In the rapidly evolving field of electric mobility, the drive system of an electric car serves as the core component dictating overall performance and energy efficiency. Traditional design approaches often focus on single objectives, such as maximizing efficiency or minimizing cost, but these methods fall short in addressing the complex trade-offs inherent in modern electric car applications. For instance, efficiency improvements frequently lead to increased manufacturing costs or thermal stability issues, highlighting the need for a holistic optimization framework. This study proposes an integrated multi-objective design and simulation framework for electric car drive systems, leveraging advanced algorithms and cross-platform co-simulation to achieve synergistic improvements in efficiency, cost, and thermal management. The innovation lies in a “dual-loop” optimization system that incorporates a dual-motor architecture, dynamic power distribution, and tolerance sensitivity modeling, validated through rigorous testing and industrial application.
The electric car industry demands drive systems that not only deliver high power density and reliability but also remain economically viable and environmentally sustainable. Conventional single-objective optimization methods, as documented in prior research, fail to capture the multi-physics couplings—electromagnetic, thermal, and mechanical—that critically influence performance. For example, ignoring mechanical deformations can lead to torque ripple prediction errors exceeding 7.8%, while electromagnetic-only models show efficiency prediction errors of up to 5.2%. Moreover, high-dimensional design spaces, encompassing up to 12 variables like permanent magnet thickness and cooling channel geometry, pose significant challenges for traditional genetic algorithms, which tend to converge to local optima. This study addresses these gaps by introducing an improved NSGA-III algorithm combined with a cross-platform simulation environment, enabling comprehensive optimization tailored for electric car drive systems. The outcomes demonstrate substantial gains: rated efficiency reaches 95.3%, costs are reduced by 12.8%, and temperature rise drops by 19.4%, with torque ripple confined within ±3.2%. These advancements not only enhance the performance of electric cars but also contribute to broader sustainability goals, such as reducing lifecycle carbon emissions by 0.92 tCO2 per unit.
The drive system architecture for an electric car is fundamentally redesigned to employ a dual permanent magnet synchronous motor (PMSM) layout on the front and rear axles. This configuration allows for specialized performance: the front-axis motor prioritizes low-speed high-torque output using a Halbach array, achieving a peak torque density of 7.2 N·m/kg, while the rear-axis motor optimizes continuous high-speed power through segmented skewing, maintaining a rated efficiency of 96.2%. The electromagnetic design follows magneto-thermal co-optimization principles, with key parameters including a stator outer diameter of 220 mm (non-uniform thickness: teeth 22 mm, yoke 18 mm), a 48-slot/8-pole combination that raises the winding factor to 0.933, and a 0.8 mm air gap length balanced for electromagnetic performance and mechanical reliability. Cooling is achieved via a hierarchical channel oil-cooling system, featuring a spiral flow channel in the stator (cross-sectional area 12 cm², flow velocity 2.3 m/s) and axial injection holes in the rotor, all enhanced with thermal grease. The power electronics incorporate a SiC MOSFET inverter with a three-level ANPC topology, supporting efficient power conversion for the electric car.
Parameterized modeling is established based on Maxwell’s electromagnetic equations and lumped parameter thermal network models, defining a 12-dimensional design variable space that captures multi-physics couplings. The optimization objectives are system efficiency \(\eta\), cost \(C\), and temperature rise \(\Delta T\), formulated as a multi-objective problem. To handle the high dimensionality and nonlinearities, an improved NSGA-III algorithm is developed, incorporating dynamic genetic operators and a Kriging surrogate model. The algorithm enhances population diversity through an entropy-based assessment function:
$$ H_t = -\sum_{i=1}^{N} p_i \log_2 p_i $$
where \(H_t\) is the entropy of generation \(t\), and \(p_i\) represents the proportion of solutions in niche \(i\). When diversity drops below a threshold \(D_{\text{max}} = 0.35\), directed mutation applies Gaussian perturbations to key variables like permanent magnet thickness and winding turns. The Kriging model approximates the design space with a covariance function:
$$ k(x, x’) = \exp\left(-\sum_{i=1}^{12} \theta_i |x_i – x’_i|^{p_i}\right) $$
where \(\theta_i\) are hyperparameters optimized via maximum likelihood estimation, reducing finite element simulation calls by 75%. This approach effectively navigates the trade-offs in electric car drive system design, such as the conflict between efficiency and torque sensitivity to permanent magnet thickness, which has a Sobol index of 0.73.
The simulation platform integrates multiple tools—Motor-CAD for electromagnetic analysis, MATLAB/Simulink for control strategy development, and Star-CCM+ for thermal-mechanical analysis—using ZeroMQ for synchronization and Protocol Buffers for data encoding. This enables real-time multi-physics coupling; for example, when winding temperature exceeds 120°C, the modulation index is dynamically reduced from 0.92 to 0.85, and switching frequency is increased to 15 kHz to suppress current harmonics. A Python script orchestrates parameter adjustments, such as permanent magnet thickness (3.5–5.2 mm) and slot opening width (1.8–2.4 mm), while the NSGA-III algorithm auto-tunes PI controller parameters. The platform supports rapid iteration, crucial for electric car development cycles.
Algorithm performance is validated on a high-performance computing setup (Xeon Gold 6348, A100 GPU, 128 GB RAM) using a case study of the HYTN-EDU220-80 drive motor. Testing under an 8-condition load spectrum per ISO 19453:2023, with introduced manufacturing tolerances, shows significant improvements. The following table summarizes key comparisons:
| Metric | NSGA-II | Improved NSGA-III | Improvement Rate (%) |
|---|---|---|---|
| Convergence Generations | 210 | 120 | 42.9 |
| Pareto Set Coverage (HV) | 0.68 | 0.86 | 26.5 |
| Temperature Rise Prediction Error (°C) | ±9 | ±4 | 55.6 |
| Single Iteration Time (min) | 38 | 5 | 86.8 |
| Tolerance Sensitivity (%) | ±12 | ±5 | 58.3 |
The improved NSGA-III reduces optimization time from 38 minutes to 5 minutes per iteration through GPU acceleration, meeting the “two-week rapid iteration” demand in electric car manufacturing. It also expands the Pareto frontier by 19.7%, yielding 65 non-dominated solutions in the 12-dimensional space versus 38 for NSGA-II, and enhances robustness to manufacturing variations, raising production yield from 92.3% to 98.6%.
Engineering validation is conducted on a 48-slot/8-pole PMSM prototype with a peak power of 110 kW, incorporating features like a 220 mm stator outer diameter, 0.8 mm air gap, N38SH permanent magnets, and flat copper windings. Testing adheres to ISO 19453:2023 and GB/T 18488—2024 standards, using a dSPACE DS1202 real-time simulator and a 250 kW dynamometer. The optimized multi-objective scheme demonstrates balanced performance, as shown in the table below:
| Parameter | Single-Objective Design | Multi-Objective Design | Change Rate (%) |
|---|---|---|---|
| Rated Efficiency (%) | 93.5 | 95.3 | +1.8 |
| Manufacturing Cost (10k CNY) | 8.6 | 7.5 | -12.8 |
| Core Loss Density (W/kg) | 2.1 | 1.7 | -19.0 |
| Peak Temperature Rise (°C) | 72 | 58 | -19.4 |
These improvements stem from a 18% reduction in copper usage and optimized cooling, lowering material costs by 12,000 CNY per unit. Robustness tests reveal that under ±10% parameter variations, torque ripple remains within ±3.2%, efficiency deviation within ±1.5%, and temperature rise error within ±4.3°C, ensuring stability for electric car operations. Durability runs over 6 hours show no demagnetization or insulation failures, with a correlation coefficient of 0.978 between simulated and measured torque-speed characteristics. The system’s rated load efficiency of 95.1% in bench tests closely matches the simulation value of 95.3%, with a temperature rise error of 2.7°C, complying with ISO 16750-3:2023 Class A accuracy.
Environmental and social benefits are substantial for electric car adoption. The optimized drive system achieves an energy efficiency ratio increase from 3.8 to 4.3, leading to lifecycle energy savings of 1500 kWh per electric car over 200,000 km, equivalent to a reduction of 0.92 tCO2 emissions. Scaling to an annual production of 2 million units, this translates to 1.84 million tCO2 mitigated yearly, accounting for 6.7% of the transportation sector’s carbon peak targets. A carbon benefit transfer function model indicates that for every 0.1-unit gain in system efficiency, overall industry carbon intensity drops by 2.7%. These advancements not only extend the driving range of electric cars to 620 km under NEDC conditions but also provide a full-chain low-carbon solution from design to manufacturing, supporting sustainable mobility.
Control strategy for the electric car drive system incorporates a dynamic torque distribution model based on fuzzy PID control, enabling feedforward-feedback compensation. The torque allocation is given by:
$$ T_{\text{dist}} = K_p (T_{\text{req}} – T_{\text{avg}}) + K_i \int w_{\text{err}} \, dt $$
where \(T_{\text{dist}}\) is the compensation torque, \(T_{\text{req}}\) is the demanded torque, \(T_{\text{avg}}\) is the average output of the dual motors, and \(w_{\text{err}}\) is the speed tracking error. By dynamically adjusting coefficients \(K_p\) and \(K_i\), transient response time is shortened to 0.12 s, and steady-state efficiency exceeds 94%. This enhances the agility and energy economy of electric cars, particularly during acceleration or regenerative braking scenarios.
Tolerance sensitivity is managed through a reverse optimization model that reduces manufacturing tolerance impacts by 68%. The model links key dimensional tolerances, such as those for permanent magnet placement and winding alignment, to performance metrics via sensitivity analysis. For an electric car drive system, this minimizes performance fluctuations across production batches, ensuring consistent quality. The sensitivity function for efficiency \(\eta\) relative to tolerance \(\delta\) is approximated as:
$$ \frac{\partial \eta}{\partial \delta} = -\alpha \delta + \beta $$
where \(\alpha\) and \(\beta\) are coefficients derived from Monte Carlo simulations. Optimizing process parameters based on this model cuts scrap rates for silicon steel by 9% and assembly time per unit by 32%, boosting cost-effectiveness.
Future directions for electric car drive systems involve extending this framework to multi-motor distributed drives and addressing extreme conditions like -40°C thermal management. Digital twin technology, integrated with SiC devices and advanced cooling methods, could further elevate power density and dynamic response. Additionally, the integration of machine learning for real-time terrain adaptation, as seen in some intelligent control systems, may offer synergies for autonomous electric cars. However, current research primarily focuses on dual-motor setups, leaving room for exploration in more complex configurations.
In summary, this study presents a comprehensive multi-objective optimization approach for electric car drive systems, combining improved algorithms, dual-motor architectures, and cross-platform simulation. The results validate significant gains in efficiency, cost, and thermal performance, with robust alignment to industry standards and environmental goals. As the electric car market grows, such methodologies will be pivotal in driving innovation toward higher performance, affordability, and sustainability, ultimately accelerating the global transition to clean transportation.