Energy–Structure–Performance Coupling in Stirred Crude Oil Desalting: RSM Optimization and Identification of an Input-Energy Window via Droplet Size Distribution Authors: Xiaolong Zhou, Liangkun Chen, Shen Wei, Muhammad Faryad Ali, Wenlong Deng, Zhang Jie Published: 2026 ABSTRACT This study introduces an innovative “energy–structure–performance” framework to quantify mixing intensity in stirred crude oil desalting, moving beyond traditional metrics such as stirring speed and duration. By analyzing droplet size distribution and estimating input energy via a Reynolds number–Power number approach, we identified a critical input-energy window centered around 50 J. Within this range, desalting efficiency exhibits a characteristic rise–fall trend that closely mirrors the evolution of the Sauter mean diameter (D32) along the energy axis, emphasizing the pivotal role of droplet structure in translating energy input into performance. Mechanistically, operating within this optimal energy window promotes droplet coalescence while minimizing shear-induced overemulsification. These results provide a physically grounded methodology for refining engineering operations, allowing precise control of mixing valves and optimization of energy input to achieve enhanced salt removal and improved phase separation efficiency. Introduction To meet feed quality standards and ensure the long-term safe and stable operation of refinery units, − water and inorganic salts, inevitably entrained during crude oil production and transportation and typically forming stable water-in-oil (W/O) emulsions, , must be effectively removed before deep refining. Inadequate desalting and dehydration can lead to fouling, blockages in heat-exchange equipment, and severe corrosion of the pipelines and columns. As demonstrated in recent multiphase petroleum studies, the continuous growth of inorganic salt precipitation and corrosive destruction poses a serious threat to production equipment, highlighting the necessity of efficient salt and water removal before deep refining. , Consequently, achieving efficient and reliable demulsification and desalting of these emulsions remains a major technical challenge in refinery pretreatment. − The stability of emulsions primarily originates from natural surface-active components such as asphaltenes, resins, and gums, which form a robust interfacial film at the oil–water interface. The stability of these asphaltene-stabilized emulsions is further complicated by the amount of mineralized formation water, where the emulsified water actively participates in the formation of stable organic–inorganic structural barriers. This film maintains water droplets in a highly dispersed state within the continuous oil phase, inhibiting their coalescence and settling. − To overcome this challenge, various enhanced demulsification and desalting strategies have been developed in both industry and academia, including chemical demulsifier addition, heating combined with wash water dilution, electrostatic desalting under an electric field, and physical or physicochemical intensification techniques such as ultrasound, centrifugation, and nanomaterials. − Recent advancements in chemical demulsification have shown that multicomponent demulsifier packages can synergistically minimize interfacial tension, break the interfacial film, and significantly increase the frequency of droplet collisions. Existing studies and reviews demonstrate that these approaches can substantially improve desalting and dehydration performances across different crude oil systems; however, their effectiveness is highly dependent on the appropriate alignment of crude properties with operating conditions. ,, In desalting and dehydration research, multifactor optimization and process modeling have become widely adopted strategies. Because emulsions are compositionally complex and operating variables are strongly interdependent, single-factor experiments often fail to capture both the main effects and interactions simultaneously. As a result, response surface methodology (RSM) and its commonly used experimental designs, such as the central composite design (CCD) and Box–Behnken design, have been extensively applied for modeling desalting processes and optimizing operating parameters. , Previous studies have shown that RSM can generate predictive empirical models with relatively few experiments, allowing evaluation of the effects of temperature, demulsifier dosage, wash water volume, electric field strength, and settling time on both desalting efficiency and energy consumption. ,, For example, Mehdizadeh et al. explored the combined effects of temperature, demulsifier injection rate, wash water dosage, and voltage on electrostatic desalting performance, proposing operating conditions that balance efficiency and energy demand. White et al. demonstrated improvements in AC electrostatic desalting, achieving stable operation while reducing both demulsifier and wash water requirements. For more complex emulsions, chemical demulsification has been coupled with physical intensification techniques, such as ultrasound , with RSM used to identify key factors and optimal operating ranges. Similarly, nanomaterial-assisted demulsification combined with an RSM-based optimization has been reported. Comparisons of experimental designs have further highlighted method-specific advantages: Dhandhi et al. reported that CCD can offer superior predictive performance in certain systems, while Ahmadi et al. applied RSM under scaled-up bottle tests and dynamic conditions, achieving high desalting and dehydration efficiency while reducing the need for additives. Overall, response-surface-based approaches provide robust tools for multivariable analysis and operating-window prediction in crude oil desalting processes. In many existing optimization studies, stirring is often treated merely as a background operating condition, described only by stirring speed or duration, with limited systematic quantification of stirring power and energy input. Furthermore, structural variables such as droplet size distribution (DSD) are rarely incorporated into a unified analytical framework, making it difficult to elucidate the process chain by which mixing intensity influences droplet size and distribution, wash water contact, and ultimately desalting efficiency. Critically, mixing intensity exhibits a window effect: insufficient intensity results in inadequate oil–water contact and limited salt transfer, whereas excessive intensity can cause overemulsification, slow phase separation, and even reduction of desalting performance. These limitations hinder comparisons of optimization results across different equipment scales, oil-phase viscosities, and crude properties, and they also restrict the transferability of findings to scale-up applications and energy evaluation. Based on these considerations, a laboratory-scale stirred desalting system was employed to systematically investigate how mixing intensity affects the crude oil desalting performance. Key factors, including stirring speed, stirring time, wash water dosage, desalting temperature, and settling time, were examined, while stirring power and input energy were quantitatively determined using a Reynolds number–Power number approach, thereby transforming the conventional speed-based description into an energy-based metric that allows direct engineering comparison. The system utilized a three-blade straight paddle impeller with a diameter of 62 mm and a cylindrical vessel with an inner diameter of 72 mm to treat an emulsion prepared from 85 g of crude oil and 15 g of water. The oil-phase viscosity was approximately 0.503 Pa·s, and the initial salt content was about 110 ppm. To characterize the mixing state and droplet-structure evolution, DSD parameters, including D32 (Sauter mean diameter, SMD), D50, and D90, were introduced, aiming to establish correlations among mixing intensity, droplet scale, and desalting efficiency. ,,− In addition, a CCD-based RSM was applied for process modeling and optimization, enabling evaluation of the trade-off between input energy and desalting performance. This study identifies and discusses an operable input-energy range that avoids overemulsification while balancing desalting efficiency and energy consumption, providing practical guidance for parameter selection in stirred desalting operations. Materials and Methods Crude Oil Sample The crude oil used in this work was obtained from Sinopec Shanghai Gaoqiao Petrochemical and is denoted as the Labi crude oil. Its physicochemical properties are summarized in Table . 1Physicochemical Properties of Crude Oilphysical propertiesunitsvaluestest standardsdensitykg/m3 842.3SH/T 0316-1998chlorine contentmg/kg110SH/T 6056-2022saturateswt %59.1SY/T 5119-2016aromaticswt %21.3SY/T 5119-2016resinswt %15.1SY/T 5119-2016asphalteneswt %5.4SY/T 5119-2016water contentmg/L200GB/T 260-2016 Physicochemical Properties of Crude Oil Emulsion Preparation and Desalting Procedure For each experiment, 85 g of crude oil and 15 g of water were mechanically stirred to prepare the emulsion for desalting. Following mixing, 20 ppm of demulsifier was added, and the mixture was subsequently transferred to a separatory bottle to allow demulsification. The desalting efficiency was then calculated using eq : Equation Desalination Efficiency 1 SRE=(S0−S1)*100%/S1 The demulsifiers used in this study were a blend of DRC-168 and DRI-0903 at a specified mass ratio. The properties of these two commercial demulsifiers are defined in Table . 2Physicochemical Properties of Demulsifiersname codecolorodorflash pointtest standardmain active ingredientsDRC-168Dark amberAromatic63 °CASTM D93Alkoxy-Nonylphenol Formaldehyde Resin, PolyacrylateDRI-0903Light yellow brownSlightly pungent30 °CASTM D93Polyacrylate Physicochemical Properties of Demulsifiers Stirring Apparatus and Experimental System A laboratory-scale mechanical stirring system was employed for the experiments. The vessel was a cylindrical container with an inner diameter of 72 mm, and a three-blade straight paddle impeller with a diameter of 62 mm was installed coaxially. The stirring speed was precisely regulated by using a variable-frequency speed controller. A schematic of the stirred desalting setup is presented in Figure . Stirred desalting tank. Calculation of Stirring Reynolds Number (Re) and Stirring Energy To quantify the mixing intensity, the stirring Reynolds number, stirring power, and input energy were employed as key evaluation parameters. Stirring Reynolds Number The stirring Re was calculated as follows. Reynolds Number Equation Reynolds Number 2 Re=ρD2Nμ where ρ is the density of the mixed phase (kg·m–3), N is the stirring speed (s–1), D is the impeller diameter (m), and μ is the viscosity of the mixed phase (Pa·s). Stirring Power and Input Energy Stirring power (P) was calculated by using the classical power correlation: Equation Stirring Power 3 P=NPρN3D5 where N p is the Power number, obtained from a power chart corresponding to the calculated Re. The calculated Reynolds numbers for the experimental conditions ranged from 1200 to 7,500, indicating that the fluid dynamics transitioned from the transitional to the fully turbulent flow regime. Within this specific flow regime and for the chosen three-blade straight paddle impeller, the Power number (N p) correlation curve is relatively stable, with N p values varying only slightly between 0.38 and 0.22. Therefore, deriving the stirring power from standard Re-N p correlations is highly justified for this system. It is important to acknowledge the limitations of assuming Newtonian behavior in these calculations. Crude oil emulsions, inherently stabilized by asphaltenes and resins, are non-Newtonian fluids that typically exhibit shear-thinning behavior. By utilizing a constant apparent viscosity (≈0.503 Pa·s) to calculate the Reynolds number across all stirring speeds, the resulting energy values represent a “nominal input energy” rather than the absolute physical dissipation energy. Specifically, at high shear rates (e.g., 650 rpm), the actual emulsion viscosity would be lower than the assumed constant, yielding a slightly higher actual Re; conversely, at low shear rates, the actual viscosity might be higher. However, for the macroscopic purpose of establishing the energy–structure–performance (E–S–P) framework, this engineering approximation is robust. The constant-viscosity assumption provides a unified, comparative metric to evaluate the critical physical transitionfrom coalescence-dominated to breakup-dominated droplet behavioracross a continuous energy spectrum without altering the fundamental “rise-and-fall” trend of the system. The input energy (E) was defined as the product of the stirring power and stirring time: Equation Stirring Energy 4 E=Pt Using the above methodology, the mixing intensity of the oil/water system was systematically evaluated. Droplet Size Analysis To quantify the mixing state, the DSD of water droplets in the emulsion was measured and analyzed. The procedure was as follows: immediately after the mixture was stirred, a sample was collected. To avoid disturbing the emulsion, no coverslip was applied. The sample was promptly observed under an optical microscope equipped with 20× and 40× objective lenses, and the visual fields were captured using a digital camera. The images were processed in ImageJ for droplet size analysis. Prior to image thresholding and particle counting, the spatial resolution was calibrated by using a standard stage micrometer. Accounting for both the optical resolution at 40× magnification and the digital pixel limitations for noise elimination, the reliable minimum detectable droplet size in this study was approximately 1 μm. Since the characteristic metrics (D50, D90, and D32) are volume-weighted, the negligible volumetric mass of submicron droplets ensures that this optical limit does not compromise the statistical accuracy of the macroemulsion size distributions. The resulting extracted size data were further analyzed with Origin software to calculate D50, D90, and D32. Furthermore, to quantitatively evaluate the correlation between the macroscopic emulsion structural parameters and desalting performance, a statistical analysis was conducted. The Pearson correlation coefficient (r) was calculated to determine the linear dependence between the Sauter mean diameter (D32) and desalting efficiency across the evaluated energy spectrum. Experimental Design and Response Surface Analysis CCD was used for the multifactor experimental design. Stirring speed, stirring time, wash water dosage, desalting temperature, and settling time were selected as independent variables, and desalting efficiency was used as the response. The CCD matrix is shown in Table , where α = 1. 3Experimental Design Parameterparametercodeunitminmax-alpha+alphastirring speedArpm150650150650stirring timeBmin3012030120wash water dosageCg10301030temperatureD°C30903090settling timeEmin01200120 Experimental Design Parameter The experimental results were analyzed using RSM for regression modeling and statistical evaluation, enabling assessment of the main effects and interactions of the variables on desalting performance and identification of the optimal combination of operating parameters. Simultaneously, the input energy corresponding to the mixing-related operating conditions in the CCD experiments was calculated to allow comparison with additional energy-based experiments. The resulting fitted model is expressed as Equation Fitting Polynomial Equations 5 y=β0+Σβixi+Σβiixi2+ΣΣβijxixj+ε where y represents the response function, which is the optimization objective; β0 denotes the constant term, while βi, βii, and βij correspond to the linear, quadratic, and interaction effects, respectively. x i values indicate the input parameter values, and ε represents the statistical error. The detailed experimental parameter combinations and corresponding desalting efficiency results for all CCD runs are presented in Table . 4Droplet Size Metrics and Desalting Efficiency under Different Input Energiesrunstirring energyD50D90D32desalting efficiencyunitJmmmmmm%1100.341.26121.814675.62200.68342.07982.74183.1043300.65672.43113.035386.082864400.61162.56673.098588.431435500.59342.36022.881888.757146600.57592.06762.843487.762867700.54611.74852.523986.891438800.51431.39642.149485.814299900.48171.4161.949884.52286101000.47241.38961.638783.92111100.44461.34021.522283.22857121200.37761.34061.655282.22857131400.37731.28041.66681.23143141600.35551.17251.550779.86857151800.32891.06661.518277.84571161900.29310.92731.346376.15429172200.26910.86271.244674.41143182400.27750.80330.1.1572.21714193200.30090.79010.984670.46286203600.33790.79030.681968.576214500.39390.80980.656568.7 Droplet Size Metrics and Desalting Efficiency under Different Input Energies 5CCD Experiment Resultsruninput parametersresponse ABCDESRE stirring speedstirring timewater dosagetemperaturesettling time rpmming°Cmin%1150120109012077.8762215030103012073.9693650301030064.928441507520606086.463454007520606082.59596650120309012050.477400752060087.477986501203030062.873194007510606087.0525104003020606084.0251116501201090069.579712650120103012069.5881134007530606084.3527144007520606078.253515150303030071.1098164007520606074.2615174007520606075.505218150301090081.4832194007520906081.38442040075206012071.03822165030109012074.3306224007520306065.0924236507520606082.4298241501201030065.329254007520606073.37452640012020606085.074227650303090045.11672865030303012053.33291501203090083.33304007520606080.003115030309012093.3332150120303012093.33 CCD Experiment Results Supplemental Mechanistic Experiment: Input Energy and DSD In the RSM analysis, the stirring speed (A) and stirring time (B) exhibited significant nonlinear effects and interaction in the desalting model, highlighting the critical role of mixing intensity in desalting behavior. However, using speed or time alone as operational variables makes it difficult to represent the mixing state on a unified basis across different conditions and limits mechanistic interpretation of why mixing-related factors affect desalting differently under varying scenarios. To further elucidate the nonlinearity and interaction associated with mixing in the CCD experiments and to investigate the role of mixing intensity from an energy–structure perspective, a supplemental experiment was conducted, focusing on input energy and DSD. It should be noted that this continuous input-energy experiment was intended for trend analysis and mechanistic discussion; no replicate measurements were performed at individual energy points. The results were used to determine the overall dependence of desalting behavior on the input energy rather than to identify a statistically optimal condition. This experiment was designed independently with test conditions that do not correspond to the CCD center point or other CCD runs, and input energy was systematically varied as the sole independent variable. The operating combinations of the stirring speed and stirring time are summarized in Table . Except for these two variables, all other conditions were fixed at the levels suggested by the RSM optimization analysis to enable comparison of trends on a unified energy scale. The input-energy range was set from 10 to 360 J, with increments of 10–20 J to balance experimental coverage and workload. These results were used to determine the overall dependence of desalting behavior on input energy rather than to identify statistically optimal conditions. Although no local replicates were performed at individual energy points, the high sampling density across the wide input-energy range (21 data points from 10 to 450 J) ensures that a continuous and reliable trend was captured. This continuous-trend analysis approach was prioritized over single-point statistical optimization to reveal the mechanistic transition between droplet coalescence and breakup. Results and Discussion Statistical Analysis and Model Validation for Desalting Efficiency The experimental results from the CCD were analyzed using the RSM to establish a quadratic regression model for desalting efficiency. Analysis of variance (ANOVA) was employed to assess the statistical significance and adequacy of the fitted model. A detailed examination of the ANOVA results in Table provides critical insights into the relative importance of the operational variables. The model’s highly significant p-value (<0.0001) confirms its overall predictive validity. Among the linear terms, stirring speed (A) emerges as the most dominant factor affecting desalting efficiency (p < 0.0001), followed by wash water dosage (C, p = 0.0255). Furthermore, the highly significant quadratic term for temperature (D2, p < 0.0001) mathematically validates the nonlinear, parabolic response of efficiency to heating. Crucially, several interaction terms, specifically stirring speed with water dosage (AC), stirring speed with temperature (AD), stirring time with settling time (BE), and water dosage with temperature (CD), demonstrate significant interaction effects (p < 0.01). This confirms that desalting performance is driven by complex multivariable coupling rather than isolated factors, thereby justifying the necessity of the subsequent 3D response surface analysis. 6ANOVA Analysis TableSourceSum of Squares d f Mean Square F-value p-value Model0.000111010.14<0.0001significantA Stirring speed01026.49<0.0001 B Stirring time1.31 exp-0711.31 exp-070.10320.7513 C Wash water dosage7.39 exp-0617.39 exp-065.830.0255 D Temperature2.54 exp-0812.54 exp-080.020.8889 E Settling time2.51 exp-0612.51 exp-061.980.1744 AB2.73 exp-0712.73 exp-070.21540.6476 AC01013.760.0014 AD01014.610.0011 BE0109.230.0065 CD01011.20.0032 D2 01028.13<0.0001 Residual0201.27 exp-06 Lack of Fit0151.57 exp-064.420.0546not significantCor Total0.000231 R 2 0.848 R 2 pred 0.6215 R 2 adj 0.7644 Adequate Precision 14.3208 ANOVA Analysis Table The lack-of-fit test was not significant at the selected confidence level, indicating that deviations between experimental data and model predictions are primarily due to random error rather than systematic bias. Additionally, Figure shows that the residuals are randomly distributed and approximately follow a normal distribution, confirming that the assumptions underlying the regression analysis are reasonably satisfied. The parity plot further demonstrates good agreement between the predicted and experimental desalting efficiencies, verifying the predictive capability of the established model. (A) Normal plot of residuals and (B) residuals vs run. Overall, the statistical analysis confirms that the RSM model provides a reliable foundation for evaluating the effects of the operating variables on the desalting performance. The Design-Expert software was used to analyze and fit the experimental data, generating the regression equations for the desalting models. Equation Desalting Fitting Equation 6 1SRE=2.9248×10−2−1.1×10−5A−1.2×10−5B−2.91×10−4C−4.03×10−4D−3.0×10−4E−1.161×10−8AB+4.175×10−7AC+1.4343×10−7AD−3.166×10−6BE+3.1396×10−6CD+2.3638×10−6D2 Effects of Operating Variables on Desalting Efficiency The fitted response surface model was employed to analyze both the individual and the interactive effects of the operating variables on desalting efficiency. Figure illustrates the single-factor effects of the stirring speed, stirring time, wash water dosage, desalting temperature, and settling time on the desalting efficiency. Effect of a single factor on desalting efficiency: (A) stirring speed, (B) stirring time, (C) wash water dosage, (D) temperature, and (E) settling time. As shown in Figure A, the desalting efficiency decreases with increasing stirring speed. Higher stirring speeds enhance local turbulence and shear intensity, promoting droplet breakup and producing finer dispersed water droplets. While smaller droplet sizes increase the interfacial area and facilitate salt transfer to the aqueous phase, excessive breakup generates highly stable emulsions composed of small droplets that are difficult to coalesce and settle during the subsequent settling stage, thereby limiting effective salt removal. Stirring time also affects the desalting performance by modulating the overall mixing state. As illustrated in Figure B, desalting efficiency increases slightly with longer stirring times, reflecting the competing effects of enhanced oil–water contact and increased emulsion stability. Since stirring speed and stirring time jointly determine overall mixing intensity, the effect of stirring time cannot be interpreted independently of stirring speed. Wash water dosage exhibits a positive effect on the desalting efficiency (Figure C). Increasing the wash water fraction enlarges the oil–water contact area and promotes mass transfer of salts from the oil phase to the aqueous phase. Additionally, a higher water fraction influences droplet collision and coalescence behavior, affecting brine removal efficiency during settling. The influence of the temperature on desalting efficiency follows a monotonic trend (Figure D). Higher temperatures reduce the viscosity of the continuous phase and weaken the interfacial film, facilitating droplet coalescence and settling. However, beyond a certain temperature, desalting efficiency decreases, indicating that temperature simultaneously affects emulsification stability and separation dynamics. As shown in Figure E, extending the settling time significantly improves the desalting efficiency by allowing sufficient time for droplet coalescence and gravity-driven separation. The effectiveness of prolonged settling, however, strongly depends on the DSD; when droplets are excessively small and emulsions are highly stable, the benefit of extended settling becomes limited. Taken together, the single-factor analysis indicates that the operating variables are interdependent and that mixing-related parameters play a central role in determining desalting behavior. Interaction Effects of Operating Variables Figure presents the response surfaces of the significant quadratic interaction terms influencing desalting efficiency, including stirring speed and stirring time (Figure A), stirring speed and wash water dosage (Figure B), stirring time and settling time (Figure C), stirring speed and desalting temperature (Figure D), and wash water dosage and desalting temperature (Figure E). Effect of coupling terms on the desalting rate: (A) stirring speed and stirring time, (B) stirring speed and wash water dosage, (C) stirring time and settling time, (D) stirring speed and temperature, and (E) wash water dosage and temperature. As shown in Figure A, higher desalting efficiencies are achieved at relatively low stirring speeds and short stirring times. Increasing either parameter results in a gradual decrease in the desalting efficiency, indicating that excessive mixing stabilizes emulsions and suppresses droplet coalescence during settling. This interaction highlights the strongly nonlinear combined effect of the stirring speed and stirring time on desalting efficiency. Figure B shows that at low stirring speeds, increasing the water dosage significantly enhances the desalting efficiency. In contrast, at higher stirring speeds, additional wash water provides only marginal improvement. This suggests that under moderate mixing conditions, increased water effectively enhances oil–water contact and salt transfer, whereas at high mixing intensity, emulsion stabilization limits the benefits of extra wash water. As illustrated in Figure C, longer settling times improve desalting efficiency, particularly when the stirring time is relatively short. However, when stirring time is prolonged, the gain from extended settling becomes less pronounced, indicating that excessive emulsification during mixing requires more favorable separation conditions to achieve comparable desalting performance. Figure D,E reveals an optimal temperature range of approximately 60–70 °C, within which higher desalting efficiencies are attained across a broad range of stirring speeds and wash water dosages. Beyond this range, the desalting efficiency declines, reflecting the dual role of temperature in affecting both emulsion stability and separation kinetics. Overall, the interaction analysis demonstrates that mixing-related variables strongly interact with other operating parameters, and their effects on the desalting efficiency cannot be fully understood by examining single variables in isolation. Effect of Input Energy and DSD on Desalting Efficiency To further elucidate the role of mixing intensity suggested by the RSM analysis, a supplemental experiment was conducted, in which input energy was treated as the sole independent variable. Input energy was systematically varied by adjusting combinations of stirring speed and stirring time, while all other operating conditions were fixed at representative levels determined from the CCD analysis. This approach allowed the mixing state to be examined on a unified energy basis independent of specific operational parameters. Figure shows the variation of the characteristic droplet sizes (D50, D90, and D32) with increasing input energy. All three metrics exhibit a similar nonmonotonic trend: droplet size initially increases with input energy, reaches a maximum near approximately 50 J, and then gradually decreases as energy is further increased, eventually stabilizing at values slightly lower than those observed at the lowest energy level. Effect of stirring energy on particle size. This trend reflects the competition between droplet coalescence and breakup during mixing. At a relatively low input energy, the turbulent kinetic energy is insufficient to overcome the resistance of the interfacial film surrounding water droplets. Under these conditions, increased mixing primarily enhances droplet collisions, with coalescence dominating over breakup, leading to larger characteristic droplet sizes. As the input energy approaches the intermediate range (∼50 J), the coalescence efficiency reaches its maximum, resulting in the largest observed droplets. When the input energy exceeds this range, the energy imparted by stirring becomes sufficient to disrupt and fragment droplets. Consequently, droplet breakup gradually dominates and characteristic droplet sizes decrease with further increases in input energy. Among the droplet size descriptors, D32 (SMD), defined as the ratio of the third to the second moment of the diameter distribution, shows the most pronounced response to input energy, reflecting its sensitivity to the interfacial area and phase-contact characteristics. The variation of the desalting efficiency with input energy is presented in Figure . Notably, the macroscopic desalting efficiency exhibits a trend that closely mirrors the microstructural evolution of D32. Statistical analysis confirms this robust relationship, yielding a highly positive Pearson correlation coefficient of r = 0.883 between D32 and the desalting efficiency. This strong quantitative alignment indicates that macroscopic droplet structure fundamentally mediates the relationship between mixing energy input and desalting performance. This close correspondence indicates that the droplet structure mediates the relationship between mixing energy input and desalting performance. Larger droplet sizes, reflected in higher D32 values, promote the release and settling of salt-laden water droplets during the subsequent quiescent stage, thereby enhancing the desalting efficiency. Furthermore, the strong agreement between the independent trends of D32, D50, D90, and desalting efficiency provides robust internal validation for the identified energy window, effectively compensating for the absence of single-point replicates. Effect of stirring energy on desalting efficiency. Overall, these results demonstrate that the influence of mixing intensity on desalting efficiency cannot be understood solely in terms of operational variables such as the stirring speed or time. Rather, input energy governs the evolution of the droplet-scale structure, which in turn determines the balance between the effective wash water contact during mixing and droplet separability during settling. Although single-point replicate measurements were not performed, the statistical reliability of the identified optimal energy window is robustly supported by two factors. First, the high sampling density across a broad energy range (21 data points from 10 to 450 J) yields a smooth and continuous physical trend, which makes it highly unlikely that the peak is a random artifact of experimental variability. Second, the synchronous peaking of completely independent measurementsdesalting efficiency (a macroscopic performance metric) and D32 (a microscopic structural metric)provides strong internal cross-validation for the existence of this critical transition point. Identification of an Input-Energy Window and Mechanistic Discussion By combining the input-energy-dependent droplet size data with the corresponding desalting efficiency results, a characteristic energy window governing desalting behavior can be identified. Within the investigated range, an intermediate input-energy region centered around approximately 50 J corresponds to both the maximum characteristic droplet size and the highest observed desalting efficiency. This region represents an optimal balance between sufficient oil–water contact during mixing and adequate droplet coalescence and separability during settling. From a mechanistic perspective, these findings highlight the existence of a mixing-intensity window rather than a monotonic relationship between the energy input and desalting performance. At low input energy, insufficient mixing limits the oil–water contact and salt transfer. As the input energy increases into the intermediate range, enhanced contact promotes salt migration, while coalescence remains dominant, resulting in larger droplets that readily settle. Beyond this range, excessive energy input intensifies droplet breakup and stabilizes the emulsion, producing finer droplets that resist gravitational separation and thereby constrain the desalting efficiency. From a fundamental perspective, the ability of water droplets to coalesce or break is governed by microscopic interfacial rheologyspecifically, the mechanical strength and elasticity of the asphaltene and resin networks at the oil–water interface. While interfacial rheology determines the intrinsic film drainage dynamics, dynamically measuring it in situ under intense turbulent stirring is technically prohibitive. Therefore, within the E–S–P framework, D32 serves as a highly reliable and measurable macroscopic proxy for these microscopic rheological outcomes. The observed D32 peak around 50 J visually signifies the energetic tipping point where the applied turbulent shear overcomes the rheological strength of the interfacial film. It should be emphasized that the identified input-energy value (≈50 J) is specific to the Labi crude oil properties, mixing geometry, and operating conditions investigated in this study. Consequently, the absolute energy values should be regarded as an engineering reference rather than a universal constant. Nonetheless, the primary contribution of this work lies in the E–S–P framework, which provides a physically meaningful basis for interpreting the effects of mixing intensity on the desalting performance. It should be emphasized that the identified input-energy value (≈50 J) is a system-specific characteristic value. This absolute numerical value will inevitably shift in actual industrial production as crude oil physicochemical properties (e.g., viscosity and asphaltene content) and operational conditions (e.g., continuous flow dynamics and desalter scale) change. The primary contribution of this work is not to establish 50 J as a universal industrial constant but to propose the energy–structure–performance (E–S–P) methodology. This framework demonstrates that an optimal energy window fundamentally exists and can be systematically located by tracking droplet structure evolution, transitioning desalting optimization from empirical trial-and-error to a quantified, energy-based protocol. To extend the E–S–P framework to diverse crude oil systems or industrial scales, the following methodologies are proposed: Methodological generalizability: it should be emphasized that the identified optimal input-energy value (≈50 J) is a system-specific characteristic. This absolute numerical value will inevitably shift as crude oil physicochemical properties (e.g., viscosity and asphaltene content) change. Therefore, the primary contribution of this work is not to establish 50 J as a universal industrial constant but to propose the energy–structure–performance (E–S–P) methodology. For different crude oils, the specific energy window can be recalibrated by correlating the Reynolds number (Re) and Power number (Np) with laboratory-scale droplet size measurements. This approach transitions desalting optimization from empirical trial-and-error to a quantified, energy-based protocol. Industrial scale-up: for industrial applications, it is critical to recognize the geometric and operational differences between laboratory stirred tanks and industrial continuous desalting. Industrially, mixing energy is typically imparted not by an impeller but by the hydrodynamic resistance across a mixing valve located upstream of the desalter vessel. While the residence time and spatial distribution of energy dissipation differ drastically between batch and continuous systems, the fundamental microscopic mechanism governing the droplet behavior remains identical. The universal competition between turbulent collision-induced coalescence and shear-induced breakup dictates that the characteristic “rise-and-fall” trend of the E–S–P framework is fundamentally valid for continuous flow as well. In this framework, the mechanical “input energy” delivered by the lab-scale impeller mathematically corresponds to the pressure drop (ΔP) across an industrial mixing valverepresenting the mechanical energy loss term in Bernoulli’s fluid energy equation. By conducting lab-scale batch energy sweeps to identify the optimal energy window for a specific crude blend, operators can calculate the equivalent pipeline pressure drop required. The optimal desalting performance can then be continuously maintained online by dynamically adjusting the mixing valve opening to match this target ΔP, effectively preventing the emulsion from entering the adverse “overshearing” regime. To further translate the empirical 50 J energy window into a fundamental fluid dynamic context, it is instructive to correlate this optimal point with critical dimensionless numbers. Due to the inherent challenges of measuring in situ dynamic interfacial tension under high-shear conditions, a representative interfacial tension for demulsifier-treated crude oil–water systems (approximately 12.4 mN/m, as reported in the recent literature) was utilized. At the optimal input energy of 50 J, the system operates at a Reynolds number of Re = 2544.6, indicating a transitional-to-early turbulent flow regime that provides sufficient mixing without extreme chaotic shear. Crucially, the system reaches a Weber number of We = 184.6 and a Capillary number of Ca = 0.834. In turbulent stirred systems, a Ca approaching unity signifies a critical threshold of droplet deformation, while a We of 184.6 indicates that disruptive inertial forces have become dominant. Collectively, these dimensionless numbers quantify the hydrodynamic boundary of the E–S–P framework: they mark the precise sweet spot where applied shear and inertial forces are perfectly tuned to deform droplets and rupture the asphaltene-stabilized interfacial film (maximizing coalescence), yet remain just below the limit that would trigger massive secondary droplet breakup and detrimental overemulsification. The physical boundaries of the identified input-energy window are fundamentally rooted in the interfacial properties of the crude oil, specifically, the role of asphaltenes. Asphaltenes are known to form robust, viscoelastic “structural barriers” at the oil–water interface, which significantly enhance emulsion stability. In the low-energy regime (<50 J), the stirring intensity is insufficient to rupture these asphaltene-stabilized films, meaning that even if droplets collide, they fail to coalesce, resulting in poor desalting performance. Conversely, in the high-energy regime (>50 J), the excessive turbulent shear forces dominate, rapidly generating a large amount of a new interfacial area. The high surface activity of asphaltenes allows them to quickly adsorb and stabilize these newly formed fine droplets, leading to the detrimental “overemulsification” observed in Figure . Thus, asphaltenes act as the “rheological gatekeeper” of the system; their concentration and film-forming kinetics dictate the minimum energy required for film rupture and the maximum energy allowed before secondary stabilization occurs. This mechanistic understanding explains why the E–S–P coupling is a universal framework, while its specific numerical window remains sensitive to crude oil composition. In conclusion, by integrating statistical modeling with energy-based and structural analyses, this study provides a coherent explanation for the nonlinear effects of mixing and underscores the importance of controlling the mixing intensity within a rational energy window. Conclusions This study establishes an E–S–P framework to investigate the role of mixing intensity in crude oil desalting. RSM analysis reveals significant nonlinearity and interactions among mixing variables. By introducing input energy as a unified metric, a critical energy window (∼50 J) was identified within which the optimal droplet structure maximizes desalting efficiency. While providing a physically meaningful approach for parameter optimization, the study is currently limited to a specific crude oil system and theoretical power correlations without industrial validation. Future research should focus on multisystem validation, direct power measurement, and industrial pilot tests to refine the E–S–P framework for diverse refinery conditions. Overall, this methodology offers a quantitative protocol to prevent overemulsification and optimize energy input in stirred desalting processes.