0.640.53301251795000.250.5Two0.739335127 Experimental response. The experimental design utilizes a Response Surface Methodology (RSM)-based 17-run scheme to investigate the machinability of WAAM Nickel–Stainless Steel bimetallic samples, considering four factors: cutting speed (8000–9500 rpm), feed rate (0.10–0.25 mm/tooth), depth of cut (0.5–1.5 mm), and tool flute count (two or four flutes), as detailed in Table 1. This design records four responses—surface roughness (Ra), material removal rate (MRR), power consumption (Pc), and cutting force (CF)—providing a comprehensive dataset that captures low, medium, and high levels of the numerical factors and both tool configurations. The design enables second-order modeling of main effects and interactions, while minimizing the number of machining trials. These selected responses are crucial for a holistic evaluation of machinability: Ra reflects surface quality and finishing performance, MRR indicates productivity and material removal efficiency, Pc reveals energy demand and machining sustainability, and CF represents mechanical loading, machining stability, and potential tool–workpiece interaction. Together, Ra, MRR, Pc, and CF provide a balanced assessment of quality, productivity, energy consumption, and process stability in the face milling of WAAM-fabricated Ni–SS bimetallic materials. Residuals diagnostics Normal probability plots of externally studentized residuals Fig. 6Externally Studentized Residuals vs. Normal % Probability, (a) Ra, (b) MRR, (c) PC, and (d) CF. Externally Studentized Residuals vs. Normal % Probability, (a) Ra, (b) MRR, (c) PC, and (d) CF. The residuals for Ra (Fig. 6(a) largely align with the reference line, with a dense cluster around zero residual, indicating that most experimental runs are well represented by the model. However, two tail departures are observed: one in the lower tail with a large negative residual and another in the upper tail with a positive residual, suggesting potentially influential runs and slightly heavier tails than an ideal normal distribution. Despite these outliers, the central portion remains linear, and no systematic curvature is present, making the normality assumption acceptable, though extreme runs should be treated as influential candidates during model validation. For MRR (Fig. 6(b), the residuals symmetrically surround zero, with no noticeable curvature or tail distortion, supporting the normal-error assumption and ensuring stable predictions. In PC (Fig. 6(c), residuals show linearity, with a mild deviation in the upper tail, indicating localized departures rather than global non-normality. CF (Fig. 6(d) displays pronounced tail structure, with clusters in both the lower and upper tails, suggesting influential observations and stronger tail effects compared to MRR and PC. Overall, normal probability diagnostics confirm that all responses are approximately normally distributed, with MRR showing the strongest conformity, and Ra and CF showing isolated tail departures, indicating potentially influential runs. Externally studentized residuals versus predicted values Fig. 7Predicted vs. Externally Studentized Residuals, (a) Ra, (b) MRR, (c) PC, and (d) CF. Predicted vs. Externally Studentized Residuals, (a) Ra, (b) MRR, (c) PC, and (d) CF. The residuals for Ra (Fig. 7(a) are centered around zero across the fitted range, with no evidence of curvature or funnel-shaped patterns, indicating no misspecification/ heteroscedasticity. One observation shows a relatively large negative residual ( ≈ − 15) near the mid-high predicted region, and another shows a moderate positive deviation, suggesting potentially influential runs. However, since these remain within the ± 18.3984 limits and the other points are tightly clustered around zero, the Ra model is statistically adequate, with the isolated deviations likely due to run-specific variability. For MRR (Fig. 7(b), residuals are randomly scattered around zero with uniform spread, supporting constant variance. A mild shift toward negative residuals at intermediate predictions is noted, but it does not form a clear trend or curvature, indicating localized experimental scatter rather than a structural issue. No points approach the control bounds, indicating no extreme outliers. In PC (Fig. 7(c), the residuals remain compact and symmetrically distributed around zero, with minimal dispersion, satisfying the assumptions of independence and homoscedasticity. CF (Fig. 7(d) residuals are also centered around zero, with a slight clustering of negative residuals in the mid-predicted region, but no systematic curvature or widening spread, indicating model adequacy and variance stability. Overall, Fig. 7 confirms random residual dispersion with no pronounced heteroscedasticity, supporting the adequacy of the predictive models. Run-order residual diagnostics (independence and process stability) Fig. 8Run Number vs. Externally Studentized Residuals, (a) Ra, (b) MRR, (c) PC, and (d) CF. Run Number vs. Externally Studentized Residuals, (a) Ra, (b) MRR, (c) PC, and (d) CF. The residuals for Ra (Fig. 8(a) remain close to zero for most runs, indicating stable behavior, with two isolated deviations: one large negative residual ( ≈ − 15) in an early run and a positive excursion ( ≈ + 6 to + 8) in a later run. Since these deviations are isolated and do not show a progressive trend across adjacent runs, they suggest run-specific abnormalities (e.g., transient vibration, chip evacuation disturbances, measurement variations) rather than systematic drift. Overall, the Ra residuals support independent errors with a few influential candidates. For MRR (Fig. 8(b), the residuals oscillate within a narrow band around zero, with no monotonic trend or clustering by run order. The amplitude remains small and uniform, confirming no run-order dependence and consistent process conditions. In PC (Fig. 8(c), the residuals are tightly distributed around zero, with minor alternating deviations. The absence of sustained trends or periodicity confirms stable error structure and supports the independence assumption. CF residuals (Fig. 8(d) show similar behavior to PC, fluctuating slightly around zero with constant spread and no extreme residuals. This confirms no time-related bias and supports adequate model representation for CF. Overall, the run-order plots show that the experiments were stable, with independent residuals for all responses. The Ra plot contains isolated influential deviations, suggesting the need for verification against experimental conditions but not indicating a systematic violation of independence. Actual versus predicted comparison of developed models The predicted Ra values (Fig. 9(a) closely align with the unity line across the full range of measured Ra, with only minor scatter observed in the mid-range, indicating small local deviations but no persistent bias. The strong clustering on the 45° line confirms that the Ra model accurately reproduces the experimental surface roughness, making it suitable for prediction and optimization within the investigated parameter domain. For MRR (Fig. 9(b), data points closely track the unity line throughout the measured range, showing nearly one-to-one correspondence between actual and predicted values. The absence of systematic curvature or offset indicates that the model effectively captures material removal behavior and maintains consistent accuracy across both low and high MRR values, reflecting a stable regression structure with minimal bias. In PC (Fig. 9(c), the plot exhibits strong linearity, with most points falling on the unity line and very limited scatter, indicating that the model reliably represents the process behavior and produces uniformly small prediction errors. For CF (Fig. 9(d), the predicted values show excellent agreement with experimental data, with points concentrated along the 45° line. Minor deviations are observed at intermediate CF values, but there is no systematic under- or over-prediction, confirming the CF model’s reliability for response analysis and multi-objective optimization. Overall, Fig. 9 demonstrates that the developed models exhibit high predictive accuracy for all responses, with close clustering of data points around the unity line and no systematic bias, validating the regression formulations for further parametric interpretation and optimization. Fig. 9Actual vs. Predicted, (a) Ra, (b) MRR, (c) PC, and (d) CF. Actual vs. Predicted, (a) Ra, (b) MRR, (c) PC, and (d) CF. The adequacy of the developed models was confirmed through multiple statistical checks. The residual plots showed no strong systematic pattern, indicating acceptable normality and variance stability. The actual-versus-predicted plots showed close agreement between experimental and predicted values. In addition, the ANOVA results confirmed model significance, while the non-significant lack-of-fit values indicated that the fitted models were adequate within the selected design space. The high R², adjusted R², predicted R², and adequate precision values further support the reliability of the models for prediction and optimization. Effects of Input proces parameters by Response surface analysis with respect to the response Ra (Microns), two & four flute for cutting speed (Rpm) vs. feed rate (mm/tooth) vs. depth of cut (mm) Figure 10(a) (A–B interaction at constant C) demonstrates that Ra decreases with increasing cutting speed and increases sharply with feed rate. The surface exhibits a monotonic slope with mild convexity toward high feed, indicating accelerated Ra deterioration as feed approaches its upper limit. The lowest Ra is observed at high speed and low feed, where chip formation is stable, and feed marks are minimized. Figure 10(b) (A–C interaction at constant B) shows that the Ra surface is nearly planar, indicating limited influence of cutting speed and depth of cut on Ra when feed is held constant. The wide contour spacing suggests weak sensitivity and negligible A–C interaction for the two-flute tool. In Fig. 10(c) (B–C interaction at constant A), Ra increases with both feed rate and depth of cut, with the steepest rise in the high-feed/high-depth region. The curvature and contour compression indicate that surface roughness amplifies due to increased cutting forces and dynamic instability, along with larger feed marks. The surface is strongly convex, rising nonlinearly towards higher feed rates and depths of cut. These findings provide insights into the interdependencies between cutting parameters and surface roughness, with feed rate being the dominant factor influencing Ra. Fig. 10Ra(Microns), Two (a) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (b) Cutting Speed (Rpm) vs. Depth of Cut (mm), (c) feed rate (mm/tooth) vs. Depth of Cut (mm), Four (d) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (e) Cutting Speed (Rpm) vs. Depth of Cut (mm), (f) feed rate (mm/tooth) vs. Depth of Cut (mm). Ra(Microns), Two (a) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (b) Cutting Speed (Rpm) vs. Depth of Cut (mm), (c) feed rate (mm/tooth) vs. Depth of Cut (mm), Four (d) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (e) Cutting Speed (Rpm) vs. Depth of Cut (mm), (f) feed rate (mm/tooth) vs. Depth of Cut (mm). Figure 10(d) (A–B interaction at constant C) shows that Ra decreases with cutting speed and increases with feed rate, similar to the two-flute configuration. However, the surface is shifted downward with a broader low-Ra region, indicating that surface finish is less sensitive to parameter escalation under the four-flute tool, especially at moderate feeds. The curve is inclined with reduced curvature and a broader low-Ra plateau. In Fig. 10(e) (A–C interaction at constant B), the surface remains relatively flat, confirming that feed is the primary driver of Ra. A slight increase in Ra with depth of cut is observed, while cutting speed provides a mild improvement. The curve appears as a weakly inclined plane with low curvature. Figure 10(f) (B–C interaction at constant A) shows that Ra increases with both feed rate and depth of cut, but the surface is less steep compared to the two-flute tool. The reduced contour compression at high feed and depth indicates that the four-flute configuration mitigates the combined adverse effects of high chip load and increased force, improving stability and surface integrity. The curve is convex but with moderated steepness relative to the two-flute case. The four-flute tool consistently yields lower Ra and a broader low-roughness window across factor pairs. This improvement is due to more distributed cutting engagement and reduced chip load per tooth, minimizing surface tearing and feed marks. The four-flute configuration also shows reduced curvature and weaker sensitivity in high-feed/high-depth regions, enhancing process stability and reducing force-induced vibration. Overall, feed rate dominates Ra, with depth of cut influencing Ra primarily through its interaction with feed. The four-flute tool offers a clear performance advantage, maintaining surface quality over a broader range of cutting parameters. MRR(mm3/min), two & four flute for cutting speed (rpm) vs. feed rate (mm/tooth) vs. depth of cut (mm) Figure 11(a) (A–B interaction at constant C) shows that MRR increases with both cutting speed and feed rate, with the surface rising toward the high-speed/high-feed corner. The contour compression along the feed axis highlights feed rate as a strong contributor to productivity, with cutting speed providing a secondary influence. The curve is monotonic, with mild curvature, steepest toward high feed. In Fig. 11(b) (A–C interaction at constant B), MRR increases with depth of cut and, to a lesser extent, cutting speed. The surface remains largely planar with a gentle slope, indicating limited A–C interaction when feed is fixed. Figure 11(c) (B–C interaction at constant A) reveals that MRR rises with both feed rate and depth of cut but flattens at higher values, indicating diminishing returns due to force and stability limitations. The contour pattern shows that increases in depth of cut raise MRR more consistently than feed, with the highest MRR in the high-feed/high-depth region. Fig. 11MRR(mm3/min), Two (a) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (b) Cutting Speed (Rpm) vs. Depth of Cut (mm), (c) feed rate (mm/tooth) vs. Depth of Cut (mm), Four (d) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (e)Cutting Speed (Rpm) vs. Depth of Cut (mm), (f) feed rate (mm/tooth) vs. Depth of Cut (mm). MRR(mm3/min), Two (a) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (b) Cutting Speed (Rpm) vs. Depth of Cut (mm), (c) feed rate (mm/tooth) vs. Depth of Cut (mm), Four (d) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (e)Cutting Speed (Rpm) vs. Depth of Cut (mm), (f) feed rate (mm/tooth) vs. Depth of Cut (mm). Figure 11(d) (A–B interaction at constant C) shows that MRR increases with both cutting speed and feed rate, similar to the two-flute tool, but the high-MRR region is broader and the surface is elevated across much of the parameter domain, indicating improved productivity robustness under the four-flute configuration. The curve is monotonic, with reduced steepness variations and a wider high-MRR plateau. In Fig. 11(e) (A–C interaction at constant B), MRR increases with depth of cut, with a mild trend toward improvement from cutting speed. The surface is nearly planar with a smooth gradient, highlighting stable effects of depth engagement. Figure 11(f) (B–C interaction at constant A) reveals that MRR rises with both feed rate and depth of cut, but the curvature indicates diminishing returns at higher depth due to torque, chip evacuation, and tool engagement limitations. Compared to the two-flute case, the four-flute tool maintains MRR more effectively at moderate-to-high feed rates. The four-flute tool consistently delivers higher MRR and a broader high-productivity region, enabling stable performance at elevated cutting speeds and feed rates. This improvement is due to multi-tooth engagement, reducing per-tooth loading and enhancing chip formation stability. The four-flute response surfaces show smoother gradients and reduced curvature, indicating lower sensitivity to parameter variations and more stable volumetric removal. MRR is primarily driven by feed rate and depth of cut, with cutting speed providing secondary enhancement. The four-flute tool significantly expands the high-MRR operating window, offering superior performance across the tested parameter space. Power consumption (w), two & four flute for cutting speed (rpm) vs. feed rate (mm/tooth) vs. depth of cut (mm) Fig. 12Power Consumption (w), Two (a) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (b) Cutting Speed (Rpm) vs. Depth of Cut (mm), (c) feed rate (mm/tooth) vs. Depth of Cut (mm), Four (d) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (e)Cutting Speed (Rpm) vs. Depth of Cut (mm), (f) feed rate (mm/tooth) vs. Depth of Cut (mm). Power Consumption (w), Two (a) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (b) Cutting Speed (Rpm) vs. Depth of Cut (mm), (c) feed rate (mm/tooth) vs. Depth of Cut (mm), Four (d) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (e)Cutting Speed (Rpm) vs. Depth of Cut (mm), (f) feed rate (mm/tooth) vs. Depth of Cut (mm). Figure 12(a) (Cutting speed vs. feed rate, C constant) shows that PC exhibits a distinct quadratic behavior, with a minimum region at intermediate cutting speeds and low-to-moderate feeds. PC increases with higher cutting speeds and feed rates, indicating that simultaneous escalation of cutting speed and feed raises torque and power demand. The curve is bowl-shaped (convex upward), with more pronounced curvature along the feed direction. In Fig. 12(b) (Cutting speed vs. depth of cut, B constant), PC increases with both cutting speed and depth of cut, displaying a predominantly additive effect of cutting speed and depth with limited curvature, indicating weak interaction. In Fig. 12(c) (Feed rate vs. depth of cut, A constant), PC increases with depth of cut and shows a rising trend with feed rate, though the rate of increase slows in the upper region. This suggests increasing engagement-driven load with depth, with feed’s contribution becoming less significant at higher settings. The curve shows mild concavity, indicating saturation at high feed and depth. Figure 12(d) (Cutting speed vs. feed rate, C constant) shows that PC follows a similar quadratic pattern to the two-flute case, with a low-PC basin at intermediate cutting speeds and lower feeds, and an increase toward higher cutting speeds and feed rates. Compared to the two-flute surface, the gradients are smoother, and the low-PC region is broader, indicating enhanced robustness. The curve is bowl-shaped (convex upward) with a wider minimum zone and reduced steepness. In Fig. 12(e) (Cutting speed vs. depth of cut, B constant), PC increases with both cutting speed and depth of cut, with the surface remaining close to planar, suggesting stable power escalation. The slope is slightly flatter than the two-flute case, indicating reduced sensitivity to parameter changes. Figure 12(f) (Feed rate vs. depth of cut, A constant) shows a flatter surface, indicating a weaker combined influence of feed rate and depth of cut on PC under the four-flute tool, with subdued B–C interaction. Across the response surfaces, the four-flute tool demonstrates smoother power consumption (PC) gradients and a broader low-PC operating basin, suggesting a reduced sensitivity of spindle power demand to simultaneous increases in process parameters. This results in a more tolerant operating window, where moderate increases in speed, feed, or depth do not cause abrupt power surges. When combined with the productivity surfaces (MRR), the four-flute configuration proves advantageous by maintaining comparable PC levels while enabling higher material removal rates, thereby improving process efficiency (higher output per unit of power). This behavior indicates more stable cutting action and reduced load concentration per cutting edge, mitigating sharp power spikes at aggressive machining settings. From a sustainability perspective, power consumption is a key indicator of energy demand during machining. In this study, PC was treated as a cost-type response, while MRR was treated as a benefit-type response. Therefore, the optimal machining condition should not only minimize power demand but also maintain sufficient material removal productivity. A machining condition with lower PC and higher MRR enhances eco-efficiency by increasing the amount of material removed per unit of energy input. Furthermore, reduced cutting force can lower mechanical loading on the tool and spindle, while improved surface roughness can reduce the need for additional finishing operations. Thus, the optimized parameter combination contributes to sustainable machining by balancing surface integrity, productivity, energy consumption, and process stability. Although a full life-cycle assessment is not included in the present study, the inclusion of PC and MRR provides valuable machining-level sustainability indicators for the post-processing of WAAM-fabricated Ni–SS bimetallic components. CF (N), two & four flute for cutting speed (rpm) vs. feed rate (mm/tooth) vs. depth of cut (mm) Fig. 13CF (N), Two (a) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (b) Cutting Speed (Rpm) vs. Depth of Cut (mm), (c) feed rate (mm/tooth) vs. Depth of Cut (mm), Four (d) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (e) Cutting Speed (Rpm) vs. Depth of Cut (mm), (f) feed rate (mm/tooth) vs. Depth of Cut (mm). CF (N), Two (a) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (b) Cutting Speed (Rpm) vs. Depth of Cut (mm), (c) feed rate (mm/tooth) vs. Depth of Cut (mm), Four (d) Cutting Speed (Rpm) vs. Feed rate (mm/tooth), (e) Cutting Speed (Rpm) vs. Depth of Cut (mm), (f) feed rate (mm/tooth) vs. Depth of Cut (mm). Figure 13(a) (Cutting speed vs. feed rate, C constant) shows that CF increases strongly with feed rate and exhibits a mild increasing trend with cutting speed, with maximum CF occurring at high speed and feed. The surface is largely planar with a dominant feed gradient, indicating that feed rate is the primary driver of CF in the two-flute configuration. In Fig. 13(b) (Cutting speed vs. depth of cut, B constant), CF increases with both cutting speed and depth of cut, with a more pronounced slope along the depth-of-cut axis, suggesting significant contribution from engagement to force escalation. The surface exhibits mild curvature, indicating a non-linear increase at higher depths. Figure 13(c) (Feed rate vs. depth of cut, A constant) shows that CF rises with both feed rate and depth of cut, with the highest force predicted when both parameters are elevated. The contour tightening toward the high-feed/high-depth region indicates a coupled amplification of force, with a convex upward surface rising more rapidly toward the high B–C corner. Figure 13(d) (Cutting speed vs. feed rate, C constant) shows that CF increases with feed rate and exhibits a mild increase with cutting speed. Compared to the two-flute plot, the gradient is smoother, and the high-force region is less abrupt, indicating reduced sensitivity to parameter escalation. The curve is an inclined plane with moderated slope and reduced curvature. In Fig. 13(e) (Cutting speed vs. depth of cut, B constant), CF increases primarily with depth of cut, with a smaller contribution from cutting speed. The surface remains smooth with mild curvature, indicating predictable force scaling with engagement. Compared to the two-flute case, the slope is slightly reduced across the domain, suggesting a more stable force response. Figure 13(f) (Feed rate vs. depth of cut, A constant) shows CF rising with both feed rate and depth of cut, but with less pronounced curvature than in the two-flute case, and the contour map is more widely spaced, indicating that the four-flute tool reduces the combined influence of chip load and engagement on cutting force. The four-flute configuration demonstrates superior stability and a broader operating window by maintaining controlled forces, with feed rate and depth of cut as the primary factors influencing cutting force (CF), while cutting speed plays a secondary role. Its smoother gradients and reduced curvature signify lower CF sensitivity to simultaneous increases in the input parameters, resulting in improved dimensional accuracy and extended tool life. Moreover, when considering material removal rate (MRR) and power consumption (PC) trends, the four-flute configuration supports higher productivity in regions with less force escalation, facilitating stable operation under aggressive machining conditions within the investigated parameter space. ANOVA Table 2Model Significance.SourcedfF ValueProb > FRemarksRaMRRPcCFRaMRRPcCFModel1121.84921.1117.87814.5360.01370.01440.01830.0246significantA: Cutting Speed131.66410.63311.8659.4920.01110.04710.04110.0541B: Feed rate125.9819.7928.69320.5420.01460.05210.06010.0201C: Depth of Cut116.698.45627.20918.1030.02650.06210.01370.0238D: No of Flute217.32113.1692.61514.5770.02250.03270.22010.0285AB113.52910.63322.66232.9490.03480.04710.01760.0105AC0––––AD219.06412.73712.23323.7260.01970.03420.03610.0145BC117.13325.61.52927.6490.02560.01490.30420.0134BD218.14427.11815.04424.8270.02110.01200.02730.0136CD0––––Residual3––––Lack of Fit11.5520.5490.7850.6740.3390.5360.4690.498not significantPure Error2Cor Total14 Table 2Model Significance.SourcedfF ValueProb > FRemarksRaMRRPcCFRaMRRPcCFModel1121.84921.1117.87814.5360.01370.01440.01830.0246significantA: Cutting Speed131.66410.63311.8659.4920.01110.04710.04110.0541B: Feed rate125.9819.7928.69320.5420.01460.05210.06010.0201C: Depth of Cut116.698.45627.20918.1030.02650.06210.01370.0238D: No of Flute217.32113.1692.61514.5770.02250.03270.22010.0285AB113.52910.63322.66232.9490.03480.04710.01760.0105AC0––––AD219.06412.73712.23323.7260.01970.03420.03610.0145BC117.13325.61.52927.6490.02560.01490.30420.0134BD218.14427.11815.04424.8270.02110.01200.02730.0136CD0––––Residual3––––Lack of Fit11.5520.5490.7850.6740.3390.5360.4690.498not significantPure Error2Cor Total14 Model Significance. Table 2 shows that the developed model was statistically significant for all four responses, namely surface roughness (Ra), material removal rate (MRR), power consumption (Pc), and cutting force (CF), since the model p-values were below 0.05. Among the main factors, cutting speed significantly influenced Ra, MRR, and Pc, while its effect on CF was marginal. Feed rate significantly affected Ra and CF, whereas depth of cut significantly influenced Ra, Pc, and CF. The number of flutes showed a significant contribution to Ra, MRR, and CF, but not to Pc. Regarding interaction effects, AB, AD, and BD were significant for all responses, indicating that the combined influence of machining parameters played an important role in determining machining performance. The BC interaction was significant for Ra, MRR, and CF, but insignificant for Pc. Furthermore, the lack-of-fit test was non-significant for all responses, confirming the adequacy and reliability of the fitted models within the investigated range52–55. Goodness of Fit Table 3Model performance.ParameterStd. Dev.MeanC.V. %R-SquaredAdj R-SquaredPred R-SquaredAdeq PrecisionRa0.0230.12618.2540.98440.95820.92048.747MRR0.1100.14774.830.96410.91810.90049.211Pc0.5210.495105.2530.95570.92830.91246.046CF0.6850.77888.0460.98580.92500.91325.113 Table 3Model performance.ParameterStd. Dev.MeanC.V. %R-SquaredAdj R-SquaredPred R-SquaredAdeq PrecisionRa0.0230.12618.2540.98440.95820.92048.747MRR0.1100.14774.830.96410.91810.90049.211Pc0.5210.495105.2530.95570.92830.91246.046CF0.6850.77888.0460.98580.92500.91325.113 Model performance. Table 3 indicates that the developed models exhibited strong fitting accuracy and satisfactory predictive capability for all machining responses. The coefficient of determination (R²) values were high, ranging from 0.9557 to 0.9858, demonstrating that the proposed models explained most of the variability in surface roughness (Ra), material removal rate (MRR), power consumption (Pc), and cutting force (CF). The adjusted and predicted R² values were also in close agreement for each response, suggesting acceptable model consistency and the absence of substantial overfitting. Furthermore, the Adeq Precision values for all responses exceeded the recommended threshold of 4, confirming an adequate signal and indicating that the models are suitable for navigating the design space. Among the responses, Ra showed the lowest coefficient of variation, reflecting comparatively stable behaviour, whereas MRR, Pc, and CF exhibited higher variability52–55. Overall, the statistical indicators confirm that the developed models are reliable for prediction, analysis, and process optimization. Model sensitivity Fig. 14Desirability Index. Desirability Index. Figure 14 presents the desirability-based optimization results for the face-milling process. The overall desirability value, approaching 0.90, indicates that the selected machining condition provides a highly satisfactory compromise among the considered responses. Although the individual desirability values vary across the responses, this variation reflects the expected trade-offs in multi-response optimization. Importantly, the composite desirability index confirms that the optimized setting does not favor one response disproportionately, but instead achieves a balanced improvement in surface quality, material removal performance, power consumption, and cutting force52–55. This demonstrates the suitability of the adopted optimization framework for identifying practically viable machining conditions within the investigated design space. Determination of criteria weightage Linguistic measures for pairwise comparisons and score index (SI) Table 4Linguistic for pairwise comparisons.ParametersAbsolutely More importanceVery high importanceHigh importanceSlightly more importanceEqually importanceSlightly low ImportanceLow ImportanceVery low ImportanceAbsolutely Low ImportanceAMIVHIHISMIEISLILIVLIALIµ, υ, π0.9,0.1,000.8,0.2,0.10.7,0.3,0.20.6,0.4,0.30.5,0.4,0.40.4,0.6,0.30.3,0.7,0.20.2,0.8,0.10.1,0.9,0.0Sore Index (SI)975311/31/51/71/9 Table 4Linguistic for pairwise comparisons.ParametersAbsolutely More importanceVery high importanceHigh importanceSlightly more importanceEqually importanceSlightly low ImportanceLow ImportanceVery low ImportanceAbsolutely Low ImportanceAMIVHIHISMIEISLILIVLIALIµ, υ, π0.9,0.1,000.8,0.2,0.10.7,0.3,0.20.6,0.4,0.30.5,0.4,0.40.4,0.6,0.30.3,0.7,0.20.2,0.8,0.10.1,0.9,0.0Sore Index (SI)975311/31/51/71/9 Linguistic for pairwise comparisons. Table 4 presents the linguistic measures used for pairwise comparisons, along with the corresponding membership functions (µ, υ, π) and score index (SI) values. These linguistic terms are employed to assess the relative importance of various factors in a fuzzy decision-making framework. “Absolutely More Importance” (AMI) is represented by a membership function of (0.9, 0.1, 0) and assigned a score index of 9, indicating the highest level of importance. “Very High Importance” (VHI), with a membership function of (0.8, 0.2, 0.1), is given a score index of 7, while “High Importance” (HI) is defined by (0.7, 0.3, 0.2) and has a score index of 5. “Slightly More Importance” (SMI) has a membership function of (0.6, 0.4, 0.3) and a score index of 3. “Equally Important” (EI) is represented by (0.5, 0.4, 0.4) and a score index of 1. Lower levels of importance, such as “Slightly Low Importance” (SLI), “Low Importance” (LI), “Very Low Importance” (VLI), and “Absolutely Low Importance” (ALI), are associated with decreasing membership values and score indices, with ALI having the lowest score of 1/9. These measures facilitate the ranking and comparison of criteria based on their relative importance. Initial Comparison matrix Table 5Initial pairwise comparison matrix for criteria evaluation.Left Criteria Is GreaterRight Criteria Is GreaterTNEAMIVHIHISMIEISLILIVLIALIA--1--1---B2A---1-----C2A-1---1---D2B-----11--C2B--11-----D2C-1---1---D2 Table 5Initial pairwise comparison matrix for criteria evaluation.Left Criteria Is GreaterRight Criteria Is GreaterTNEAMIVHIHISMIEISLILIVLIALIA--1--1---B2A---1-----C2A-1---1---D2B-----11--C2B--11-----D2C-1---1---D2 Initial pairwise comparison matrix for criteria evaluation. Table 5 Initial Comparison Matrix presents the pairwise comparison results for different criteria based on their relative importance. The matrix uses linguistic terms such as “Absolutely More Importance” (AMI), “Very High Importance” (VHI), “High Importance” (HI), and others to express the dominance of one criterion over another. Each pairwise comparison assigns a value (1, 2, etc.) indicating the strength of preference, with the left and right criteria compared for each relationship. For instance, criterion A is compared to B, C, and D with specific values, signifying its relative importance in the context of the decision-making process. The matrix helps to establish the comparative significance of factors, guiding the subsequent decision-making steps in the fuzzy multi-criteria optimization framework. This comparison matrix is crucial in calculating weight coefficients, enabling further analysis through methods like Spherical Fuzzy AHP and Fuzzy MARCOS for evaluating and ranking machining alternatives. Integrated spherical fuzzy comparison matrix Table 6Integrated spherical fuzzy comparison matrix for criteria evaluation.ABCDA0.5000.4000.4000.5290.4870.2670.4240.5880.2470.5660.4650.247B0.4240.5880.2470.5000.4000.4000.3460.6550.2510.6480.3550.258C0.5290.4870.2670.6480.3550.2580.5000.4000.4000.5660.4650.247D0.3460.6710.1940.3460.6550.2510.3460.6710.1940.5000.4000.400 Table 6Integrated spherical fuzzy comparison matrix for criteria evaluation.ABCDA0.5000.4000.4000.5290.4870.2670.4240.5880.2470.5660.4650.247B0.4240.5880.2470.5000.4000.4000.3460.6550.2510.6480.3550.258C0.5290.4870.2670.6480.3550.2580.5000.4000.4000.5660.4650.247D0.3460.6710.1940.3460.6550.2510.3460.6710.1940.5000.4000.400 Integrated spherical fuzzy comparison matrix for criteria evaluation. Table 6 Integrated Spherical Fuzzy Comparison Matrix presents the pairwise comparisons of criteria A, B, C, and D using spherical fuzzy values. The matrix reflects the relative importance of each criterion, where each cell shows the fuzzy membership values for the comparisons. For instance, the comparison between criteria A and B yields a value of 0.500 for A’s importance over B, while B over A is represented by 0.400. Similar comparisons are shown for the other criteria, capturing the nuances of each criterion’s influence in the decision-making process. These values are derived from the fuzzy AHP methodology, allowing for a more flexible and robust evaluation compared to traditional crisp values. The matrix facilitates the subsequent assignment of weight coefficients to each criterion, serving as a foundation for further multi-criteria decision-making processes in the optimization of machining parameters for WAAM-fabricated bimetallic materials. Spherical fuzzy weights criteria Table 7Spherical fuzzy weights.CriteriaC1C2C3C4W10.5090.5010.5660.393W20.480.4830.4230.586W30.2990.30.2980.284 Table 7Spherical fuzzy weights.CriteriaC1C2C3C4W10.5090.5010.5660.393W20.480.4830.4230.586W30.2990.30.2980.284 Spherical fuzzy weights. Table 7 Spherical Fuzzy Weights presents the fuzzy weights for criteria C1, C2, C3, and C4, which are derived through the Spherical Fuzzy AHP method. Each criterion is associated with three membership values, representing different levels of importance. For instance, C1 has values of 0.509, 0.480, and 0.299, while C2 is represented by 0.501, 0.483, and 0.300. Similarly, C3 has fuzzy weights of 0.566, 0.423, and 0.298, and C4 has values of 0.393, 0.586, and 0.284. These weights provide a fuzzy representation of each criterion’s significance, incorporating uncertainty and subjectivity in the decision-making process. The values derived from the spherical fuzzy approach help to prioritize the criteria more accurately, forming the basis for the subsequent ranking and selection of optimal machining parameters in the multi-criteria optimization framework. Consistency ratio The consistency ratio (CR) is a crucial measure for evaluating the reliability of pairwise comparisons in the decision-making process. In this study, the calculated CR value is 0.0189, which falls within the acceptable threshold of 0.1, indicating that the pairwise comparisons used in the analysis are consistent. A CR value below 0.1 signifies that the judgments made during the comparison process are sufficiently consistent and reliable, ensuring the robustness of the Spherical fuzzy AHP methodology employed for the optimization of machining parameters. This consistency verification enhances the credibility of the weight assignments and supports the overall decision-making framework for the machining of WAAM-fabricated bimetallic materials. Determination of ranking alternatives by fuzzy MARCOS Linguistic scale and triangular fuzzy numbers for evaluating ranking alternatives Table 8Linguistic Scale for evaluating the ranking alternatives.Linguistic TermExtremely PoorVery PoorPoorMedium PoorMediumMedium GoodGoodVery GoodExtremely GoodEPVP P MPMMGGVGEGTFN1,1,11,1,31,3,33,3,53,5,55,5,75,7,77,7,97,9,9 Table 8Linguistic Scale for evaluating the ranking alternatives.Linguistic TermExtremely PoorVery PoorPoorMedium PoorMediumMedium GoodGoodVery GoodExtremely GoodEPVP P MPMMGGVGEGTFN1,1,11,1,31,3,33,3,53,5,55,5,75,7,77,7,97,9,9 Linguistic Scale for evaluating the ranking alternatives. Table 8 Linguistic Scale for Evaluating the Ranking Alternatives presents the fuzzy linguistic terms and their corresponding triangular fuzzy numbers (TFN) used to assess the ranking of machining alternatives in the Fuzzy MARCOS methodology. These linguistic terms, ranging from “Extremely Poor” (EP) to “Extremely Good” (EG), allow for a nuanced evaluation of the alternatives based on subjective judgment. For instance, “Extremely Poor” is assigned the TFN (1,1,1), indicating a consistently low evaluation, while “Extremely Good” corresponds to the TFN (7,9,9), reflecting a highly favorable evaluation. The other terms, such as “Very Poor” (1,1,3), “Poor” (1,3,3), “Medium Poor” (3,3,5), and others, provide a spectrum of evaluations, facilitating the ranking process. These fuzzy values are used in the multi-criteria decision-making framework to assign scores to different alternatives based on their performance across multiple machining criteria. This approach aids in addressing uncertainty and imprecision in the decision-making process, aligning with the objectives of optimizing machining parameters for WAAM-fabricated bimetallic materials and improving the overall machinability of the components. Evaluation matrix of alternatives for ranking based on criteria performance Table 9Evaluation of machining alternatives based on criteria performance.Alternative1st criteria4th criteriaEPVP P MPMMGGVGEGEPVP P MPMMGGVGEGA1001224411012453000A2024531000001344300A3002552100001651200A4045421000001234500A5000414600025530000A6002453010002453100A7000134520126221000A8012552000023622000A9221224200000244500A10000234420012453000A11000024621046321000A12003453000000334320A13002453010002543100A14005322200000235500A15003453000036431000A16006342000000255300A17004623000023721000 Table 9Evaluation of machining alternatives based on criteria performance.Alternative1st criteria4th criteriaEPVP P MPMMGGVGEGEPVP P MPMMGGVGEGA1001224411012453000A2024531000001344300A3002552100001651200A4045421000001234500A5000414600025530000A6002453010002453100A7000134520126221000A8012552000023622000A9221224200000244500A10000234420012453000A11000024621046321000A12003453000000334320A13002453010002543100A14005322200000235500A15003453000036431000A16006342000000255300A17004623000023721000 Evaluation of machining alternatives based on criteria performance. Table 9 Evaluation of Alternatives Based on Criteria Performance presents the evaluation of 17 machining alternatives (A1 to A17) across 8 linguistic criteria ranging from “Extremely Poor” (EP) to “Extremely Good” (EG) for two different sets of criteria. Each alternative is assessed on its performance against these criteria, with numerical values indicating the degree of match for each linguistic term. For example, A1’s performance across various criteria ranges from 0 (EP) to 4 (Good), with higher values indicating better alignment with preferred outcomes. Similarly, A2 shows varying performances, with the highest values in the “Medium” (M) and “Medium Good” (MG) categories for certain criteria. The performance for each alternative is presented as a sequence of values corresponding to each linguistic term, reflecting the degree to which each alternative meets the expected criteria. These evaluations provide a comprehensive comparison across all 17 alternatives, allowing for a detailed analysis of their performance in the decision-making process. The linguistic terms and numerical scores serve as a foundation for ranking the alternatives, which is essential for optimizing machining parameters in the context of WAAM-fabricated bimetallic materials. Integrated evaluation matrix for ranking alternatives based on criteria performance Table 10Integrated matrix.Alt.-1-10.5090.480.2990.3930.5860.284X1X4A139375.000252614.4852935697.14751568.5616875.00134491.23A2181.12153507.402944647.053413071.32104176.46439338.01A32716.82234583.4588173627.42984705.6740919.70224152.05A460.3731568.558346398.578921785.53188287.72615073.22A548713.929172568.5462939539.862481.002716.8229228.36A64150.01544647.0534232945.64153507.4044647.05205438.75A788045.176439338.01411429301.002020.12905.615357.65A81215.00013071.3188113665.8342405.003507.4068199.50A91125.0004546.633444557.007037733.65243078.40794056.11A1068199.500287614.24361207979.82311568.5616875.00134491.23A11173627.429900375.00002268945.000034.861568.5635940.18A121568.55829228.3574134491.228852827.11188287.721020929.29A134150.01544647.0534232945.64153507.4034583.46205438.75A14389.711414175.000042699.641448713.93243078.40939539.86A151568.55829228.3574134491.2288104.576075.00103750.38A16233.826917537.014452827.106029228.36173627.43671099.90A17905.607513584.1130104176.4579313.712716.8257639.10Min60.37381568.558346398.578920.1246905.60755357.6464Max173627.429900375.00002268945.000052827.1060188287.7201020929.2872 Table 10Integrated matrix.Alt.-1-10.5090.480.2990.3930.5860.284X1X4A139375.000252614.4852935697.14751568.5616875.00134491.23A2181.12153507.402944647.053413071.32104176.46439338.01A32716.82234583.4588173627.42984705.6740919.70224152.05A460.3731568.558346398.578921785.53188287.72615073.22A548713.929172568.5462939539.862481.002716.8229228.36A64150.01544647.0534232945.64153507.4044647.05205438.75A788045.176439338.01411429301.002020.12905.615357.65A81215.00013071.3188113665.8342405.003507.4068199.50A91125.0004546.633444557.007037733.65243078.40794056.11A1068199.500287614.24361207979.82311568.5616875.00134491.23A11173627.429900375.00002268945.000034.861568.5635940.18A121568.55829228.3574134491.228852827.11188287.721020929.29A134150.01544647.0534232945.64153507.4034583.46205438.75A14389.711414175.000042699.641448713.93243078.40939539.86A151568.55829228.3574134491.2288104.576075.00103750.38A16233.826917537.014452827.106029228.36173627.43671099.90A17905.607513584.1130104176.4579313.712716.8257639.10Min60.37381568.558346398.578920.1246905.60755357.6464Max173627.429900375.00002268945.000052827.1060188287.7201020929.2872 Integrated matrix. Table 10 Integrated Matrix for Evaluation of Alternatives Based on Criteria Performance presents the integrated evaluation of alternatives (A1 to A17) based on various criteria. Each alternative is assessed across different metrics, where numerical values represent the performance of the alternatives against the corresponding criteria. The values are calculated using a combination of fuzzy and numerical evaluation, with the matrix indicating the range of performance for each alternative. For instance, A1 has values such as 39,375 for one criterion and 935,697 for another, showing the variation in its performance across the factors. The matrix also includes the minimum and maximum values for each criterion, providing an overview of the range of performances across all alternatives. The integrated matrix facilitates a comprehensive comparison of alternatives, which serves as the foundation for ranking and selecting the optimal machining parameters in the context of WAAM-fabricated bimetallic materials, enabling decision-making based on multi-criteria optimization. Extended integrated matrix for evaluation and ranking of alternatives Table 11Extended integrated matrix.Alt.-1-10.5090.480.2990.3930.5860.284X1X4A (AI)1114315.811114315.811114315.81420681.37420681.37420681.37A139375.00252614.49935697.151568.5616875.00134491.23A2181.123507.4044647.0513071.32104176.46439338.01A32716.8234583.46173627.434705.6740919.70224152.05A460.371568.5646398.5821785.53188287.72615073.22A548713.93172568.55939539.8681.002716.8229228.36A64150.0244647.05232945.643507.4044647.05205438.75A788045.18439338.011429301.0020.12905.615357.65A81215.0013071.32113665.83405.003507.4068199.50A91125.004546.6344557.0137733.65243078.40794056.11A1068199.50287614.241207979.821568.5616875.00134491.23A11173627.43900375.002268945.0034.861568.5635940.18A121568.5629228.36134491.2352827.11188287.721020929.29A134150.0244647.05232945.643507.4034583.46205438.75A14389.7114175.0042699.6448713.93243078.40939539.86A151568.5629228.36134491.23104.576075.00103750.38A16233.8317537.0152827.1129228.36173627.43671099.90A17905.6113584.11104176.46313.712716.8257639.10A (ID)16009.1716009.1716009.172094.462094.462094.46 Table 11Extended integrated matrix.Alt.-1-10.5090.480.2990.3930.5860.284X1X4A (AI)1114315.811114315.811114315.81420681.37420681.37420681.37A139375.00252614.49935697.151568.5616875.00134491.23A2181.123507.4044647.0513071.32104176.46439338.01A32716.8234583.46173627.434705.6740919.70224152.05A460.371568.5646398.5821785.53188287.72615073.22A548713.93172568.55939539.8681.002716.8229228.36A64150.0244647.05232945.643507.4044647.05205438.75A788045.18439338.011429301.0020.12905.615357.65A81215.0013071.32113665.83405.003507.4068199.50A91125.004546.6344557.0137733.65243078.40794056.11A1068199.50287614.241207979.821568.5616875.00134491.23A11173627.43900375.002268945.0034.861568.5635940.18A121568.5629228.36134491.2352827.11188287.721020929.29A134150.0244647.05232945.643507.4034583.46205438.75A14389.7114175.0042699.6448713.93243078.40939539.86A151568.5629228.36134491.23104.576075.00103750.38A16233.8317537.0152827.1129228.36173627.43671099.90A17905.6113584.11104176.46313.712716.8257639.10A (ID)16009.1716009.1716009.172094.462094.462094.46 Extended integrated matrix. Table 11 Extended Integrated Matrix for Evaluation and Ranking of Alternatives presents the extended evaluation of 17 machining alternatives (A1 to A17) based on various criteria, including additional considerations for A (AI) and A (ID). Each alternative is assessed across multiple metrics, with numerical values representing performance in relation to the established criteria. For example, A1 has values such as 39,375 for one criterion and 935,697 for another, while A2 shows varying performance across the different criteria. The matrix also includes values for A (AI) and A (ID), which are calculated based on the aggregated performance of all alternatives, providing a comprehensive evaluation. The minimum and maximum values for each criterion are provided, offering a range of the overall performance across all alternatives. This extended integrated matrix is essential for the further analysis and ranking of alternatives, facilitating the decision-making process in optimizing machining parameters for WAAM-fabricated bimetallic materials. Normalized performance matrix for ranking alternatives Table 12Normalized matrix.Alt.-1-10.5090.480.2990.3930.5860.284X1X4 A (AI) 0.01440.01440.01440.00500.00500.0050 A1 0.01710.06340.40660.01560.12411.3353 A2 0.35864.564488.38910.00480.02010.1602 A3 0.09220.46295.89260.00930.05120.4451 A4 0.345010.2063265.16740.00340.01110.0961 A5 0.01700.09280.32860.07170.770925.8575 A6 0.06870.35863.85760.01020.04690.5972 A7 0.01120.03640.18180.39092.3128104.0745 A8 0.14081.224813.17630.03070.59725.1715 A9 0.35933.521114.23040.00260.00860.0555 A10 0.01330.05570.23470.01560.12411.3353 A11 0.00710.01780.09220.05831.335360.0875 A12 0.11900.547710.20630.00210.01110.0396 A13 0.06870.35863.85760.01020.06060.5972 A14 0.37491.129441.07960.00220.00860.0430 A15 0.11900.547710.20630.02020.344820.0292 A16 0.30300.912968.46590.00310.01210.0717 A17 0.15371.178517.67780.03630.77096.6764 A (ID) 1.000001.000001.000001.000001.000001.00000 Table 12Normalized matrix.Alt.-1-10.5090.480.2990.3930.5860.284X1X4 A (AI) 0.01440.01440.01440.00500.00500.0050 A1 0.01710.06340.40660.01560.12411.3353 A2 0.35864.564488.38910.00480.02010.1602 A3 0.09220.46295.89260.00930.05120.4451 A4 0.345010.2063265.16740.00340.01110.0961 A5 0.01700.09280.32860.07170.770925.8575 A6 0.06870.35863.85760.01020.04690.5972 A7 0.01120.03640.18180.39092.3128104.0745 A8 0.14081.224813.17630.03070.59725.1715 A9 0.35933.521114.23040.00260.00860.0555 A10 0.01330.05570.23470.01560.12411.3353 A11 0.00710.01780.09220.05831.335360.0875 A12 0.11900.547710.20630.00210.01110.0396 A13 0.06870.35863.85760.01020.06060.5972 A14 0.37491.129441.07960.00220.00860.0430 A15 0.11900.547710.20630.02020.344820.0292 A16 0.30300.912968.46590.00310.01210.0717 A17 0.15371.178517.67780.03630.77096.6764 A (ID) 1.000001.000001.000001.000001.000001.00000 Normalized matrix. Table 12 Normalized Matrix for Evaluation of Alternatives presents the normalized performance values of 17 machining alternatives (A1 to A17) across various criteria. The normalization process transforms the original values into a common scale, facilitating the comparison of alternatives on an equal basis. Each alternative is evaluated based on its performance relative to the criteria, with the normalized values ranging from 0 to 1. For example, A1 has normalized values such as 0.0171 for the first criterion and 1.3353 for the fourth criterion. The normalization allows for the direct comparison of each alternative’s performance, with values like 1.0000 assigned to the aggregated criteria for A (ID), representing the reference standard for the evaluation. The normalized matrix helps to quantify the relative strengths and weaknesses of each alternative, supporting the decision-making process in selecting optimal machining parameters for WAAM-fabricated bimetallic materials. Weighted normalized matrix for evaluation and ranking of alternatives Table 13Weighted normalized matrix.Alt.X1X4A (AI)0.00730.00690.00430.00200.00290.0014A10.00870.03040.12160.00610.07270.3792A20.18252.190926.42830.00190.01180.0455A30.04690.22221.76190.00370.03000.1264A40.17564.899079.28500.00130.00650.0273A50.00870.04450.09830.02820.45187.3435A60.03500.17211.15340.00400.02750.1696A70.00570.01750.05440.15361.355329.5572A80.07170.58793.93970.01210.34991.4687A90.18291.69014.25490.00100.00500.0158A100.00670.02670.07020.00610.07270.3792A110.00360.00850.02760.02290.782517.0648A120.06060.26293.05170.00080.00650.0113A130.03500.17211.15340.00400.03550.1696A140.19080.542112.28280.00090.00500.0122A150.06060.26293.05170.00790.20205.6883A160.15430.438220.47130.00120.00710.0204A170.07820.56575.285700.01430.45181.8961A (ID)0.50900.48000.29900.39300.58600.2840 Table 13Weighted normalized matrix.Alt.X1X4A (AI)0.00730.00690.00430.00200.00290.0014A10.00870.03040.12160.00610.07270.3792A20.18252.190926.42830.00190.01180.0455A30.04690.22221.76190.00370.03000.1264A40.17564.899079.28500.00130.00650.0273A50.00870.04450.09830.02820.45187.3435A60.03500.17211.15340.00400.02750.1696A70.00570.01750.05440.15361.355329.5572A80.07170.58793.93970.01210.34991.4687A90.18291.69014.25490.00100.00500.0158A100.00670.02670.07020.00610.07270.3792A110.00360.00850.02760.02290.782517.0648A120.06060.26293.05170.00080.00650.0113A130.03500.17211.15340.00400.03550.1696A140.19080.542112.28280.00090.00500.0122A150.06060.26293.05170.00790.20205.6883A160.15430.438220.47130.00120.00710.0204A170.07820.56575.285700.01430.45181.8961A (ID)0.50900.48000.29900.39300.58600.2840 Weighted normalized matrix. Table 13 Weighted Normalized Matrix for Ranking Alternatives presents the weighted performance values of 17 machining alternatives (A1 to A17) across various criteria, derived from the normalized matrix. The weights are applied to the normalized values, resulting in a weighted score for each alternative under each criterion. For example, A1’s weighted scores across different criteria are 0.0087, 0.0304, and 0.1216, reflecting its performance in relation to the weights assigned to each criterion. Similarly, A (ID) serves as the reference with its own weighted values for comparison. This weighted normalized matrix enables a more refined evaluation of the alternatives by factoring in the relative importance of each criterion, facilitating more accurate decision-making in the optimization process for WAAM-fabricated bimetallic materials. The matrix helps identify the best-performing alternatives based on the weighted contributions of each criterion, further supporting the selection of the optimal machining parameters. Fuzzy MARCOS-based evaluation and ranking of machining alternatives Table 14Ranking of machining alternatives.Alt.SiFuzzy Ki-Fuzzy Ki+TiA (AI)0.02730.02490.0157A10.213.48107.5213.49140.133934.500.181.7754.6113.67141.903989.11A20.202.2726.6612.5391.24975.690.171.1513.5412.7092.39989.23A30.090.433.025.5817.14110.620.070.221.545.6517.36112.16A40.265.3280.0016.68213.822927.480.222.7040.6316.91216.522968.11A50.111.3843.146.7255.501578.700.090.7021.916.8156.201600.61A60.050.251.513.2710.0355.380.040.130.773.3110.1656.15A70.283.3341.5117.94133.801519.030.241.6921.0818.18135.481540.11A80.141.077.828.7142.99286.240.120.543.978.8343.53290.21A90.252.408.5615.6896.68313.400.211.224.3515.8997.89317.75A100.182.3183.3511.4393.083050.130.151.1742.3311.5894.253092.46A110.050.9217.443.2937.05638.160.040.478.863.3337.52647.01A120.221.7823.6713.9971.58865.990.190.9012.0214.1772.49878.01A130.080.382.745.1515.11100.360.070.191.395.2215.30101.75A140.343.9274.0821.88157.732710.790.291.9937.6222.18159.722748.41A150.100.579.076.2222.94331.880.080.294.616.3123.23336.49A160.190.5821.4511.9323.45785.040.160.3010.9012.0923.75795.94A170.141.227.618.7149.05278.440.120.623.868.8249.67282.30 Table 14Ranking of machining alternatives.Alt.SiFuzzy Ki-Fuzzy Ki+TiA (AI)0.02730.02490.0157A10.213.48107.5213.49140.133934.500.181.7754.6113.67141.903989.11A20.202.2726.6612.5391.24975.690.171.1513.5412.7092.39989.23A30.090.433.025.5817.14110.620.070.221.545.6517.36112.16A40.265.3280.0016.68213.822927.480.222.7040.6316.91216.522968.11A50.111.3843.146.7255.501578.700.090.7021.916.8156.201600.61A60.050.251.513.2710.0355.380.040.130.773.3110.1656.15A70.283.3341.5117.94133.801519.030.241.6921.0818.18135.481540.11A80.141.077.828.7142.99286.240.120.543.978.8343.53290.21A90.252.408.5615.6896.68313.400.211.224.3515.8997.89317.75A100.182.3183.3511.4393.083050.130.151.1742.3311.5894.253092.46A110.050.9217.443.2937.05638.160.040.478.863.3337.52647.01A120.221.7823.6713.9971.58865.990.190.9012.0214.1772.49878.01A130.080.382.745.1515.11100.360.070.191.395.2215.30101.75A140.343.9274.0821.88157.732710.790.291.9937.6222.18159.722748.41A150.100.579.076.2222.94331.880.080.294.616.3123.23336.49A160.190.5821.4511.9323.45785.040.160.3010.9012.0923.75795.94A170.141.227.618.7149.05278.440.120.623.868.8249.67282.30 Ranking of machining alternatives. Table 14 Results of Fuzzy MARCOS for Ranking Alternatives presents the fuzzy MARCOS (Multi-criteria Aggregated Ranking with Compromise Solution) results for the evaluation and ranking of 17 machining alternatives (A1 to A17). The table includes values such as Si, Fuzzy Ki- (lower fuzzy bound), Fuzzy Ki+ (upper fuzzy bound), and Ti (total index). These fuzzy values reflect the relative performance of each alternative across the evaluated criteria, with higher values indicating better performance. For instance, A1 has an Si value of 0.21, Fuzzy Ki- values ranging from 3.48 to 107.52, and Ti values between 13.49 and 141.90, indicating its ranking within the alternatives. Similarly, each alternative is evaluated with respect to fuzzy lower and upper bounds, facilitating a robust comparison. The results help identify the best-performing alternatives based on the integration of fuzzy logic and multi-criteria decision-making, supporting the selection of optimal machining parameters for WAAM-fabricated bimetallic materials. Final ranking of alternatives based on fuzzy MARCOS Table 15Ranking alternatives.Fuzzy F(K-)Fuzzy F(K+)Crisp K-CrispK+Crisp F(K-)CrispF(K+)(1-f(K-))/f(K-)(1-f(K+))/f(K+)f(K)Ranking0.00020.00230.07170.020.185.17751.4210.310.010.9972.890.0110.3110.00020.00150.01780.020.121.28225.533.050.000.30248.602.380.9170.00010.00030.0020.010.020.1530.790.410.000.041846.1823.740.02150.00030.00350.05330.020.283.84633.248.610.010.8387.510.207.2420.00010.00090.02880.010.072.07301.244.130.010.40183.291.531.6460.00010.00020.0010.000.010.0716.460.220.000.023466.4545.270.00170.00030.00220.02770.020.181.99345.364.680.010.45161.821.212.1350.00020.00070.00520.010.060.3877.821.040.000.10729.418.790.11130.00030.00160.00570.020.130.41119.301.570.000.16483.395.390.25110.00020.00150.05560.020.124.00572.317.860.010.7595.870.335.9730.00010.00060.01160.000.050.84131.611.800.000.17423.374.790.31100.00020.00120.01580.020.091.14194.392.640.000.26287.972.920.6880.00010.00030.00180.010.020.1327.650.370.000.042054.5826.540.01160.00040.00260.04940.030.213.56560.607.650.010.7498.640.365.6840.00010.00040.0060.010.030.4471.640.970.0010.09780.789.630.09140.00020.00040.01430.020.031.03148.462.040.0010.19372.484.130.4090.00020.00080.00510.01140.06440.36580.561.070.00140.105707.048.450.1112 Table 15Ranking alternatives.Fuzzy F(K-)Fuzzy F(K+)Crisp K-CrispK+Crisp F(K-)CrispF(K+)(1-f(K-))/f(K-)(1-f(K+))/f(K+)f(K)Ranking0.00020.00230.07170.020.185.17751.4210.310.010.9972.890.0110.3110.00020.00150.01780.020.121.28225.533.050.000.30248.602.380.9170.00010.00030.0020.010.020.1530.790.410.000.041846.1823.740.02150.00030.00350.05330.020.283.84633.248.610.010.8387.510.207.2420.00010.00090.02880.010.072.07301.244.130.010.40183.291.531.6460.00010.00020.0010.000.010.0716.460.220.000.023466.4545.270.00170.00030.00220.02770.020.181.99345.364.680.010.45161.821.212.1350.00020.00070.00520.010.060.3877.821.040.000.10729.418.790.11130.00030.00160.00570.020.130.41119.301.570.000.16483.395.390.25110.00020.00150.05560.020.124.00572.317.860.010.7595.870.335.9730.00010.00060.01160.000.050.84131.611.800.000.17423.374.790.31100.00020.00120.01580.020.091.14194.392.640.000.26287.972.920.6880.00010.00030.00180.010.020.1327.650.370.000.042054.5826.540.01160.00040.00260.04940.030.213.56560.607.650.010.7498.640.365.6840.00010.00040.0060.010.030.4471.640.970.0010.09780.789.630.09140.00020.00040.01430.020.031.03148.462.040.0010.19372.484.130.4090.00020.00080.00510.01140.06440.36580.561.070.00140.105707.048.450.1112 Ranking alternatives. Table 15 Ranking Alternatives Based on Fuzzy MARCOS Results presents the final ranking of the 17 machining alternatives based on various fuzzy and crisp metrics, including Fuzzy F(K-), Fuzzy F(K+), Crisp K-, Crisp K+, and other derived values such as (1 - f(K-))/f(K-), (1 - f(K+))/f(K+), and the total fuzzy score (f(K)). The table facilitates the comparison of the alternatives by providing fuzzy lower (Fuzzy F(K-)) and upper (Fuzzy F(K+)) bounds, along with crisp values that help identify the best-performing alternatives. For instance, Alternative 1 (A1) ranks first with a high crisp score of 10.31, indicating its strong performance across the criteria[42–46]. The ranking is derived based on the comprehensive analysis of fuzzy lower and upper bounds, as well as the crisp evaluation scores, with higher values representing better performance. This ranking is essential for selecting the optimal machining parameters and making informed decisions in the context of WAAM-fabricated bimetallic materials[42–50]. The influence of machining parameters on surface integrity and cutting performance was evaluated through controlled experimental runs varying cutting speed, feed rate, depth of cut, and number of flutes. In Run 1 (9500 rpm, 0.1 mm/tooth, 1.5 mm depth of cut, four flutes), superior performance was achieved, yielding a low surface roughness (Ra) of 0.6 μm, a material removal rate (MRR) of 35.6 mm³/min, power consumption (Pc) of 320 W, and cutting force (CF) of 120 N, indicating an optimal balance between productivity and surface quality. In contrast, Run 6 (8750 rpm, 0.175 mm/tooth, 1 mm depth of cut, two flutes) resulted in comparatively inferior performance, with increased surface roughness (0.8 μm), slightly reduced MRR (34 mm³/min), and marginally lower power and cutting force values (318 W and 119 N, respectively). The results suggest that higher cutting speed combined with lower feed rate and increased flute number enhances surface finish and material removal efficiency, while maintaining stable cutting forces. The improved performance observed under higher spindle speed and multi-flute configuration can be attributed to reduced chip load per tooth, improved chip evacuation, and minimized tool–workpiece interaction instability, thereby confirming the sensitivity of machining performance to process parameter optimization. The optimized machining condition provides practical benefits by balancing surface quality, productivity, energy demand, and cutting stability. Lower Ra improves the surface finish and reduces the need for additional finishing operations. Higher MRR improves productivity, while lower Pc supports energy-efficient machining. Controlled CF indicates reduced mechanical loading and improved process stability. Therefore, the selected optimal condition is not only statistically favourable but also practically useful for the post-processing of WAAM-fabricated Ni–SS bimetallic components. The optimized machining condition can be interpreted based on the microstructural characteristics of the WAAM-fabricated Ni–SS bimetallic material. The heterogeneous structure, consisting of dendritic regions near the substrate, mixed cellular–dendritic zones, and refined equiaxed grains toward the top, results in spatial variation in hardness, thermal conductivity, and deformation behavior. In addition, the Ni-rich region exhibits higher strength and cutting resistance, while the stainless-steel-rich region is more prone to adhesion and strain hardening. At the bimetallic interface, the compositional gradient creates variable tool–workpiece interaction conditions. Under these conditions, high cutting speed promotes smoother shearing and reduces built-up edge formation, while low feed rate minimizes feed marks and surface roughness. The use of a four-flute tool distributes the cutting load across multiple edges, reducing force concentration and improving stability. Therefore, the optimized parameter combination can be physically explained as a condition that stabilizes chip formation and reduces force and energy fluctuations across the heterogeneous Ni–SS interface, leading to improved surface quality and machining performance. Validation of product quality Surface finish quality (Ra) Fig. 15Surface Roughness Conditions, (a) Best, (b) Worse. Surface Roughness Conditions, (a) Best, (b) Worse. Figure 15 illustrates the surface roughness profiles of bimetallic Wire Arc Additive Manufacturing (WAAM) samples after face milling under the best in Fig. 15(a) and worse Fig. 15(b) machining conditions. The best condition (Run 1: cutting speed = 9500 rpm, feed rate = 0.1 mm/tooth, depth of cut = 1.5 mm, four-flute cutter) produced a smoother surface with Ra ≈ 0.6 μm along with a stable material removal rate (35.6 mm³/min), power consumption of 320 W, and cutting force of 120 N. The improved finish is attributed to the higher spindle speed combined with a lower feed per tooth and increased flute engagement, which facilitated uniform chip formation, reduced force variation, and minimized surface damage across the bimetallic interface. Conversely, the worse condition (Run 6: 8750 rpm, 0.175 mm/tooth, 1 mm depth of cut, two-flute tool) resulted in a higher roughness value (Ra ≈ 0.8 μm) despite comparable MRR and power values, indicating that the increased feed rate and reduced flute count elevated the load per cutting edge, promoting vibration, ploughing, and irregular material removal within the heterogeneous WAAM microstructure. Overall, Fig. 15 confirms that higher cutting speed, lower feed rate, greater flute engagement, and adequate depth of cut contribute to improved surface integrity by stabilizing cutting forces and mitigating microstructural heterogeneity effects in bimetallic additively manufactured components. Chip morphology Fig. 16Chip Morphology Conditions, (a) Best, (b) Worse. Chip Morphology Conditions, (a) Best, (b) Worse. Figure 16 presents the chip morphology of bimetallic WAAM samples during face milling under the best and worse machining conditions corresponding to the selected parameter sets. Under the best condition (Run 1: cutting speed = 9500 rpm, feed rate = 0.1 mm/tooth, depth of cut = 1.5 mm, four-flute cutter), the chips (Fig. 16(a) appear more continuous and uniformly curled, indicating stable shearing action and consistent chip flow. The lower surface roughness (Ra ≈ 0.6 μm), higher material removal rate (35.6 mm³/min), and moderate power and cutting force values (320 W and 120 N) suggest efficient energy utilization and reduced tool load per cutting edge due to multi-flute engagement. The formation of regular ribbon-like chips reflects controlled plastic deformation and limited adhesion at the tool–workpiece interface, which is beneficial for maintaining surface integrity in heterogeneous bimetallic WAAM structures. In contrast, the worse condition (Run 6: 8750 rpm, 0.175 mm/tooth, 1 mm depth of cut, two-flute tool) produced fragmented and irregular chips (Fig. 16(b), signifying unstable cutting characterized by higher instantaneous chip thickness and increased load per flute. The elevated feed rate combined with reduced flute count promoted intermittent shearing, chip breakage, and localized tearing, which correlates with the higher roughness value (Ra ≈ 0.8 μm) despite similar MRR and power levels. Overall, Fig. 16 demonstrates that higher cutting speed, lower feed rate, greater flute engagement, and adequate depth of cut facilitate continuous chip formation and stable material removal, thereby mitigating the influence of microstructural heterogeneity in bimetallic additively manufactured components. Tool life assessment Fig. 17Tool Life Assessment Conditions, (a) Best, (b) Worse. Tool Life Assessment Conditions, (a) Best, (b) Worse. Figure 17 presents the tool wear characteristics observed during face milling of bimetallic WAAM samples under the optimized (best) and non-optimal (worse) machining conditions, highlighting the influence of cutting speed, feed rate, depth of cut, and flute configuration on tool life. Under the best condition (Run 1: 9500 rpm, 0.1 mm/tooth, 1.5 mm depth of cut, four-flute cutter), the tool edge shown in Fig. 17(a) exhibits relatively minor flank wear and limited edge chipping, indicating stable cutting and uniform load distribution across multiple flutes. The lower feed rate reduced instantaneous chip thickness, while the higher flute engagement dispersed cutting forces, thereby minimizing thermal concentration and adhesion wear. This condition aligns with the improved surface roughness (Ra ≈ 0.6 μm) and continuous chip morphology reported earlier, confirming efficient shearing and controlled tool–workpiece interaction that prolongs tool life. In contrast, the worse condition (Run 6: 8750 rpm, 0.175 mm/tooth, 1 mm depth of cut, two-flute tool) displays pronounced edge degradation in Fig. 17(b), including micro-chipping and abrasive wear marks along the cutting edge. The increased feed rate elevated chip load per tooth, while the reduced flute count intensified force concentration and vibration, accelerating mechanical and thermal wear mechanisms. Furthermore, the heterogeneous microstructure and hardness mismatch inherent to bimetallic WAAM deposits promote intermittent cutting and adhesion–delamination cycles, which contribute to rapid tool deterioration despite similar power consumption and material removal rates. Overall, Fig. 17 validates that optimized machining parameters not only enhance surface integrity and chip formation but also significantly improve tool life by stabilizing cutting forces, reducing thermal stress, and mitigating abrasive interaction with the anisotropic WAAM microstructure. These findings emphasize that experimental tool life assessment is essential to confirm that model-predicted optimal settings achieve sustainable machining performance without excessive tool wear or premature failure. Advanced validation for RSM predictive models k-fold cross-validation Table 165-fold cross-validation results for RSM models.ResponseRa (µm)MRR (mm³/min)Pc (W)CF (N)Mean MAE0.04852.07013.70823.4362Mean RMSE0.05632.44324.74254.0093 Table 165-fold cross-validation results for RSM models.ResponseRa (µm)MRR (mm³/min)Pc (W)CF (N)Mean MAE0.04852.07013.70823.4362Mean RMSE0.05632.44324.74254.0093 5-fold cross-validation results for RSM models. A five-fold cross-validation procedure was employed to assess the generalization capability and robustness of the developed RSM models beyond the CCD design space in Table 16. The experimental dataset comprising 17 machining trials was randomly divided into five subsets, where in each iteration approximately 80% of the data were used for model training and the remaining 20% for validation, ensuring that each subset was tested once. Independent RSM models were developed for surface roughness (Ra), material removal rate (MRR), power consumption (Pc), and cutting force (CF) using cutting speed, feed rate, depth of cut, and flute number as predictors, and model performance was evaluated using mean absolute error (MAE) and root mean square error (RMSE). The cross-validation results revealed consistently low prediction errors, with mean RMSE values of 0.056 μm for Ra, 2.44 mm³/min for MRR, 4.74 W for Pc, and 4.01 N for CF, indicating stable predictive accuracy without fold-dependent bias or overfitting. The comparatively lower error in Ra reflects the strong deterministic influence of kinematic and geometric cutting parameters on surface formation, whereas the slightly higher yet acceptable errors in MRR, Pc, and CF are attributed to their sensitivity to nonlinear chip load variation, thermal effects, and dynamic tool–workpiece interactions inherent to WAAM bimetallic machining. Importantly, the absence of error amplification across validation folds confirms that the models capture the underlying machining physics rather than merely fitting CCD boundary points. Overall, the cross-validation analysis provides strong statistical and physical justification for the reliability of the RSM models, validates their suitability for integration with the fuzzy MCDM optimization framework, and confirms that the proposed hybrid methodology can support robust, uncertainty-aware decision-making in advanced machining applications. Leave-one-out cross-validation (LOOCV) Table 17LOOCV Results (n = 17).ResponseLOOCV MAELOOCV RMSEPRESSCV-R²Ra (µm)0.03830.05230.04650.7323MRR (mm³/min)1.84742.420699.61050.6013Pc (W)3.58694.7549384.36030.6429CF (N)2.59773.4099197.66340.4508 Table 17LOOCV Results (n = 17).ResponseLOOCV MAELOOCV RMSEPRESSCV-R²Ra (µm)0.03830.05230.04650.7323MRR (mm³/min)1.84742.420699.61050.6013Pc (W)3.58694.7549384.36030.6429CF (N)2.59773.4099197.66340.4508 LOOCV Results (n = 17). Leave-One-Out Cross-Validation (LOOCV) was employed to rigorously assess the robustness and generalization capability of the developed RSM models under data-sparse conditions, which is critical for experimental datasets limited to 17 trials. In this procedure, each experimental run was sequentially excluded, the model was trained on the remaining sixteen runs, and the excluded response was predicted, with errors aggregated across all iterations in Table 17. The LOOCV results demonstrated low and stable prediction errors for all responses, with RMSE values of 0.052 μm for surface roughness (Ra), 2.42 mm³/min for material removal rate (MRR), 4.75 W for power consumption (Pc), and 3.41 N for cutting force (CF), accompanied by satisfactory cross-validated coefficients of determination (CV-R²) of 0.73, 0.60, 0.64, and 0.45, respectively. The superior LOOCV performance of the Ra model reflects the strong deterministic dependence of surface finish on cutting kinematics and tool geometry, while the comparatively lower CV-R² values for MRR, Pc, and CF are attributed to their sensitivity to nonlinear chip formation, thermal effects, and dynamic tool–workpiece interactions inherent in WAAM-fabricated bimetallic machining. Importantly, the absence of large PRESS values or error amplification confirms that the models are not overfitted to individual CCD points and remain stable when subjected to extreme data exclusion. Overall, the LOOCV analysis provides strong statistical and physical evidence of model robustness, validates predictive reliability under severe data sparsity, and justifies the integration of the RSM models with the fuzzy MCDM framework for uncertainty-aware machinability optimization in advanced manufacturing applications. Prediction error metrics Table 18Prediction error metrics for RSM models.ResponseMAERMSEMAPE (%)PRESSR²Ra (µm)0.02590.03413.447310.01980.8862MRR (mm³/min)1.23461.54143.675340.38830.8383Pc (W)2.39843.10440.7522163.82910.8478CF (N)1.73052.19621.456581.99580.9722 Table 18Prediction error metrics for RSM models.ResponseMAERMSEMAPE (%)PRESSR²Ra (µm)0.02590.03413.447310.01980.8862MRR (mm³/min)1.23461.54143.675340.38830.8383Pc (W)2.39843.10440.7522163.82910.8478CF (N)1.73052.19621.456581.99580.9722 Prediction error metrics for RSM models. Prediction error metrics were employed to comprehensively evaluate the accuracy, robustness, and predictive credibility of the developed RSM models for all machinability responses in Table 18. Using the full experimental dataset, regression models were constructed for surface roughness (Ra), material removal rate (MRR), power consumption (Pc), and cutting force (CF), and their performance was quantified using mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE), coefficient of determination (R²), and the prediction error sum of squares (PRESS). The results revealed consistently low prediction errors, with Ra exhibiting the highest accuracy (MAE = 0.0259 μm, RMSE = 0.0341 μm, MAPE = 3.45%, R² = 0.886), confirming strong deterministic control of surface finish by cutting kinematics and tool geometry. MRR showed slightly higher yet acceptable errors (RMSE = 1.54 mm³/min, MAPE = 3.68%, R² = 0.838), reflecting its nonlinear dependence on speed, feed, and depth of cut. Pc and CF demonstrated low relative errors (MAPE < 1% and < 1.5%, respectively) with satisfactory R² values, indicating reliable capture of energy and force-related machining behavior despite inherent dynamic and thermal variability in WAAM-fabricated bimetallic systems. Importantly, the low PRESS values across all responses confirm minimal residual dispersion and strong internal model consistency, thereby strengthening confidence in predictive stability. Overall, the combined error metrics provide strong statistical and physical justification for the reliability of the RSM models, validate their suitability for integration with the hybrid fuzzy MCDM optimization framework, and support their application for robust, uncertainty-aware decision-making in advanced machining of WAAM bimetallic materials. Uncertainty propagation Table 19Monte Carlo Simulation.ResponseMean PredictionStd. Dev.95% CI Lower95% CI UpperRa (µm)0.76980.04600.68610.8540MRR (mm³/min)34.13061.431231.556736.7227Pc (W)318.59432.8634313.3420323.8476CF (N)118.52491.6054115.6077121.4554 Table 19Monte Carlo Simulation.ResponseMean PredictionStd. Dev.95% CI Lower95% CI UpperRa (µm)0.76980.04600.68610.8540MRR (mm³/min)34.13061.431231.556736.7227Pc (W)318.59432.8634313.3420323.8476CF (N)118.52491.6054115.6077121.4554 Monte Carlo Simulation. Monte Carlo simulation was employed to quantify input–output uncertainty propagation and assess model sensitivity under stochastic variation of machining parameters in Table 19. A total of 10,000 random realizations were generated by introducing ± 5% uncertainty in cutting speed, feed rate, and depth of cut, while flute configuration was randomly assigned, and the trained RSM models were used to predict the corresponding responses. The probabilistic results indicate narrow standard deviations and well-defined 95% confidence intervals for all responses, with Ra exhibiting a standard deviation of 0.046 μm and a 95% confidence interval of 0.686–0.854 μm, confirming strong surface finish stability against moderate parameter fluctuations. Similarly, MRR, Pc, and CF showed limited dispersion, with confidence intervals of 31.56–36.72 mm³/min, 313.34–323.85 W, and 115.61–121.46 N, respectively, indicating robust energy and force response prediction. The narrow confidence bounds and absence of extreme outliers confirm low model sensitivity to small input perturbations and minimal uncertainty amplification. Scientifically, this behavior reflects the dominance of stable cutting mechanics and consistent chip formation regimes within the investigated parameter space. Overall, the Monte Carlo analysis provides strong probabilistic validation of the RSM models, complements deterministic error metrics and cross-validation results, and reinforces the suitability of the proposed hybrid optimization framework for reliable, uncertainty-aware decision-making in WAAM bimetallic machining. External validation Table 20Confirmation results.Run (Point)Ra ActualRa PredRa PE%MRR ActualMRR PredMRR PE%1 (Optimal)0.60.5685.340835.641.566116.75866 (Worst)0.80.79410.74283433.82670.5096 Run (Point) Pc Actual Pc Pred Pc PE% CF Actual CF Pred CF PE% 1 (Optimal)320328.16052.5502120125.26724.38946 (Worst)318317.70570.0925119117.90780.9178 Table 20Confirmation results.Run (Point)Ra ActualRa PredRa PE%MRR ActualMRR PredMRR PE%1 (Optimal)0.60.5685.340835.641.566116.75866 (Worst)0.80.79410.74283433.82670.5096 Run (Point) Pc Actual Pc Pred Pc PE% CF Actual CF Pred CF PE% 1 (Optimal)320328.16052.5502120125.26724.38946 (Worst)318317.70570.0925119117.90780.9178 Confirmation results. External validation through independent confirmation runs was carried out to verify the industrial reliability of the developed RSM models by comparing experimentally measured responses with model predictions at both optimal and non-optimal operating conditions in Table 20. In this procedure, selected confirmation points were excluded from model fitting, the regression models were recalibrated using the remaining experimental data, and the excluded responses were subsequently predicted and evaluated using percentage error as the validation metric. The results demonstrated close agreement between predicted and experimental values for surface roughness, power consumption, and cutting force, with prediction errors consistently within the accepTables 5, 6, 7, 8, 9 and 10% range, confirming strong predictive fidelity for surface quality, energy demand, and force response. The worst-condition confirmation run exhibited prediction errors below 1% for all responses, indicating excellent robustness of the models under adverse machining conditions. Although the material removal rate at the optimal condition showed a comparatively higher deviation, this behavior is attributed to the inherently nonlinear dependence of MRR on the combined effects of cutting speed, feed rate, and depth of cut, which is difficult to fully capture using linear regression under limited experimental data. Importantly, no systematic bias or instability was observed, and the overall confirmation results substantiate that the models maintain reliable predictive capability beyond the calibration dataset. From a scientific perspective, the strong agreement arises from the dominant influence of deterministic cutting kinematics on Ra, Pc, and CF, whereas the higher variability in MRR reflects multiplicative chip formation dynamics. Overall, the external validation confirms the practical applicability and industrial reliability of the developed RSM models and justifies their integration into the hybrid fuzzy MCDM optimization framework for robust, uncertainty-aware machining decision-making. Advanced validation for hybrid MCDM framework Sensitivity analysis with different approaches Fig. 18Sensitivity analysis with different approaches. Sensitivity analysis with different approaches. Figure 18, presents the sensitivity analysis of alternative rankings obtained using different FAHP-based hybrid decision-making approaches, namely FAHP–FMARCOS, FAHP–FGRA, FAHP–FVIKOR, FAHP–FTOPSIS, and FAHP–FELECTRE. The comparison shows a high level of agreement in the ranking patterns across all methods, particularly for the best- and worst-performing alternatives. Alternatives A1, A4, and A10 consistently occupy the lowest rank positions, while A6, A13, and A15 remain among the highest-ranked options irrespective of the applied method, indicating strong consensus at the extremes. Minor rank variations are observed for mid-ranked alternatives such as A3, A9, A11, and A16, which can be attributed to differences in aggregation mechanisms, normalization schemes, and compromise strategies inherent to each method. These variations remain limited in magnitude and do not alter the overall decision structure. The absence of abrupt rank reversals across approaches confirms that the ranking results are not method-dependent but are driven by the underlying performance of the alternatives. Overall, the strong convergence of rankings across multiple FAHP-integrated techniques demonstrates the robustness and stability of the proposed evaluation framework. This consistency reinforces the reliability of the decision outcomes and validates the suitability of the adopted hybrid methodology for multi-criteria evaluation under uncertainty. Sensitivity analysis with different scenarios Fig. 19Sensitivity analysis with different scenario. Sensitivity analysis with different scenario. Figure 19, illustrates the sensitivity analysis of alternative rankings under systematic variations in criteria weightages, considering ± 10%, ± 15%, and ± 20% perturbations from the baseline (current approach). The analysis shows that the overall ranking structure remains largely stable across all scenarios. High-ranked alternatives (A6, A13, and A15) consistently retain their positions, while low-ranked alternatives (A1, A4, and A10) remain unchanged, demonstrating strong robustness of the decision outcomes to weight fluctuations. Moderate rank shifts are observed for a few mid-performing alternatives, particularly A8, A9, and A11, as the level of perturbation increases. These variations are limited in magnitude and primarily occur among alternatives with closely competing performance values, rather than indicating instability of the model. Importantly, even under the most severe perturbation (± 20%), no extreme rank reversals are observed. The persistence of the optimal and least-preferred alternatives across all scenarios confirms that the proposed decision-making framework is not sensitive to reasonable changes in criteria importance. This robustness validates the reliability of the obtained rankings and supports the applicability of the model for decision-making under uncertainty in stakeholder preferences. Integrated hybrid model validation Coupled RSM–MCDM consistency check A coupled RSM–MCDM consistency check is performed to verify that the RSM-derived optimal region (continuous optimum from regression surfaces/desirability) corresponds to the top-ranked alternative produced by the MCDM method (e.g., Fuzzy MARCOS). The check is done by (i) identifying the RSM optimum (or high-desirability zone) in the factor space, (ii) selecting the closest experimental alternative(s) to that region, and (iii) confirming that these alternatives receive the highest MCDM utility score/rank; agreement indicates that the decision model is coherent with the underlying machining physics captured by RSM, while disagreement indicates inconsistent weighting, normalization, or dominance of one criterion. Using your two-point dataset as a demonstrator, criteria directions were defined as cost-type for Ra, Pc, and CF (lower is better) and benefit-type for MRR (higher is better). A standard ratio normalization was applied (benefit: value/max, cost: min/value) and an illustrative equal-weight aggregation was used to test coherence. Consistency demonstration using Run 1 (Best) vs. Run 6 (Worst) Table 21Coupled RSM–MCDM consistency analysis.RunRa (cost)MRR (benefit)Pc (cost)CF (cost)Composite utilityRank11.00001.00000.99380.99170.9964160.75000.95511.00001.00000.92632 Table 21Coupled RSM–MCDM consistency analysis.RunRa (cost)MRR (benefit)Pc (cost)CF (cost)Composite utilityRank11.00001.00000.99380.99170.9964160.75000.95511.00001.00000.92632 Coupled RSM–MCDM consistency analysis. A coupled RSM–MCDM consistency check was performed to verify physical decision coherence between the statistically optimized regions identified by Response Surface Methodology and the alternative rankings obtained through the multi-criteria decision-making framework. In this procedure, the RSM-derived optimal condition, characterized by low surface roughness and energy consumption with high material removal rate, was mapped to the closest experimental alternatives and evaluated using normalized cost-type (Ra, Pc, CF) and benefit-type (MRR) criteria in Table 21. The composite utility values obtained from the MCDM aggregation clearly ranked the RSM-optimal alternative (Run 1) as the top-performing solution, while the experimentally identified worst condition (Run 6) consistently occupied a lower rank. The close agreement between the RSM optimum and the highest MCDM utility score confirms that the decision model reinforces, rather than contradicts, the underlying machining physics captured by the regression surfaces. Scientifically, this coherence arises because the dominant process mechanisms—stable chip formation, reduced frictional interaction, and efficient material removal at higher cutting speed and optimal tool engagement—are simultaneously favored by both the statistical optimization and the decision-ranking criteria. The absence of rank conflict or reversal indicates balanced criterion weighting, correct benefit–cost classification, and consistent normalization. Overall, the coupled analysis validates the integrity of the hybrid framework, demonstrating that the decision-making layer faithfully translates the RSM-based physical optimization into reliable rankings, thereby strengthening confidence in the robustness and industrial applicability of the proposed hybrid RSM–MCDM methodology. Pareto front analysis A Pareto front analysis was conducted using the experimental responses surface roughness (Ra), material removal rate (MRR), power consumption (Pc), and cutting force (CF) to evaluate multi-objective trade-offs and to verify whether the selected optimal condition lies on or near the Pareto frontier. In this analysis, Ra, Pc, and CF were treated as cost-type objectives to be minimized, while MRR was treated as a benefit-type objective to be maximized. Based on dominance criteria, an experimental run is considered Pareto-optimal if no other run simultaneously improves all objectives without deterioration in at least one response. The comparison between the best condition (Run 1: Ra = 0.6 μm, MRR = 35.6 mm³/min, Pc = 320 W, CF = 120 N) and the worst condition (Run 6: Ra = 0.8 μm, MRR = 34 mm³/min, Pc = 318 W, CF = 119 N) shows that Run 1 dominates Run 6 in terms of surface quality and productivity, while maintaining comparable energy and force requirements. As a result, Run 1 lies on the Pareto frontier, whereas Run 6 is Pareto-dominated and excluded from the optimal solution set. Scientifically, this outcome reflects the ability of the optimal cutting speed, feed rate, depth of cut, and flute configuration to achieve a balanced compromise between surface integrity, material removal efficiency, and mechanical–thermal loading. The Pareto analysis therefore confirms that the selected optimal run represents a non-dominated, physically meaningful trade-off solution, reinforcing the validity of the RSM optimization and its consistency with the multi-criteria decision-making results, and demonstrating that the chosen operating condition is both statistically and practically optimal within the investigated parameter space. Fig. 20Pareto analysis. Pareto analysis. The Pareto parallel-coordinates analysis was employed to evaluate the four-response trade-offs among surface roughness (Ra), material removal rate (MRR), power consumption (Pc), and cutting force (CF), thereby identifying non-dominated machining solutions and validating the optimality of the selected operating condition in Fig. 20. In this procedure, Ra, Pc, and CF were treated as cost-type objectives and transformed into benefit-scaled metrics, while MRR was retained as a benefit-type response, enabling a unified comparison across criteria. Each experimental run was represented as a polyline across the four normalized axes, and Pareto-optimal solutions were distinguished from dominated alternatives. The resulting parallel-coordinates plot clearly reveals a distinct envelope formed by the Pareto front, within which the optimal run is located, characterized by simultaneously low Ra, Pc, and CF values combined with high MRR. In contrast, dominated runs exhibit pronounced trade-offs, with improvements in one response accompanied by deterioration in others, as reflected by intersecting and inward-shifted polylines. Scientifically, the smooth and compact spread of Pareto-optimal lines indicates stable cutting mechanics and efficient chip formation, arising from favorable combinations of cutting speed, feed rate, depth of cut, and flute configuration. The alignment of the RSM-identified optimum with the Pareto frontier confirms that the statistical optimization captures physically meaningful multi-objective trade-offs rather than isolated single-response extrema. Overall, the Pareto parallel-coordinates analysis provides strong visual and analytical confirmation of decision robustness, reinforces the coherence between regression-based optimization and multi-criteria ranking, and substantiates the industrial relevance of the proposed hybrid RSM–MCDM framework for reliable, multi-response machining optimization. Conclusion This study optimized the machinability of WAAM-fabricated Nickel–Stainless Steel bimetallic samples with silicon enhancement using a hybrid RSM–Spherical Fuzzy AHP–Fuzzy MARCOS framework. Mechanical characterization confirmed uniform hardness values of 202–220 HV, strong metallurgical bonding, and ductile tensile properties, with an ultimate tensile strength of approximately 450 MPa and elongation of 32–35%, indicating good structural integrity and favorable machinability. Response Surface Methodology identified cutting speed and depth of cut as important factors influencing surface roughness and material removal rate, while feed rate and flute configuration significantly affected power consumption and cutting force. The optimal machining condition of 9500 rpm cutting speed, 0.1 mm/tooth feed rate, 1.5 mm depth of cut, and a four-flute tool produced low surface roughness, high productivity, and stable energy-force characteristics. Statistical validation through ANOVA, residual diagnostics, actual–predicted comparison, and confirmation experiments verified the adequacy and predictive capability of the developed models. The agreement between RSM optimization and fuzzy MCDM ranking confirmed the robustness of the proposed hybrid framework. From an industrial perspective, the optimized condition can support improved surface integrity, enhanced productivity, reduced power consumption, and better cutting stability in the post-processing of WAAM-fabricated multi-material components. These benefits are relevant to aerospace, marine, energy, automotive, and tooling industries, where Ni–SS bimetallic structures may be used for components requiring corrosion resistance, thermal stability, mechanical strength, and dimensional accuracy. However, the study is limited to the selected Ni–SS bimetallic material with silicon content, the investigated face-milling parameter range, and a 17-run RSM-based design, which may not fully capture all higher-order interactions. Direct tool wear progression, chip morphology, cutting temperature, and interface-specific machining behavior were also not examined in detail. Future work should include expanded experimental designs, tool wear and chip morphology analysis, SEM/EDS evaluation of worn tools, interface-level force measurement, thermal analysis, and sustainability indicators such as energy per unit material removed and carbon footprint.