An efficient open-source framework for high-fidelity 3D surface topography and roughness prediction in milling
Hadi Bakhshan, Sima Farshbaf, Adrián Travieso-Disotuar, Luciano Mijaíl Villarreal, Fernando Rastellini Canela, Josep Maria Carbonell · 2026
Read An efficient open-source framework for high-fidelity 3D surface topography and roughness prediction in milling by Hadi Bakhshan, Sima Farshbaf, Adrián Travieso-Disotuar, Luciano Mijaíl Villarreal, Fernando Rastellini Canela, Josep Maria Carbonell (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-28160.txt.
Introduction
Bakhshan and five co-authors build an open-source tool that predicts milled surface texture in 3D. Training data-driven models needs large surface datasets, but measuring real parts is slow and costly. Their framework simulates the milling process and outputs surface topography plus roughness values. An optimized solver made the forward model 42 times faster on average. Checks against two experimental datasets show sound accuracy, so the tool can feed surrogate models at scale.
Catalog record
| CREATOR | Hadi Bakhshan, Sima Farshbaf, Adrián Travieso-Disotuar, Luciano Mijaíl Villarreal, Fernando Rastellini Canela, Josep Maria Carbonell |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57375 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.28160
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
An efficient open-source framework for high-fidelity 3D surface topography and roughness prediction in milling Hadi Bakhshan, Sima Farshbaf, Adrián Travieso-Disotuar, Luciano Mijaíl Villarreal, Fernando Rastellini Canela, Josep Maria Carbonell Abstract With the emergence of data-driven approaches in science, there is growing interest in their application to manufacturing, particularly in surface precision engineering. However, generating large datasets required for model training is often impractical experimentally due to high costs and the time-intensive nature of measurements. High-fidelity synthetic datasets offer a viable alternative if they can be generated both efficiently and accurately. To address this challenge, this paper presents an efficient framework for generating accurate 3D surface topographies and roughness indicators in milling operations using numerical methods. First, a conventional topography prediction model is developed based on the forward solution method (FSM). Building on this, an optimized computational algorithm is proposed to establish an efficient FSM with significantly improved performance. The model is validated against two independent sets of experimental results, assessing both prediction accuracy and computational efficiency. The results demonstrate acceptable prediction errors and an average computational speedup of 42.2×. The proposed open-source model provides a generalizable framework for large-scale analysis, enabling the generation of extensive datasets for data-driven surrogate modeling. 1 Introduction The rapid advancements of industries such as aerospace [ 1 ] , energy systems [ 2 ] , optics [ 3 ] , and semiconductor manufacturing [ 4 , 5 ] necessitate increasingly stringent requirements for the surface quality of critical components. Many high-performance parts operate in extreme environmental conditions characterized by high temperatures, moisture, and corrosive environments, which highlights the significant role of the surface integrity of the machined surface in determining component reliability and operational lifespan [ 6 , 7 , 8 ] . Surface topography, which describes the three-dimensional characteristics of a machined surface, directly influences key functional properties such as friction [ 9 ] , contact behavior [ 10 ] , wear resistance [ 11 , 12 , 13 ] , and fatigue performance [ 14 , 15 ] . Therefore, surface roughness and related topographical features are commonly employed as important indicators of machining quality. The milling process is a widely used subtractive manufacturing method due to its versatility, productivity, and economic advantages [ 16 ] . However, achieving a target surface quality in milling typically requires numerous experimental trials, which can be both time-consuming and costly [ 8 ] . Consequently, reliable predictive approaches for surface topography and roughness are essential to improve process understanding and enable more efficient selection of machining parameters [ 17 , 18 ] . Several methods have been proposed to predict surface roughness and topography, which can be broadly categorized into four main groups. First, empirical methods aim to establish statistical relationships between machining parameters and surface quality indicators through regression analysis based on experimentally measured data. Various linear, polynomial, or power-law relationships are formulated to correlate surface roughness parameters with machining variables such as cutting conditions, tool parameters, workpiece properties, and environmental factors [ 19 ] . However, approaches that rely heavily on experimental observations are often limited in generality and dependent on significant resources. This makes them inefficient for covering a wide range of process conditions, especially considering the large number of influential parameters involved in machining processes. Second, analytical and mechanistic approaches describe surface generation based on the kinematics of the cutting tool, tool geometry, and machining parameters. These models capture the relative motion between the cutter and the workpiece, where the machined surface is considered as the envelope of the tool trajectory, resulting in the generation of surface topography [ 16 , 20 , 21 ] . Analytical formulations are often used to derive simplified expressions for common roughness parameters, such as average areal or line roughness, by approximating the surface profile generated by the periodic engagement of the cutting edges. Third, data-driven methods rely on large datasets and are widely employed for surface roughness prediction [ 22 ] , while their application to full surface topography prediction remains relatively limited. These approaches typically learn a functional mapping between machining parameters such as cutting speed, feed rate, and depth of cut and surface roughness values using machine learning (ML) or deep learning (DL) techniques. Representative models include artificial neural networks (ANNs) [ 23 , 24 ] , support vector machines (SVMs) [ 25 , 26 ] , and other learning-based algorithms. In such cases, both input and output data are generally numerical. For image-based data, techniques such as convolutional neural networks (CNNs) [ 27 , 28 ] , recurrent neural networks (RNNs) [ 29 , 30 ] , and generative models [ 31 ] have been employed to extract features from surface topography images. These models can process two-dimensional pixel-based images or three-dimensional voxel-based representations of surface morphology, and in some cases incorporate machining parameters as additional inputs to predict surface characteristics. However, data-driven methods typically require large datasets for training, and generating sufficient experimental data is often impractical. Consequently, synthetic datasets generated from analytical or numerical approaches are frequently used to support the development of such models. Fourth, numerical approaches provide an effective framework for modeling machined surface topography when analytical formulations become insufficient to capture the complex interactions occurring during the cutting process. In general, these methods discretize the spatial and temporal domains into finite intervals, allowing the surface generation process to be reconstructed step by step. Such approaches enable the incorporation of additional factors in the machining process, including tool run-out, tool geometry, multi-tooth engagement, and complex tool kinematics, which are difficult to represent in purely analytical formulations. Several studies have attempted to classify numerical methods for machined surface topography prediction [ 16 , 32 , 33 , 34 ] . A commonly adopted classification categorizes numerical approaches according to their computational flow into forward solution methods (FSM), backward solution methods, and forward–backward solution methods, as suggested by [ 35 ] . Forward methods track the cutting-edge trajectory over discretized time steps and generate the surface by selecting the minimum height at each workpiece point. These methods offer straightforward implementation but often involve high computational cost due to the fine temporal discretization required [ 36 , 37 ] . Backward methods, in contrast, start from discretized workpiece points and solve for the corresponding tool position and cutting time. This approach reduces the need for trajectory discretization but typically requires iterative solutions of nonlinear or transcendental equations [ 20 , 38 , 39 ] . Hybrid forward–backward methods combine both strategies by representing the swept surface of the cutting-edge and computing intersections with surface points. In doing so, they partially reduce computational effort while maintaining geometric accuracy [ 40 , 41 ] . Despite these advancements, existing approaches still face challenges in balancing surface resolution, computational efficiency, and accurate correspondence between tool motion and generated surface points. Surface topography models are generally evaluated based on two key aspects: prediction accuracy and computational efficiency. Research on model accuracy is relatively mature, as many existing approaches achieve comparable levels of accuracy under similar machining conditions [ 35 ] . In contrast, computational efficiency has received comparatively less attention, particularly for large-scale milling simulations where high-resolution surfaces must be generated for cutters with multiple