A new energy vehicle motor feeding characteristic test system, method and bench

Abstract

The application relates to the technical field of automobile testing, in particular to a motor feeding characteristic test system, method and bench for a new energy automobile, which comprises a physical environment simulation unit, which is used for providing a composite environment field containing temperature, humidity and mechanical vibration for a test motor; an integrated load simulation unit, which comprises a mechanical inertia module, a road surface characteristic simulation module and a braking load simulation module, and is used for applying a dynamic load coupled with vehicle inertia, road surface resistance characteristics and braking strength to the test motor in the composite environment field. The physical environment simulation unit realizes the cooperative simulation of a temperature and humidity range of -40 DEG C to 85 DEG C and six-degree-of-freedom mechanical vibration, and the integrated load simulation unit can accurately reproduce the dynamic load characteristics under complex working conditions such as ice and snow road surfaces and gravel roads through the coupling control of mechanical inertia, road surface characteristics and braking load.

Description

Technical Field

[0001] This invention relates to the field of automotive testing technology, specifically to a testing system, method, and test bench for the power supply characteristics of motors used in new energy vehicles. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the regenerative braking power supply performance of the motor, as a core power component, directly affects the vehicle's driving range and energy efficiency, becoming a key focus of industry research and testing. Currently, new energy vehicles face complex and varied driving scenarios, requiring them to cope with different environmental conditions such as temperature, humidity, and vibration, as well as dynamic load conditions such as differences in road surface adhesion coefficients, fluctuations in braking intensity, and vehicle inertial coupling. This places stringent requirements on comprehensive and accurate testing of the motor's power supply characteristics.

[0003] However, existing motor power supply characteristic testing technologies have many limitations and cannot meet the needs of practical applications. In terms of environmental simulation, traditional test systems mostly adopt single-factor simulation methods, which can only achieve static control of a single temperature or humidity. They lack the construction of a composite environmental field that couples temperature, humidity and mechanical vibration, and cannot reproduce the real operating state of vehicles in complex environments such as extreme cold, high temperature and high humidity, and bumpy roads. This results in a disconnect between the test environment and the actual vehicle operating conditions, and the test results have limited reference value.

[0004] At the load simulation level, existing equipment typically controls mechanical inertia, road resistance, and braking load separately and independently, failing to achieve synergistic coupling among the three. This makes it difficult to reproduce the dynamic load characteristics under complex conditions such as emergency braking and driving on gravel roads. In particular, road characteristic simulation relies heavily on mechanical structure adjustments, resulting in slow response speeds and cumbersome switching, making it impossible to accurately simulate the impact of wheel slip rate fluctuations caused by changes in different road surface adhesion coefficients on power supply stability. Regarding data acquisition and analysis, traditional systems have a single acquisition dimension, mainly focusing on electrical or mechanical parameters, lacking simultaneous acquisition of thermal parameters and multi-axial vibration parameters. Furthermore, the lack of precise timestamp alignment across data dimensions leads to poor multi-source data fusion. At the same time, existing technologies struggle to achieve refined decomposition of losses, failing to accurately locate key loss sources such as copper losses, iron losses, and permanent magnet eddy current losses. Moreover, they struggle to identify additional electromagnetic losses caused by mechanical vibrations, leading to blind spots in control strategy optimization. Summary of the Invention

[0005] To address the problems in the existing technology, this invention provides a test system, method, and test bench for the power supply characteristics of motors used in new energy vehicles. The technical solution adopted by the present invention to solve its technical problem is: a test system for the power supply characteristics of a motor for new energy vehicles, including a physical environment simulation unit, used to provide a composite environmental field including temperature, humidity and mechanical vibration for the test motor; The integrated load simulation unit includes a mechanical inertia module, a road surface characteristic simulation module, and a braking load simulation module, which are used to apply a dynamic load coupled with vehicle inertia, road surface resistance characteristics, and braking intensity to the tested motor in the composite environment field. A multi-dimensional data sensing and fusion unit is used to synchronously and in real time collect the electrical, mechanical, thermal and vibration parameters of the tested motor under the composite environmental field and dynamic load. The intelligent power supply performance diagnosis and closed-loop optimization unit is communicatively connected to the multi-dimensional data sensing and fusion unit and the controller of the tested motor, and is used for: Based on the fused multi-dimensional data, the energy flow distribution during the power feeding process is analyzed in real time and key loss sources are located. Based on the characteristics of the key loss sources, optimized parameters for the regenerative braking control strategy of the tested motor are dynamically generated. The optimized parameters are sent to the controller of the motor under test, and the integrated load simulation unit is driven to perform a new round of testing to verify the optimization effect.

[0006] The beneficial effects of this invention are: (1) This invention constructs a test system that couples a composite environmental field with a dynamic load. The physical environment simulation unit achieves coordinated simulation of temperature and humidity range of -40℃ to 85℃ and six-degree-of-freedom mechanical vibration. The integrated load simulation unit can accurately reproduce the dynamic load characteristics under complex working conditions such as icy and snowy roads and gravel roads through the coupled control of mechanical inertia, road surface characteristics and braking load, thus solving the core problem of the disconnect between single-factor simulation and actual vehicle working conditions. At the same time, the multi-dimensional data perception and fusion unit covers four major categories of parameters: electrical, mechanical, thermal and vibration. Through timestamp alignment and coherence function analysis, it achieves for the first time accurate diagnosis of additional electromagnetic losses caused by mechanical vibration, with high accuracy in locating the loss source.

[0007] (2) This invention significantly improves the optimization efficiency and robustness of motor power supply control strategy through intelligent optimization and collaborative technology. The power supply performance closed-loop optimization unit integrates a loss tracing model and reinforcement learning algorithm, which can accurately decompose the five major losses such as copper loss and iron loss and generate differentiated optimization schemes, thereby improving the motor power supply efficiency by an average of 5% to 8%. The digital twin accelerated optimization engine reduces the number of physical iterations from the traditional 50 rounds to less than 10 rounds through high-fidelity simulation models and digital fingerprint parameter correction. Attached Figure Description

[0008] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0009] Figure 1 This is an overall flowchart of a test system for the power supply characteristics of a motor for new energy vehicles according to the present invention; Figure 2This is a flowchart of the test method for motor power supply characteristics according to the present invention; Figure 3 This is a flowchart of the closed-loop optimization process for the power supply performance of this invention. Figure 4 This is a flowchart of the cloud-based collaborative optimization process of the present invention; Figure 5 This is a schematic diagram of a three-dimensional structure of a test bench for the power supply characteristics of a motor for new energy vehicles according to the present invention; Figure 6 For the present invention Figure 5 Internal structure diagram; Figure 7 This is a schematic diagram of the three-dimensional structure of the fixed base of the present invention; Figure 8 This is a schematic diagram of the three-dimensional structure of the adjustable bracket of the present invention.

[0010] In the diagram: 100, simulated housing; 200, fixed base; 210, Y-axis moving drive component; 211, guide rail; 212, Y-axis moving drive rod; 220, adjusting housing; 230, Z-axis adjusting seat; 300, cast iron platform; 310, support base; 320, adjustable support; 330, moving slot. Detailed Implementation

[0011] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0012] Example 1: As Figure 1 - Figure 8 As shown, the present invention provides a test system for the power supply characteristics of a motor for new energy vehicles, which includes a physical environment simulation unit for providing a composite environmental field including temperature, humidity and mechanical vibration for the test motor. The integrated load simulation unit includes a mechanical inertia module, a road surface characteristic simulation module, and a braking load simulation module, which are used to apply a dynamic load coupled with vehicle inertia, road surface resistance characteristics, and braking intensity to the tested motor in the composite environment field. A multi-dimensional data sensing and fusion unit is used to synchronously and in real time collect the electrical, mechanical, thermal and vibration parameters of the tested motor under the composite environmental field and dynamic load. The intelligent power supply performance diagnosis and closed-loop optimization unit is communicatively connected to the multi-dimensional data sensing and fusion unit and the controller of the tested motor, and is used for: Based on the fused multi-dimensional data, the energy flow distribution during the power feeding process is analyzed in real time and key loss sources are located. Based on the characteristics of the key loss sources, optimized parameters for the regenerative braking control strategy of the tested motor are dynamically generated. The optimized parameters are sent to the controller of the motor under test, and the integrated load simulation unit is driven to perform a new round of testing to verify the optimization effect.

[0013] First, the physical environment simulation unit constructs a composite environmental field based on the test requirements, including a temperature range of -40℃ to 85℃, a humidity range of 10%RH to 90%RH, and a mechanical vibration frequency range of 0 to 2000Hz. The test motor is placed in this environment and stabilized for a preset time, usually 2 hours, to ensure that the motor reaches the environmental equilibrium temperature. Subsequently, the mechanical inertia module in the integrated load simulation unit simulates the rotational inertia corresponding to the vehicle's curb weight (in this embodiment, simulating a vehicle mass of 1800kg) through a flywheel assembly. The road surface characteristic simulation module generates the corresponding road surface resistance signal, and the braking load simulation module applies the load according to the braking intensity requirements. The three components are coupled to form a dynamic load that acts on the test motor. During this process, the multi-dimensional data perception and fusion unit synchronously collects the electrical, mechanical, thermal, and vibration parameters of the motor, achieving multi-source data fusion through timestamp alignment. Finally, the power supply performance closed-loop optimization unit analyzes the energy flow distribution based on the fused data, locates the loss source, generates control strategy optimization parameters, sends them to the motor controller, and drives the load simulation unit to execute a new round of testing to verify the optimization effect. Compared with the shortcomings of traditional motor test systems that can only perform single-factor or static testing, the composite environmental field simulation realizes coupled testing of the environment, load, and motor, which is closer to the actual operating conditions of the vehicle and solves the problem of the disconnect between traditional bench testing and actual vehicle performance. The integrated load simulation unit realizes the coordinated control of inertia, road surface, and braking load, and can reproduce complex power supply scenarios such as emergency braking and bumpy road recovery. The closed-loop optimization function enables testing and optimization to be carried out simultaneously, avoiding the cumbersome process of "testing first, then optimizing, and then retesting" in traditional testing, and greatly improving R&D efficiency. According to actual measurements, the error between the test results and the actual vehicle power supply efficiency is ≤3%, and the optimization iteration cycle is shortened by more than 40%.

[0014] In a winter low-temperature environment of -20℃ and 60%RH on an icy and snowy road surface, the physical environment simulation unit lowered the temperature of the sealed chamber to -20℃ and stabilized it for 2 hours. Simultaneously, random vibrations with a frequency of 5–20Hz and an acceleration of 1.5g were applied via a vibration table to simulate the bumps and vibrations of an icy and snowy road surface. In the integrated load simulation unit, the mechanical inertia module combined three sets of flywheels to simulate the inertia of a 1800kg vehicle. The road surface characteristic simulation module generated the resistance harmonic signal corresponding to an icy and snowy road surface with an adhesion coefficient of 0.2. The braking load simulation module applied a load with a braking intensity of 0.8g. Multi-dimensional data sensing unit… The sampling frequency for the three-phase current of the motor was 10kHz, the speed was 1kHz, the stator winding temperature was 10Hz, and the X / Y / Z triaxial vibration acceleration was 2kHz. The closed-loop optimization unit analysis of the power supply performance found that the change in the permeability of the permanent magnet at low temperature caused the iron loss to account for 45% of the total loss. The optimization parameters were then generated to adjust the switching frequency from 10kHz to 12kHz. After the parameters were issued and the test was repeated, the iron loss ratio dropped to 32%, and the power supply efficiency increased from 78% at room temperature to 72% at -20℃, which verified the optimization effect.

[0015] Ice water impact test: According to GB / T 18488-2024, the test motor was placed in an ice water environment at -20℃ for 30 minutes. After the impact, the insulation resistance was retested and found to be 120MΩ (meeting the requirement of ≥100MΩ). The power supply efficiency dropped from 82% to 80.5%. By adjusting the weak magnetic angle to 13°, the efficiency was restored to 81.8%.

[0016] As a preferred technical solution, the road surface characteristic simulation module is implemented through a programmable current harmonic injection unit to simulate the wheel slip ratio change caused by different road surface adhesion coefficients. This change is converted into a harmonic interference signal with a specific frequency and amplitude to the motor current loop.

[0017] The programmable current harmonic injection unit consists of an FPGA controller (Xilinx Kintex-7), a D / A converter (16-bit resolution, 1MHz sampling rate), a power amplifier (output bandwidth 0~10kHz, maximum output current 10A), and a harmonic signal conditioning circuit. It is connected to the current loop control port of the motor controller via a CANoe bus and can receive the current reference signal and superimpose harmonics in real time.

[0018] The core of road surface characteristic simulation is to convert changes in the road surface adhesion coefficient into harmonic interference in the motor current loop. Different road surfaces, such as dry asphalt, wet asphalt, and ice / snow, exhibit fixed characteristics in their adhesion coefficient-slip ratio curves. For example, the peak adhesion coefficient of dry asphalt is 0.85, corresponding to a slip ratio of 15%; the peak adhesion coefficient of ice / snow is 0.2, corresponding to a slip ratio of 20%. The system pre-stores standard curves for various road surfaces. During testing, based on the target road surface type, the FPGA controller calculates the slip ratio change curve over time. Then, using a preset algorithm (in this embodiment, Fourier series decomposition), this curve is converted into a current harmonic signal with a specific frequency of 0.1–100 Hz and an amplitude of 0–5 A. After D / A conversion and power amplification, this signal is superimposed on the current reference signal of the motor controller, causing corresponding fluctuations in the motor output torque, thereby simulating the impact of changes in the road surface adhesion coefficient on the power supply process.

[0019] Road surface characteristics are simulated by injecting current harmonics. The simulation has a fast response speed during the experiment, with the generation and superposition delay of harmonic signals ≤10μs, which can track dynamic road surface changes in real time. Moreover, it has high accuracy, with the adjustment accuracy of harmonic frequency and amplitude reaching 0.01Hz and 0.01A, respectively, which can accurately reproduce the subtle differences between different road surfaces. Different road surface types can be switched by modifying the FPGA program without replacing mechanical parts, which improves the testing efficiency by more than 60%.

[0020] As a preferred technical solution, the road surface characteristic simulation module includes a programmable harmonic current injector. It is configured to: calculate and generate corresponding current harmonic disturbance signals in real time based on the adhesion coefficient slip ratio curve corresponding to the input target road surface type, and superimpose them on the current reference signal of the motor controller to simulate the impact of the road condition on the electromagnetic torque and energy recovery stability of the motor.

[0021] The programmable harmonic current injector operates in three steps: The first step is curve import and analysis. The tester imports the adhesion coefficient-slip ratio curve of the target road surface via a host computer. The main CPU discretizes the curve, obtaining 1000 discrete data points within the slip ratio range of 0-100%. The second step is harmonic signal generation. Based on the discretized slip ratio data, the main CPU uses the least squares method to fit the time-domain function of the slip ratio. Then, it decomposes this function into the fundamental frequency and harmonic components using a fast Fourier transform. These components are then proportionally converted into current harmonic parameters (frequency, amplitude, and phase) according to the motor power rating, generating a digital harmonic signal. The third step is signal output and calibration. The digital signal is converted to a current reference signal in the motor controller after D / A conversion and power amplification. Simultaneously, the feedback calibration module collects the actual injected current signal in real time and compares it with the target signal. If the error exceeds ±0.5%, the power amplifier gain is adjusted via PID control to ensure the accuracy of the harmonic signal.

[0022] It supports importing custom road surface curves, breaking through the limitations of traditional systems that can only simulate preset road surfaces. It can reproduce the power supply characteristics of special road surfaces such as gravel roads and washboard roads. The addition of a feedback calibration module ensures that the amplitude error of harmonic current is ≤±0.3% and the frequency error is ≤±0.01Hz, significantly improving the accuracy of road surface simulation and ensuring the reliability of power supply test results. Comparative tests show that the road surface simulation error using this injector is reduced by 80% compared to an uncalibrated injector.

[0023] For the power supply test of the gravel road surface in the mountainous area, the testers first collected the adhesion coefficient-slip ratio curve of the gravel road surface through actual vehicle testing, saved it as a CSV file, and then imported it into a programmable harmonic current injector. The main CPU discretized the curve to obtain the fluctuation pattern of the slip ratio in the range of 0-30%, with an average slip ratio of 12% and a fluctuation amplitude of ±8%. From the main CPU, the fundamental frequency of 3Hz was decomposed by FFT to obtain the vehicle speed of 40km / h and the amplitude of 2.8A, the third harmonic frequency of 9Hz and the amplitude of 0.8A, and the fifth harmonic frequency of 9Hz and the amplitude of 0.8A. The harmonic parameters were set at 15Hz and 0.3A. After amplification, the signal was superimposed on the current reference signal. The feedback calibration module monitored in real time and found that the actual fundamental amplitude was 2.78A with an error of -0.71%. Then, the gain of the power amplifier was increased by 0.7% through PID adjustment to stabilize the fundamental amplitude within the range of 2.8A±0.01A. During the test, the regenerative braking torque of the motor fluctuated with the harmonic current. The fluctuation pattern matched the torque fluctuation curve of the actual vehicle on the gravel road surface with 95%, accurately simulating the influence of the gravel road surface on the power supply stability.

[0024] As a preferred technical solution, the multi-dimensional data perception and fusion unit includes: The electrical sensing subunit is used to collect motor terminal voltage, current, DC bus voltage, and high-frequency switching harmonics. The mechanical sensing subunit is used to collect torque, rotational speed, and multi-axial vibration acceleration. The thermal sensing subunit uses a combination of an infrared thermal imager and an embedded thermocouple to acquire the spatial temperature field distribution of the motor stator winding, permanent magnet and controller; The core of the data fusion is to align multi-source data based on timestamps and utilize the correlation between the vibration parameters and high-frequency current harmonics to diagnose specific electromagnetic losses caused by mechanical vibration.

[0025] The multi-dimensional data sensing and fusion unit consists of an electrical sensing subunit, a mechanical sensing subunit, a thermal sensing subunit, and a data fusion core. The electrical sensing subunit employs a high-precision power analyzer (Yokogawa WT3000, accuracy 0.01%) and a high-frequency oscilloscope with a bandwidth of 2GHz; the mechanical sensing subunit uses a torque sensor (accuracy ±0.1%FS), a photoelectric encoder (resolution 1024 lines), and a triaxial accelerometer (range ±50g, accuracy 0.001g); the thermal sensing subunit uses an infrared thermal imager (resolution 640×480, temperature range -40℃~1500℃, accuracy ±2%) and an embedded K-type thermocouple (accuracy ±0.5℃); and the data fusion core uses an industrial-grade server (CPU: Intel Xeon E5-2699v4, memory 64GB) equipped with custom data fusion algorithm software.

[0026] The multi-dimensional data perception and fusion unit operates in two phases: data acquisition and data fusion. In the data acquisition phase, each subunit synchronously acquires parameters at a preset sampling frequency. Electrical parameters such as motor terminal voltage and current are sampled at 10kHz; mechanical parameters such as torque and speed are sampled at 1kHz; thermal parameters at 10Hz; and vibration parameters such as acceleration are sampled at 2kHz. All parameters are provided by a GPS timing module with a high-precision timestamp, achieving a synchronization accuracy of 1μs. In the data fusion phase, the core first preprocesses the data acquired by each subunit, including outlier removal and noise reduction. Then, based on the timestamp, it aligns data from different sampling frequencies to a 1kHz time axis. Next, it uses correlation analysis between vibration parameters and high-frequency current harmonics to calculate coherence functions. For example, if the coherence coefficient of the 50Hz harmonic component in the current is ≥0.8 when the mechanical vibration frequency is 50Hz, it indicates the presence of additional electromagnetic losses caused by mechanical vibration, thus achieving deep fusion and loss diagnosis of multi-source data.

[0027] It offers comprehensive coverage, encompassing four major categories of parameters: electrical, mechanical, thermal, and vibration, fully reflecting the motor's power supply status. Precise synchronization of multi-source data is achieved through GPS timing and timestamp alignment, avoiding analysis errors caused by data delays. For the first time, it achieves accurate diagnosis of electromagnetic losses caused by mechanical vibration through correlation analysis of vibration and current harmonics, solving the problem of ambiguous loss source location in traditional testing. Testing shows that the loss source location accuracy of this unit exceeds 90%.

[0028] During the abnormal loss test of the motor power supply, the multi-dimensional data sensing unit synchronously collected data and found that when the motor speed was 1500 r / min, corresponding to a fundamental frequency of 25 Hz, the stator winding temperature rose abnormally, 12 ℃ higher than the normal operating condition. After data preprocessing, the data fusion core calculated the coherence function of vibration parameters and current harmonics, and found that the amplitude of the X-axis vibration acceleration reached 2.5g at 50 Hz, and the coherence coefficient of the 50 Hz harmonic component in the current was 0.86 with a threshold of 0.7. Further analysis revealed that the vibration originated from rotor eccentricity caused by motor bearing wear. Eccentricity caused air gap magnetic flux distortion, resulting in additional electromagnetic loss at 50 Hz, leading to temperature rise. Based on this diagnostic result, the testers replaced the bearing in time, and after retesting, the temperature returned to normal, verifying the effectiveness of this unit.

[0029] As a preferred technical solution, the power supply performance closed-loop optimization unit includes: The loss tracing model takes multi-dimensional data as input and outputs a quantitative proportion of total loss, which is decomposed into copper loss, iron loss, permanent magnet eddy current loss, mechanical loss and controller switching loss. The strategy optimization engine, based on the output of the loss tracing model, generates optimized parameters for adjusting the switching frequency or PWM modulation strategy if the iron loss ratio is high; and generates optimized parameters for adjusting the field weakening control strategy or current phase if the permanent magnet eddy current loss is abnormal. The strategy optimization engine integrates a reinforcement learning algorithm, which treats the motor controller parameters as actions of the intelligent agent and uses the reduction of the total power supply efficiency and specific loss components as rewards. Through continuous interaction with the environment, it autonomously seeks to obtain a more robust set of control strategy parameters.

[0030] The loss tracing model is built on MATLAB / Simulink and uses a neural network algorithm with 12 nodes in the input layer, 3 hidden layers, and 5 nodes in the output layer. The policy optimization engine uses an embedded processor (NVIDIA Jetson AGXXavier) and integrates reinforcement learning algorithms. The communication interface module supports EtherCAT and CAN bus and can communicate with the motor controller and integrated load simulation unit in real time.

[0031] First, the loss tracing model receives 12 feature parameters from multi-dimensional data fusion, including voltage, current, torque, speed, temperature, and vibration acceleration. It then uses a neural network algorithm to decompose the total loss, outputting the quantitative proportions of copper loss, iron loss, permanent magnet eddy current loss, mechanical loss, and controller switching loss. Next, the strategy optimization engine determines the dominant loss source based on the loss proportions. If the iron loss proportion is >30%, the switching frequency or PWM modulation strategy is adjusted; if the permanent magnet eddy current loss is >10%, the field weakening control strategy or current phase is adjusted. Simultaneously, the reinforcement learning algorithm treats the motor controller parameters as intelligent agent actions, using the improvement in total power supply efficiency and the reduction in dominant losses as reward values ​​(10 points for a 1% efficiency improvement, 5 points for a 10% loss reduction). Through continuous interaction with the test environment, 100 parameter combinations are generated per test round, autonomously optimizing to obtain the most robust parameter set. Finally, the optimized parameters are sent to the motor controller via the communication interface module, triggering the integrated load simulation unit to perform a retest.

[0032] The unit offers precise loss quantification, achieving accurate decomposition of the five major losses through a neural network model, providing clear targets for optimization. Its intelligent optimization strategy employs differentiated optimization schemes for different loss sources, avoiding blind adjustments. Through continuous interaction with the environment, it autonomously seeks optimization, resulting in a parameter set that is not only applicable to the current operating conditions but also adaptable to different environments and load changes, improving robustness by over 50%. Actual measurements show that this unit can improve motor feeding efficiency by an average of 5%–8%, and reduce optimization time to one-third of traditional methods.

[0033] After analyzing the data from the loss tracing model of the motor's high-temperature (60℃) power supply optimization test, the output loss ratio was found to be: copper loss 25%, iron loss 40%, permanent magnet eddy current loss 15%, mechanical loss 10%, and switching loss 10%, with iron loss being the dominant loss source. The strategy optimization engine first adjusted the switching frequency from 10kHz to 12kHz. After retesting, the iron loss ratio decreased to 35%, and the power supply efficiency improved by 3%. Subsequently, the reinforcement learning algorithm was initiated to optimize the switching frequency (8~15kHz) and PWM modulation method as the action space. After 50 rounds of interaction, the optimal parameters were obtained as a switching frequency of 13kHz + SVPWM modulation. After retesting with these parameters, the iron loss ratio decreased to 28%, and the power supply efficiency increased from the initial 75% to 82%. Moreover, it maintained high efficiency within the temperature range of 50~70℃ and the load range of 20%~100%, verifying the robustness of the optimization strategy.

[0034] As a preferred technical solution, the system also includes a cloud-based collaborative analysis platform, which is used to collect historical test data and optimized parameter sets of motors from different locations and of different models; When the power supply performance closed-loop optimization unit starts optimization, it can call the optimal initial parameter set of similar motor models from the cloud collaborative analysis platform and accelerate the local optimization process based on transfer learning technology.

[0035] The platform utilizes a cloud server cluster (8 servers, each configured with an Intel Xeon Gold 6248 processor, 128GB of memory, and a 10TB hard drive), and is equipped with a distributed database (MySQL cluster) and big data analytics software (Hadoop). The platform communicates with various test benches (supporting up to 100 units) via 4G / 5G or fiber optic networks, enabling data uploading and parameter distribution with a data transmission rate ≥100Mbps and latency ≤50ms.

[0036] First, during testing, each test bench uploads information such as the model of the motor under test, test environment parameters, multi-dimensional data, optimization parameters, and effects to the cloud database in real time, forming a historical test dataset that currently stores 100 motor models and 5,000 sets of test data. When a test bench starts testing a new motor, the power supply performance closed-loop optimization unit searches the cloud database for historical data of similar motor models based on key features such as motor model, rated power, and test conditions. A match is considered successful if the similarity is ≥80%. If a match is successful, the platform calls a transfer learning algorithm to fine-tune the historical optimal parameter set by 5% to 15% according to the current motor characteristics, generating an initial optimized parameter set and sending it to the test bench. The test bench directly uses these initial parameters to start testing, eliminating the need to iterate from the baseline parameters and accelerating the optimization process.

[0037] This approach solves the problems of "information silos" in traditional test benches and slow start-up in new motor optimization. It enables resource sharing, centralized management of test data and optimization experience from various test benches, avoids redundant testing, and reduces R&D costs. It accelerates the optimization process by reusing optimization results from similar motors through transfer learning, reducing the number of optimization iterations for new motors by more than 60%, making it particularly suitable for the rapid development of new motor models. Statistics show that adopting cloud-based collaboration has shortened the average R&D cycle for new motor power supply optimization by two months.

[0038] The first power supply optimization test of the permanent magnet synchronous motor (model YJS-150, rated power 150kW) was conducted. After the bench test was started, the cloud platform searched and found that the motor matched the stored YJS-120 motor (rated power 120kW, structurally similar, similarity 85%). The platform called the transfer learning algorithm to fine-tune the optimal parameter set of YJS-120 (switching frequency 12kHz, magnetic weakening angle 15°) according to the power characteristics of YJS-150, and obtained the initial parameters as switching frequency 13kHz and magnetic weakening angle 14°. The bench test was started with these parameters, and the power supply efficiency reached 78% in the first test. Compared with the traditional optimization starting from the baseline parameters, 12 fewer test iterations were required, and the optimization time was shortened from 3 days to 1 day, which greatly improved the R&D efficiency.

[0039] As a preferred technical solution, the physical environment simulation unit includes: A temperature and humidity-controlled sealed chamber is used to enclose the tested motor and its controller. A multi-exciter vibration table, installed in the temperature and humidity controllable sealed chamber, is used to apply six-degree-of-freedom road spectrum vibration to the motor mounting base.

[0040] The temperature and humidity controllable sealed chamber has a volume of 5m×3m×2.5m and adopts a dual temperature control method of liquid nitrogen refrigeration + electric heating, with a temperature control range of -40℃~85℃ and an accuracy of ±0.5℃. Humidity control uses steam humidification + dehumidifier dehumidification, with a humidity control range of 10%RH~90%RH and an accuracy of ±3%RH. The sealed chamber wall has a 100mm thick insulation layer with a thermal conductivity ≤0.03W / (m·K) to ensure temperature stability. The multi-vibrator vibration table adopts a 6-DOF structure (XYZ translation + XYZ rotation), driven by 6 electro-hydraulic servo vibrators, with an excitation frequency range of 0~2000Hz, a maximum acceleration of 50g, a displacement range of ±50mm, and a vibration control accuracy of ±1%.

[0041] First, the testers set the target temperature, humidity, and vibration parameters (frequency, acceleration, waveform) via a host computer according to the test requirements, such as high temperature and humidity or low temperature vibration. Then, the temperature and humidity controllable sealed chamber initiates its temperature and humidity control program. If the target temperature is lower than normal, the liquid nitrogen cooling valve is opened to cool the chamber through the evaporator, while the humidity control system is activated to adjust the humidity. If the target temperature is higher than normal, the electric heating element is activated to raise the temperature. The multi-vibrator vibration table, based on the set vibration parameters, is controlled by the controller to have six vibrators working collaboratively, applying vibration through force control mode. Once the temperature and humidity inside the chamber reach the target values ​​and the fluctuation is ≤±0.5℃ / ±3%RH, and the vibration parameters stabilize, the system enters a stable maintenance phase. PID feedback adjustment corrects deviations in real time, ensuring the stability of environmental parameters during the test.

[0042] The specific implementation method of the physical environment simulation unit was clarified, solving the problems of narrow temperature control range and insufficient vibration degrees of freedom in traditional environmental simulation equipment. It features a wide and highly accurate temperature and humidity control range, covering all working conditions from extreme cold to high temperature and from dry to high humidity, meeting the testing needs of different regions and climates. The 6-DOF vibration table can accurately reproduce the three-dimensional vibration during vehicle operation, which is closer to actual working conditions than traditional single-DOF or three-DOF vibration tables. Thirdly, the sealed chamber has excellent thermal insulation performance and small temperature fluctuations, ensuring that the motor is tested in a stable environment and improving data repeatability.

[0043] In the power supply test under high temperature and humidity environment (45℃, 85%RH) in tropical regions, after the testers set the target parameters, the temperature and humidity controllable sealed chamber activated the electric heating tube to raise the temperature to 45℃, and simultaneously turned on the steam humidifier to raise the humidity to 85%RH. The insulation layer effectively prevented heat loss, and the temperature fluctuation inside the chamber was maintained at 45±0.2℃, and the humidity fluctuation was maintained at 85±2%RH. The multi-exciter vibration table was based on the road spectrum of paved roads in tropical regions, collected from actual vehicles, and applied random vibrations with a frequency of 10-50Hz and an acceleration of 0.8g. The tested motor was stably tested in this environment for 2 hours. The collected data showed that the insulation resistance of the motor decreased by 5% and the power supply efficiency decreased by 2% under high temperature and humidity conditions, providing accurate data support for the high temperature and humidity adaptability design of the motor.

[0044] As a preferred technical solution, the system also includes a digital twin acceleration optimization engine; the engine contains a high-fidelity simulation model of the tested motor and its controller, which can perform simulation tests of a large number of control parameter combinations in parallel in the virtual space before the intelligent diagnosis and closed-loop optimization unit of the power supply performance starts physical iterative optimization, quickly pre-screen a subset of potential parameters, thereby greatly reducing the number of subsequent physical iterative optimizations and time. The simulation model of the digital twin acceleration optimization engine integrates digital fingerprint parameters that characterize the individual characteristics of the tested motor, learned from the historical data of the multi-source heterogeneous data synchronous sensing and fusion unit, so that the virtual simulation results are highly consistent with the physical measurement results.

[0045] The digital twin acceleration and optimization engine consists of a high-fidelity simulation model, a digital fingerprint parameter library, and a simulation calculation module. The high-fidelity simulation model is jointly built based on ANSYS Maxwell and Simplier, covering four sub-models of motor electromagnetics, mechanics, thermals, and control, with a model resolution of 0.1mm. The digital fingerprint parameter library stores individual characteristic parameters of each motor, such as air gap deviation, winding resistance deviation, and permanent magnet magnetic flux density deviation, which correspond one-to-one with the motor serial number. The simulation calculation module uses a GPU cluster, with a parallel computing capability of 100 TFLOPS.

[0046] The working logic of the digital twin accelerated optimization engine is virtual physical verification: First, historical test data of the tested motor is collected through a multi-dimensional data sensing unit. Machine learning algorithms are used to extract digital fingerprint parameters that characterize the individual characteristics of the motor, such as air gap non-uniformity and winding temperature coefficient deviation, and update them to the digital fingerprint parameter library. Then, the digital fingerprint parameters are input into a high-fidelity simulation model to correct the model, so that the error between the simulation results and the physical measurement results is ≤5%. When optimization is started, the simulation calculation module generates 1000 sets of control parameter combinations in the virtual environment according to a preset range, such as switching frequency 8~15kHz and current phase 0~30°, and executes simulation tests in parallel to calculate the power supply efficiency and loss distribution of each parameter combination. The parameter combinations with the highest efficiency are selected as a potential parameter subset and sent to the power supply performance closed-loop optimization unit. Finally, the closed-loop optimization unit only performs physical testing verification on this subset, which greatly reduces the number of physical iterations.

[0047] By using a digital fingerprint parameter correction model, the virtual simulation is highly consistent with the physical measurement, ensuring the reliability of the pre-screening results; the GPU cluster executes thousands of parameter simulations in parallel, completing the workload that would take 10 days for traditional physical testing in just 1 hour; the number of physical iterations is reduced from the traditional 50 rounds to less than 10 rounds, the testing cost is reduced by more than 70%, and the optimization cycle is shortened from 30 days to 7 days. Taking the power supply optimization test of mass-produced motors as an example, each batch consists of 10 motors, which need to be optimized one by one: First, extract the digital fingerprint parameters of the first batch of 1 motors, such as air gap non-uniformity of 0.05mm and winding resistance deviation of +2%. After correcting the high-fidelity simulation model, the simulation and measured efficiency error is 3.2%. Then, generate 1000 sets of control parameter combinations in a virtual environment, and after parallel simulation, select 10 sets of potential parameters with an efficiency of over 80%. Perform physical tests on these 10 sets of parameters to obtain the optimal parameter set, which takes 2 days. For the remaining 9 motors in the same batch, it is only necessary to extract their respective digital fingerprint parameter correction models and reuse the above potential parameter subset for virtual verification. Each motor only needs 2 sets of physical tests to determine the optimal parameters. The optimization time for a single motor is shortened to 4 hours, and the optimization time for the entire batch is shortened from the traditional 30 days to 5 days, which greatly improves the efficiency of batch testing.

[0048] Example 2: A test method for the power supply characteristics of a motor for new energy vehicles, comprising the following steps: S1: Environmental Coupling Settings: Set the parameters of the physical environment simulation unit and set the target working conditions in the integrated load simulation unit to form a composite test scenario; The environmental parameters were set as follows: temperature 25℃, humidity 50%RH (normal temperature and humidity); vibration parameters: frequency 10~30Hz, acceleration 0.5g (simulating urban paved road surface); load parameters: braking intensity 0.6g, vehicle speed reduced from 80km / h to 20km / h, to construct a composite test scenario. S2: Multidimensional data synchronous acquisition: Drive the system to operate and synchronously acquire all-dimensional parameters of the tested motor under the composite test scenario through the multidimensional data sensing and fusion unit; Start the motor power supply test and continuously collect data for 15 minutes to obtain data such as current, voltage, torque, speed, temperature, and vibration. After fusion, generate a standardized data file. S3: Loss Source Tracing and Problem Localization: Based on fused data, the loss source tracing model is used to quantitatively analyze the composition and proportion of various losses in real time, and to locate the dominant loss source; The loss source model outputs the following loss proportions: copper loss 30%, iron loss 35%, permanent magnet eddy current loss 15%, mechanical loss 10%, and switching loss 10%, with iron loss being the dominant loss source. Correlation analysis shows that excessive iron loss is related to low switching frequency. S4: Closed-loop optimization of control strategy: Based on the characteristics of the dominant loss source, the strategy optimization engine dynamically generates control parameter adjustment suggestions and sends them to the motor controller; The strategy optimization engine adjusts the switching frequency from 10kHz to 13kHz, generates optimization parameters, and distributes them. S5: Optimization effect verification and iteration: Under the premise of keeping the environment and load conditions unchanged, retest using new control parameters, compare the performance indicators before and after optimization, and repeat steps S3 to S5 if the target is not met until the optimal result is achieved.

[0049] With the environment and load unchanged, the iron loss ratio was reduced to 28%, and the power supply efficiency increased from 76% to 80%, meeting the target (efficiency improvement ≥3%), and the optimization was completed. The entire process took 2 days, which is 5 days shorter than the traditional method.

[0050] As a preferred technical solution, in step S3, the specific method for locating the dominant loss source includes: analyzing the coherence function of the vibration acceleration amplitude at a specific frequency and the corresponding frequency harmonic component in the current; if the coherence coefficient is higher than the threshold, it is determined that there is additional iron loss caused by mechanical vibration.

[0051] The coherence function γ²(f) measures the degree of linear correlation between two signals at a specific frequency, ranging from 0 to 1. The closer γ²(f) is to 1, the stronger the correlation between the two signals. In this method, mechanical vibration acceleration signals (such as X, Y, and Z directions) and motor current signals are first extracted, and Fourier transforms are performed on both signals to obtain the frequency domain signals. Then, the coherence function at a specific frequency f is calculated. The formula is: γ²(f) = |G(f)|² / [G(f)G(f)], Where G(f) is the cross power spectral density of vibration and current signals, and G(f) and G(f) are the self power spectral densities of vibration and current signals, respectively. A coherence coefficient threshold is set (usually 0.7~0.8, adjusted according to the test accuracy requirements). If the coherence coefficient at a certain frequency is higher than the threshold, it is determined that the mechanical vibration at that frequency has triggered the current harmonics at the corresponding frequency, thereby generating additional iron loss.

[0052] As a preferred technical solution, in step S1, when the set target operating condition has similar historical optimization records in the cloud knowledge base, the system automatically recommends using the historical optimal parameters adjusted by transfer learning as the initial strategy, thereby skipping unnecessary benchmark degradation tests and directly entering the fine optimization stage.

[0053] The method achieves a quantitative assessment of the correlation between vibration and electromagnetic loss through the calculation of coherence functions, avoiding the subjectivity of empirical judgment. It can clearly identify the vibration frequency and direction that causes additional iron loss, providing a precise basis for fault diagnosis. It is applicable to the testing of power supply losses in various types of motors, without being limited by motor model or power rating. Testing shows that this method achieves an accuracy rate of over 95% in locating additional iron loss caused by mechanical vibration.

[0054] X / Y / Z triaxial vibration acceleration signals and A-phase current signals were collected. Fourier transforms were performed on the signals, and the coherence function was calculated. The results showed that the Z-axis vibration acceleration amplitude reached 3.0g at 75Hz, and the coherence coefficient of the 75Hz harmonic component in the A-phase current was 0.82 (threshold set to 0.75). Further investigation revealed that 75Hz was three times the rotor speed (1500 r / min, fundamental frequency 25Hz), and the vibration originated from insufficient rotor dynamic balance accuracy (imbalance of 0.5g·cm). After dynamic balance correction, retesting showed that the 75Hz vibration amplitude decreased to 0.5g, the coherence coefficient decreased to 0.3, and the additional iron loss decreased from 120W to 30W, verifying the effectiveness of this method.

[0055] As a preferred technical solution, in step S1, when the set target operating condition has similar historical optimization records in the cloud knowledge base, the system automatically recommends using the historical optimal parameters adjusted by transfer learning as the initial strategy, thereby skipping unnecessary benchmark degradation tests and directly entering the fine optimization stage.

[0056] First, the cloud-based knowledge base stores all historical test data, including information such as motor model, test environment parameters, load conditions, optimization parameters, and effects. An index is created with fields including motor power, speed range, test temperature, and braking intensity. After setting the target operating condition in step S1, the system automatically extracts key feature parameters of the target condition (motor model YJS-150, temperature 25℃, braking intensity 0.6g) and performs similarity matching with the cloud-based knowledge base index using a cosine similarity algorithm. Similarity = Dot product of target feature and historical feature / (target feature magnitude × historical feature magnitude); If a historical optimization record with a similarity of ≥80% is matched, the system calls the transfer learning algorithm (based on a domain adaptive network) to fine-tune the historical optimal parameter set according to the individual characteristics of the current motor (reflected by digital fingerprint parameters) to obtain the initial strategy parameters adapted to the current motor. After confirmation by the testers, the test is started directly using these initial parameters, skipping the traditional benchmark degradation test, i.e., starting the test from the worst parameters, and directly entering the fine optimization stage.

[0057] Historical data reuse and transfer learning shorten the testing cycle, skipping baseline degradation testing and directly using optimized initial parameters, reducing the number of test rounds by more than 50%. It also improves initial efficiency, as the initial parameters already possess high performance, avoiding risks such as motor overheating due to parameter degradation during testing. Thirdly, knowledge reuse enables the effective transfer of historical optimization experience, making it particularly suitable for batch testing of motors in the same series. Statistics show that using this method reduces the average testing time for a single motor by 40%.

[0058] Taking the batch testing of a series of motors (YJS-120, YJS-150, YJS-180, with the same structure but different power) as an example: When testing the YJS-180 motor (rated power 180kW), step S1 sets the target operating conditions as a temperature of 25℃, braking intensity of 0.6g, and vehicle speed of 60~20km / h; after the system extracts the feature parameters, it matches them with the cloud knowledge base and finds that the similarity with the historical test records of the YJS-150 motor is 88% (historical parameters: switching frequency 13kHz, magnetic weakening angle 14°, efficiency 82%); the transfer learning algorithm is called, and the parameters are fine-tuned to a switching frequency of 14kHz and a magnetic weakening angle of 13° according to the power characteristics of the YJS-180 (power increase of 20%); the test is started directly with these parameters, and the initial power supply efficiency reaches 81%, which is 8 fewer rounds of testing than the traditional method of starting from the baseline parameter efficiency of 70%, and the test time is shortened from 2 days to 1 day; after fine-tuning, the final efficiency reaches 83%, which meets the optimization target.

[0059] Furthermore, based on the aforementioned road surface characteristic simulation module, the road surface adhesion characteristics are not directly converted into current harmonics according to the preset amplitude, but are sequentially processed through vehicle longitudinal force calculation, wheel end braking torque conversion, motor shaft end braking torque conversion, and current disturbance amount back calculation, so that the disturbance amount injected into the motor controller has a definite correspondence with the actual force characteristics of the target road surface.

[0060] Before the test begins, the relationship between the adhesion coefficient and slip ratio of the target road surface is stored as discrete data. During vehicle operation, the current slip ratio is first calculated based on the vehicle's equivalent linear velocity and the wheel's circumferential velocity. Specifically, the wheel's circumferential velocity is obtained by multiplying the wheel's angular velocity by the effective rolling radius of the tire. The difference between the vehicle's equivalent linear velocity and the wheel's circumferential velocity is then used as the speed difference. The current slip ratio is obtained by dividing the speed difference by the vehicle's equivalent linear velocity.

[0061] Considering the relatively small equivalent linear velocity of the vehicle when it approaches a stop, direct calculation would amplify the error of a single encoder count. Therefore, this implementation sets a low-speed calculation cutoff value. This cutoff value is not based on a pre-fixed empirical speed, but is determined according to the lowest speed change that the current speed measurement link can resolve. Specifically, first, the vehicle travel distance corresponding to one encoder count is obtained by dividing the circumference of the tire by the effective count generated per encoder revolution; then, this travel distance is divided by the current speed sampling period to obtain the lowest resolvable vehicle speed corresponding to one count; finally, three times this lowest resolvable vehicle speed is taken as the low-speed calculation cutoff value.

[0062] For example, if the effective rolling radius of a tire is 0.32 meters, and the encoder generates 4096 valid counts per revolution after quadrupling the frequency, then the circumference of one tire revolution is approximately 2.01 meters, and the vehicle travel distance corresponding to each valid count is approximately 0.00049 meters. When the rotational speed sampling period is 1 millisecond, the speed resolution corresponding to one count is approximately 0.49 meters per second. Therefore, the low-speed calculation cutoff is approximately 1.47 meters per second. When the vehicle's equivalent linear velocity is higher than this value, the slip ratio is calculated based on the actual vehicle's equivalent linear velocity; when the vehicle's equivalent linear velocity is lower than this value, 1.47 meters per second is used as the denominator for the slip ratio calculation. After this processing, the slip ratio will not suddenly increase due to encoder quantization errors when the vehicle approaches a stop. After obtaining the current slip ratio, two adjacent data points located on either side of the current slip ratio are found in the pre-stored discrete data of adhesion coefficient and slip ratio. The current adhesion coefficient is then calculated based on the positional ratio of the current slip ratio between the two data points. The resulting adhesion coefficient changes continuously without abrupt changes due to the switching of discrete data points.

[0063] Subsequently, the current adhesion coefficient is multiplied by the vehicle's equivalent mass corresponding to the mechanical inertia module and the gravitational acceleration to obtain the longitudinal adhesion force that the target road surface can generate. For example, if the mechanical inertia module corresponds to a vehicle mass of 1800 kg and the current adhesion coefficient of the icy road surface is 0.2, then 1800 is first multiplied by 9.81 to obtain an equivalent vertical load of 17658 Newtons, and then multiplied by 0.2 to obtain a target longitudinal adhesion force of 3531.6 Newtons.

[0064] After determining the target longitudinal adhesion force, multiply this adhesion force by the effective rolling radius of the tire to obtain the braking torque corresponding to the wheel end. Then, based on the actual transmission ratio and transmission efficiency used between the mechanical inertia module and the tested motor, convert the wheel end braking torque to the shaft end of the tested motor. During the conversion, the wheel end braking torque is divided by the transmission ratio and then by the transmission efficiency to obtain the target braking torque that the tested motor should generate.

[0065] Taking a tire effective rolling radius of 0.32 meters as an example, the wheel-end braking torque corresponding to a longitudinal adhesion force of 3531.6 Newtons is approximately 1130.1 Newton-meters. When the test bench transmission ratio is 9.1 and the transmission efficiency is 95%, dividing 1130.1 by 9.1 and 0.95 successively yields a target braking torque of approximately 130.7 Newton-meters at the motor shaft end. This calculation result serves as the target quantity for subsequent back-calculation of current disturbances.

[0066] The calculation of current disturbance uses the pre-calibrated torque-current relationship of the tested motor at the current speed and under the current field weakening state. During calibration, stable electromagnetic torque is recorded under different speeds, different direct-axis currents, and different quadrature-axis currents to form a torque calibration data table. During testing, adjacent calibration points are called according to the current speed and the current field weakening state, and the torque increment generated by each unit increase in quadrature-axis current under the current state is obtained by interpolation.

[0067] For example, under the current speed and field weakening conditions, each 1 ampere increase in quadrature-axis current increases the electromagnetic torque by 1.6 Newton-meters. Therefore, the aforementioned target braking torque of 130.7 Newton-meters corresponds to approximately 81.7 amperes of quadrature-axis current. Subtracting the quadrature-axis current used to achieve the basic braking torque under the current undisturbed road surface conditions from 81.7 amperes yields the difference, which is the amount of quadrature-axis current disturbance that needs to be added by the road surface characteristic simulation module.

[0068] Through the above processing, the target road surface adhesion coefficient is first converted into the vehicle's longitudinal adhesion force, and then into the wheel-end braking torque, the motor shaft-end braking torque, and the cross-axis current disturbance, so that the amplitude of the current disturbance has a clear source of vehicle dynamics.

[0069] For the continuously acquired quadrature-axis current disturbance sequence, further frequency component decomposition is performed. Each frequency component is sorted from largest to smallest contribution to the original disturbance signal, and the main frequency components are retained. The number of harmonics retained is not fixed but determined based on the actual reconstruction error of the current-injected link.

[0070] Specifically, before testing, the current injection link is calibrated using a standard signal source. Target disturbance signals are input at multiple frequencies and current amplitudes, and the actual output values ​​are recorded. The relative error at each test point is obtained by dividing the difference between the target value and the actual value by the target value. The maximum relative error among all test points is taken as the upper limit of the allowable reconfiguration error for the current current injection link.

[0071] When performing frequency component screening, first calculate the average of the squares of all sampled values ​​in the original perturbation sequence to characterize the total energy of the original signal; then calculate the average of the squared differences between the reconstructed signal and the original signal after retaining several major frequency components to characterize the residual energy of the unretained parts; finally, take the square root of both the total energy and the residual energy, and divide the square root of the residual energy by the square root of the total energy. Stop adding frequency components when the resulting ratio is no greater than the maximum relative error obtained from the actual calibration of the current injection link.

[0072] Therefore, for icy and snowy roads with a relatively simple spectrum, retaining only a small number of major frequency components is sufficient to meet the reconstruction requirements; for roads with a wider spectrum, such as gravel roads or washboard roads, the number of frequency components participating in the reconstruction is automatically increased. This process avoids oversimplification or invalid calculations caused by using a fixed number of harmonics for different road surfaces.

[0073] Before the output current disturbance, peak constraints are applied. First, the peak value of the current base control current is subtracted from the maximum allowable phase current of the tested motor to obtain the current margin. Then, the maximum absolute value of the generated disturbance signal is compared. When the peak value of the disturbance signal does not exceed the current margin, the original amplitude is output. When the peak value exceeds the current margin, the current margin is divided by the original peak value of the disturbance signal to obtain a uniform scaling ratio, and the amplitude of all frequency components is multiplied by this scaling ratio. Because each frequency component is scaled using the same ratio, their relative spectral relationships remain unchanged, while preventing the motor from entering overcurrent protection mode during the injection process.

[0074] This establishes a continuous physical mapping relationship from road surface adhesion characteristics to motor current disturbances, enabling the influence of different road surface conditions on regenerative braking torque and power supply process to be reproduced through controlled current disturbances.

[0075] Furthermore, regarding the synchronous processing of multi-source data during variable-speed power supply, during the regenerative braking process of new energy vehicles, the speed of the tested motor continuously changes. If the 10 kHz current data, 2 kHz vibration data, and 1 kHz speed data are directly reduced to the same sampling frequency for fixed-frequency analysis, the actual frequency corresponding to the same mechanical or electromagnetic order will continuously change during deceleration, easily forming an extended band in the spectrum, thus affecting the identification of loss sources.

[0076] To this end, the data fusion core first performs fixed delay correction on each acquisition channel.

[0077] Before the test, the same hardware trigger signal was simultaneously input to the electrical sensing subunit, mechanical sensing subunit, and vibration acquisition channel, and the time when each acquisition channel received the trigger signal was recorded. The channel that received the trigger signal earliest was used as the time reference, and the difference between the other channels and this time reference was used as their respective fixed transmission delay. During the formal test, the original timestamp of each channel was reduced by the fixed transmission delay of that channel to obtain the corrected sampling time. After the fixed delay correction was completed, instead of directly reducing all high-speed data to 1 kHz, the electrical angle reference was constructed using the speed of the tested motor. Specifically, the mechanical angular displacement between two adjacent sampling times was obtained by multiplying the mechanical angular velocity within that time period by the sampling time interval; the mechanical angular displacement was then multiplied by the number of pole pairs of the tested motor to obtain the corresponding electrical angle increment; the electrical angle increments of each sampling period were continuously accumulated to obtain the continuously changing electrical angle position throughout the test.

[0078] Subsequently, current, torque, and vibration data are resampled at equal electrical angle intervals, ensuring that each electrical cycle corresponds to the same number of data points. For example, if each electrical cycle contains 256 resampling points, then regardless of whether the motor is operating at 1500 rpm or 3000 rpm, the same electromagnetic order will be located at the same order position. Reducing the speed only changes the interval between these points on the actual time axis, not their relative position on the electrical angle axis.

[0079] The minimum number of resampling points required for each electrical cycle is determined based on the highest order that needs to be analyzed. First, the common analysis bandwidth is determined based on the highest effective frequency that can be reliably analyzed by both the current and vibration channels. Then, the common analysis bandwidth is divided by the lowest electrical frequency during the test to obtain the highest possible effective order. Each electrical cycle should have at least twice this highest effective order as the number of data points. To improve the efficiency of fast spectrum processing, after meeting the above requirements, the nearest power of 2 can be selected as the actual number of resampling points.

[0080] For example, if the common effective analysis frequency is limited to 1000 Hz, and the lowest electrical frequency during the test is 25 Hz, then 1000 divided by 25 equals 40, therefore at least 40 orders need to be identified. The number of sampling points per electrical cycle should be at least higher than 80 points; in practice, 128 or 256 points can be used. Using 256 points can provide a stable data length for subsequent spectrum calculations while retaining sufficient order resolution.

[0081] Since the rate of change of thermal parameters is significantly lower than that of current, vibration, and torque parameters, high-frequency interpolation is not used to generate spurious temperature changes. Instead, linear interpolation is performed between two adjacent measured temperature points based solely on their timestamps, mapping the temperature data to the corresponding analysis time window. Therefore, thermal data is only used to determine the thermal state and temperature rise trend within the corresponding time window and is not involved in high-frequency correlation judgments exceeding its actual sampling bandwidth.

[0082] After the above processing, during the entire power supply process from 80 km / h to 20 km / h, the same mechanical order and electromagnetic order can be compared under a unified angular reference, providing a stable data basis for the correlation analysis between mechanical vibration and current harmonics.

[0083] It should be noted that the method for determining additional electromagnetic losses caused by mechanical vibration is not based on the conclusion that a single coherence coefficient exceeds a fixed value. Instead, the determination of the benchmark threshold, frequency correlation screening, and loss increment verification are performed sequentially.

[0084] First, before applying the target road vibration, the tested motor was operated under the same speed, torque, temperature, and control parameters as the target test, without applying active excitation at the frequency to be analyzed to the vibration table. One hundred equal-length analysis time windows were continuously obtained under this steady-state condition.

[0085] For each analysis time window, the vibration acceleration signal and the corresponding phase current signal in a certain direction are obtained, and frequency domain analysis is performed on both. For the frequency to be analyzed, the power spectrum of the vibration signal itself, the power spectrum of the current signal itself, and the cross-power spectrum between the vibration and current signals are calculated. The cross-power spectrum amplitude is first squared, then divided by the product of the vibration signal power spectrum and the current signal power spectrum to obtain the coherence level at that frequency. The obtained value is between 0 and 1; the closer it is to 1, the more obvious the linear correlation between the two signals at that frequency.

[0086] After obtaining the coherence degree for each of the 100 benchmark analysis time windows, the 100 values ​​are arranged in ascending order, and the 99th value is used as the correlation threshold for the frequency under the current rotational speed, current load, and current measurement link conditions.

[0087] This threshold is not fixed at 0.7 or 0.8. For example, under a certain test condition, if the 99th coherence value in 100 reference windows is 0.68, then in the formal test, the coherence at that frequency must be greater than 0.68 to proceed to the next step. If, under another test condition, the 99th reference coherence value reaches 0.76 due to strong mechanical background vibration, then 0.76 is used as the threshold for that test condition in the formal test. Thus, the judgment threshold varies with the background noise of the test bench, the sensor status, and the operating conditions of the tested motor.

[0088] Even if the correlation threshold is exceeded, additional electromagnetic losses cannot be directly identified. Further, keeping the motor speed, average braking torque, ambient temperature, and control parameters constant, only the excitation amplitude of the vibration table at the target frequency is adjusted.

[0089] For example, the 50 Hz vibration acceleration is increased from 0.5 times the gravitational acceleration to 2.5 times the gravitational acceleration, and the mechanical input power and DC-side feedback power are recorded in both states. The mechanical input power is obtained by multiplying the real-time torque by the mechanical angular velocity, and the DC-side feedback power is obtained by multiplying the DC bus voltage by the DC bus current. Subtracting the DC-side feedback power from the mechanical input power yields the total loss in the current state.

[0090] Assuming an average mechanical input power of 40 kW and an average DC feedback power of 36.20 kW under 0.5 times gravitational acceleration, the total loss is 3.80 kW. When the vibration is increased to 2.5 times gravitational acceleration, the average mechanical input power remains at 40 kW, while the average DC feedback power decreases to 36.05 kW, resulting in a total loss of 3.95 kW. The total loss increase between the two states is 150 kW.

[0091] Since enhanced vibration can also directly increase mechanical losses such as bearing friction, it is necessary to deduct the mechanical-side increment. Mechanical-side calibration can be performed on the same test bench under a mechanically driven state that does not generate effective electromagnetic torque. The aforementioned two vibration levels are applied at the same rotational speed, and the increase in pure mechanical loss between the two states is measured. For example, if the measured increase in pure mechanical loss is 30 watts, then after deducting the 30 watts of mechanical loss from the aforementioned total loss increase of 150 watts, the additional electromagnetic loss related to vibration of 120 watts is obtained.

[0092] To determine whether the 120-watt increment exceeded the test fluctuation range, 100 total loss data points were continuously acquired under a baseline condition without targeted vibration excitation. First, the average of the 100 data points was calculated. Then, the difference between each data point and the average was calculated. The squares of each difference were then averaged. Finally, the square root of the resulting average was taken to obtain the standard deviation of the baseline total loss.

[0093] For example, if the standard deviation of the baseline total loss is 18 watts, then three times the standard deviation is 54 watts. Since the additional electromagnetic loss calculated above is 120 watts, which is greater than 54 watts, and the coherence between the vibration and the current at 50 Hz simultaneously exceeds the correlation threshold corresponding to the current operating condition, it is determined that the 50 Hz mechanical vibration has formed a distinguishable additional electromagnetic loss.

[0094] Only when both the correlation condition and the loss increment condition are met is the corresponding frequency determined as the effective loss excitation frequency. This determination method can exclude cases where a high degree of coherence exists solely due to the joint response of the mechanical structure, but no discernible electromagnetic loss is actually caused.

[0095] For triaxial vibration signals, the above judgment is performed on the transverse, longitudinal, and vertical vibrations respectively. When multiple directions exceed the correlation threshold at the same frequency, the vibration amplitude in each direction is controlled to change by the same amount, and the magnitude of the resulting additional electromagnetic loss increment is used to determine the main vibration direction.

[0096] Furthermore, regarding the implementation method for determining the dominant loss source and adjusting the control parameters, after the loss tracing model outputs copper loss, iron loss, permanent magnet eddy current loss, mechanical loss, and controller switching loss, it does not directly use iron loss exceeding 30% or permanent magnet eddy current loss exceeding 10% as a fixed optimization trigger condition. Instead, it combines the dispersion of the current test data to determine whether there is a statistically distinguishable dominant loss source.

[0097] After the current operating conditions stabilize, 100 analysis time windows are continuously obtained. For each time window, the proportion of the five types of losses to the total loss is calculated, and then the average proportion and standard deviation of each type of loss are calculated for the 100 time windows.

[0098] The standard deviation is calculated as follows: First, calculate the average of 100 percentage data for a certain type of loss. Then, calculate the difference between each percentage and the average. Square all the differences and calculate the average. Finally, take the square root of the result.

[0099] The loss with the highest average percentage is designated as the first candidate loss, and the loss with the second highest average percentage is designated as the second candidate loss. First, calculate the difference between their average percentages; then, square their standard deviations respectively, add the two squares, take the square root, and multiply the result by 3. Only when the difference in average percentage between the first and second candidate losses is not less than this result is the first candidate loss identified as the single dominant loss source.

[0100] For example, across 100 analysis time windows, iron loss accounted for an average of 38% with a standard deviation of 1.2 percentage points; copper loss accounted for an average of 31% with a standard deviation of 1.0 percentage point. The average percentage difference between iron and copper losses is 7 percentage points. Square 1.2 gives 1.44, square 1.0 gives 1, adding them together gives 2.44. Taking the square root of this is approximately 1.56, multiplying by 3 gives approximately 4.68 percentage points. Since 7 percentage points is greater than 4.68 percentage points, iron loss is determined to be the single dominant loss source.

[0101] If the average proportion of iron loss is 34% and the average proportion of copper loss is 32%, and the standard deviations are still 1.2 percentage points and 1.0 percentage points respectively, the difference between the two is only 2 percentage points, which is less than 4.68 percentage points. Therefore, iron loss will not be identified as the dominant loss alone, but both iron loss and copper loss will be considered as optimization targets. This avoids incorrect optimization direction caused by a single measurement fluctuation when the proportions of each loss are close.

[0102] After determining the dominant loss, the strategy optimization engine determines the effective direction of the parameter to be adjusted through local bidirectional perturbation, rather than pre-defining that the switching frequency must be increased when the iron loss is high.

[0103] Taking the switching frequency as an example, when the current switching frequency is 10 kHz, 1 kHz is used as the initial trial step size. Keeping other control parameters, speed, load and environmental conditions unchanged, short-time stability tests are performed at 9 kHz and 11 kHz respectively, and the dominant losses in the two states are calculated.

[0104] If the iron loss is 1.50 kW at 9 kHz and 1.34 kW at 11 kHz, then increasing the switching frequency by 2 kHz reduces the iron loss by 0.16 kW. This indicates that increasing the switching frequency can reduce the dominant iron loss within the current local range. The next round of optimization will search for higher frequencies from 10 kHz.

[0105] Conversely, if the iron loss is 1.30 kW at 9 kHz and 1.42 kW at 11 kHz, then increasing the switching frequency will increase the dominant loss, and the search should be directed to a lower frequency. This directional determination is based on the actual loss changes of the tested motor and the current operating conditions, rather than on a fixed adjustment direction according to preset rules.

[0106] The same bidirectional probing process is used for the weak magnetic angle, current phase, and other continuous control parameters.

[0107] After the parameters are adjusted, it is necessary to determine whether the performance improvement in this round exceeds the fluctuation of the test itself. First, obtain 100 feed efficiency data points in the stable state before optimization, and calculate the efficiency standard deviation using the same method as described above. Subtract the average efficiency of the old parameters from the average efficiency obtained from the new parameter test. Only when the increase in efficiency is greater than 3 times the efficiency standard deviation is the parameter adjustment considered to have formed an effective improvement.

[0108] For example, if the average efficiency of the optimized feedforward is 76.0% with a standard deviation of 0.25 percentage points, then three times the standard deviation would be 0.75 percentage points. If adjusting a certain parameter increases the average efficiency to 76.4%, although this represents a 0.4 percentage point increase, it is still within the test fluctuation range, and therefore the optimization is not considered effective. If the adjusted average efficiency reaches 77.2%, an increase of 1.2 percentage points, exceeding 0.75 percentage points, and the motor current, DC bus voltage, winding temperature, and controller temperature are all below the protection limits, then this parameter is retained.

[0109] When parameter adjustments cause a decrease in efficiency or an increase in losses, the parameter change in this round is halved, and a bidirectional trial is performed again. For example, if the initial adjustment step size of the switching frequency is 1 kHz, and the direction is reversed, the step size is reduced to 0.5 kHz in the next round; if the direction is reversed again, it is reduced to 0.25 kHz. When the parameter step size is reduced to the minimum adjustment resolution that the controller can perform, or when the efficiency improvement obtained from three consecutive rounds of optimization does not exceed three times the efficiency standard deviation, the search for the current parameter is stopped.

[0110] This method allows the identification of dominant loss, the direction of parameter adjustment, and the optimization termination conditions to be determined by the current measured data.

[0111] Specifically, the training data for the loss tracing model does not use manually labeled data whose sources cannot be verified. The training target values ​​for the five types of loss are obtained based on actual measurements, parameter identification results, and calibrable loss models, respectively.

[0112] Copper losses are calculated based on real-time phase currents and the actual resistance of the stator windings at the current temperature. The stator resistance is first measured at a reference temperature and then converted to the current winding temperature based on the temperature coefficient of the winding material. When calculating the three-phase copper losses, the effective value of the current in each phase is squared, multiplied by the resistance at the corresponding reference temperature, and finally the three phase results are added together.

[0113] Mechanical losses are obtained through drag calibration. With the motor not outputting effective electromagnetic torque, the test motor is driven by a load motor at multiple speed points, and the shaft-end power required to maintain that speed is recorded. The shaft-end power obtained at each speed point is stored to form calibration data showing the mechanical losses as a function of speed. During formal testing, the mechanical losses are obtained by interpolation based on the current speed.

[0114] The controller switching loss is determined by combining the difference between the controller's DC input power and AC output power with the conduction loss calibration results. First, the calibration relationship between the power device's conduction loss and current is obtained through low switching frequency and different current conditions. Then, the total controller loss is measured at the target switching frequency. The switching loss is obtained by subtracting the conduction loss under the same current conditions from the total controller loss.

[0115] After obtaining the copper loss, mechanical loss, and controller loss, the difference between the mechanical input power and the actual feedback power on the DC side is taken as the total loss. The total loss is then subtracted from the copper loss, mechanical loss, and controller loss, which can be directly measured or calibrated, and the remaining portion is taken as the total electromagnetic loss inside the motor.

[0116] Iron losses and permanent magnet eddy current losses are then separately calibrated through changes in operating conditions. During calibration, the torque is kept nearly constant while the motor speed is varied, and the variation of electromagnetic losses with electrical frequency is recorded. Subsequently, while keeping the speed constant, the harmonic magnetic field on the permanent magnet is changed by adjusting the current phase, and the changes in rotor temperature rise and electromagnetic losses are recorded simultaneously. By using data from multiple operating conditions for joint fitting, the iron loss component is made to primarily correspond to changes in electrical frequency, while the permanent magnet eddy current loss component is constrained by both high-frequency magnetic field and rotor temperature rise changes.

[0117] Only when the difference between the sum of the five types of losses and the total loss calculated from the mechanical input power and the DC-side feedback power is within the allowable error range of the power measurement link, can this set of data be used as a valid training sample for the loss tracing model.

[0118] For example, if the mechanical input power is 50 kW and the DC-side feedback power is 45 kW, the total loss is 5 kW. According to the component model, the copper loss is 1.4 kW, the iron loss is 1.8 kW, the permanent magnet eddy current loss is 0.6 kW, the mechanical loss is 0.5 kW, and the controller loss is 0.65 kW. Adding these five components together gives 4.95 kW, which is 0.05 kW less than the total loss, i.e., 50 kW. If the allowable range of the current power measurement link's overall error is 75 kW, then this sample meets the energy conservation error requirement and can be included in the model training dataset. If the sum of the five components is only 4.6 kW, which is 400 kW less than the total loss, then this set of data will not be directly used as a training sample, and the sensor synchronization status and loss calibration results will be re-examined.

[0119] This process ensures that the output of the loss tracing model is constrained by the measured energy conservation relationship, avoiding the lack of verifiable physical basis when relying solely on the data model to directly output the proportions of the five types of losses.

[0120] Regarding the acquisition and reliability determination of digital fingerprint parameters in the digital twin model: the digital fingerprint parameters are not arbitrarily extracted directly from historical test data, but are obtained by parameter inversion based on the error between the measured output of the tested motor and the output of the digital twin model.

[0121] For the same test motor, the stator winding resistance, direct-axis and quadrature-axis inductance, equivalent flux linkage of permanent magnets, moment of inertia, main thermal resistance, and main heat capacity are selected as parameters to be corrected. During the test, the actual measured values ​​of voltage, current, speed, torque, and temperature are recorded, and the same control input and load input are sent to the digital twin model to obtain the corresponding simulation output.

[0122] For each measurement channel, the measurement dispersion of that channel is first determined by repeated sampling under stable operating conditions. Specifically, at least 100 stable data points are continuously obtained, their average value is calculated, the average of the squared differences between each data point and the average value is calculated, and finally the square root is taken to obtain the standard deviation of that channel.

[0123] When performing digital fingerprint parameter inversion, the difference between each simulated output value and the measured value is calculated, and this difference is divided by the standard deviation of the corresponding measurement channel. The result represents the multiple of the current simulation error relative to the measurement fluctuation of the sensing channel itself. Then, this multiple obtained from all channels at all sampling times is squared and summed.

[0124] The digital twin acceleration optimization engine changes the parameters to be calibrated and repeats the simulation until the sum of squared normalized errors is reduced to the minimum value within the current parameter range. The corresponding parameters to be calibrated at this time are then saved as the digital fingerprint parameters of the tested motor.

[0125] After adopting the above method, the sensing channel with higher accuracy and better repeatability has a smaller standard deviation, so the same prediction deviation will produce a larger normalization error, which has a higher constraint effect on parameter correction; the sensing channel with more noise has a larger standard deviation, and its occasional measurement fluctuations will not have an excessive impact on the digital fingerprint parameters.

[0126] After the digital fingerprint parameters are updated, it is also necessary to determine whether the current digital twin model meets the conditions for entering the virtual parameter pre-screening stage.

[0127] In specific judgment, the difference between all simulation outputs and measured values ​​is divided by the standard deviation of the corresponding sensing channel, the resulting values ​​are squared and averaged, and then the square root of the average value is taken to obtain the comprehensive normalized residual.

[0128] When the overall normalized residual is no greater than 1, it indicates that the average prediction bias of the digital twin model on all observations involved in the correction has entered the repeated measurement fluctuation range of the corresponding sensor, allowing the model to be used for virtual pre-screening of control parameters. When the overall normalized residual is greater than 1, the digital twin model continues to receive new physical test data to update the digital fingerprint parameters, and does not directly use the current virtual simulation results as candidate data for subsequent physical optimization.

[0129] After the model meets the pre-screening criteria, batch virtual tests are conducted on the control parameter combinations. First, parameter combinations exceeding the maximum motor current, allowable DC bus voltage, and winding temperature limits are eliminated. Then, the parameters are sorted from high to low according to their power supply efficiency. For parameter combinations with similar efficiencies, the efficiency dispersion under temperature and load variations is further compared.

[0130] For example, the power supply efficiencies of one parameter combination under five adjacent operating conditions are 81.8%, 82.1%, 81.9%, 82.0%, and 82.2%, respectively, while those of another parameter combination are 82.5%, 82.3%, 80.4%, 79.8%, and 82.6%. Although the second group has higher peak efficiency under some operating conditions, its fluctuation with changing operating conditions is significantly larger. By calculating the average efficiency and standard deviation of each parameter combination under multiple operating conditions, the parameter combination with higher average efficiency and smaller standard deviation is preferentially retained as the physical verification object, thus avoiding virtual pre-screening that only pursues the highest efficiency under a single operating condition.

[0131] By using digital fingerprint inversion, sensor error normalization, and simulation credibility determination, the digital twin model can be corrected based on the manufacturing deviations and actual test conditions of the specific tested motor, and the pre-screening of power supply control parameters can be introduced after the specified data credibility is achieved.

[0132] Further processing when calling historical parameters in the cloud: When the cloud-based collaborative analysis platform calls historical optimal parameters, it not only determines similar objects based on motor model or rated power, but also uses motor rated power, base speed, maximum speed, current ambient temperature, braking intensity, equivalent vehicle inertia, and digital fingerprint parameters for matching.

[0133] To avoid distortion in similarity calculations caused by different dimensions of parameters, each feature is first normalized according to the minimum and maximum values ​​in historical cloud data. Specifically, the current feature value is subtracted from the minimum value of the feature in the historical database, and then divided by the difference between the maximum and minimum values ​​of the feature, so that different physical quantities are all converted to the same numerical range of 0 to 1.

[0134] The weights of each feature are determined based on the sensitivity of the feature to changes in historical optimization results. For a given feature, historical data are arranged in ascending order of feature value, and the change in the optimal control parameter caused by the feature change is calculated among samples with similar main operating conditions. The more significant the change in the optimal parameter caused by the feature change, the higher its matching weight.

[0135] During matching, the difference between the current motor's normalized features and the historical motor's normalized features is calculated item by item, and then accumulated according to the corresponding weights. The smaller the accumulated result, the closer the two motors are in terms of features related to power supply optimization.

[0136] Whether historical parameters can be used as the initial strategy for the current motor is not determined solely by a fixed 80% similarity score. The cloud platform uses historical records for replay verification. Specifically, it simulates optimization using different similarity records as initial parameters within historical motor data that already has final optimization results, and calculates the minimum similarity score required to enter the neighborhood of the final optimal parameters within a specified number of test rounds. This similarity score is then used as the call threshold for the current database version.

[0137] As the number of historical samples increases, the above replay verification is re-executed and the call threshold is updated. Thus, the historical parameter call conditions are determined by reverse engineering from existing experimental results, rather than being manually specified.

[0138] After historical parameters are sent to the current test bench, they serve only as the initial search center. First, a verification test is conducted using a constrained load not exceeding 30% of the full regenerative braking power. Once it is confirmed that the current, DC bus voltage, and temperature rise do not exceed the limits, the test bench proceeds to the target load condition. If the feeder efficiency under constrained load conditions is lower than the current baseline control parameters, and the difference exceeds three times the standard deviation of efficiency measurement, then the historical parameters are discarded and the local baseline parameters are restored.

[0139] This allows us to reuse historical optimization data while avoiding the error in optimization direction caused by migration parameters being directly applied to high-load conditions due to individual differences in different motors.

[0140] Example 3: A test bench for the power supply characteristics of a motor for new energy vehicles, including a simulation chamber 100 for adjusting various ambient temperatures, and a fixed base 200 and a cast iron platform 300 respectively provided inside the simulation chamber 100, with the fixed base 200 and the cast iron platform 300 arranged flush. The fixed base 200 is used to support the motor 240 to be tested; A Y-axis moving drive component 210 is provided on one side of the fixed base 200. The Y-axis moving drive component 210 drives the Y-axis moving drive rod 212 to move the guide rail 211. An adjustment box 220 is provided in the middle of the upper surface of the guide rail 211. A Z-axis adjustment seat 230 is provided in the middle of the adjustment box 220.

[0141] The adjustment housing 220 is hollow and contains a drive assembly for adjusting the Z-axis adjustment seat 230 up and down. The upper surface of the cast iron platform 300 is provided with a movable groove 330 for placing the support base 310, and the upper surface of the support base 330 is connected to multiple sets of adjustable supports 320.

[0142] The temperature and humidity of the motor during operation are simulated by an adjustable physical environment simulation unit set inside the simulation chamber 100. Before the test, the Y-axis moving drive component 210 drives the Y-axis moving drive rod 212 to move the guide rail 211 to adjust the position of the motor along the Y-axis. The adjustment chamber 220 is equipped with a drive component for adjusting the Z-axis adjusting seat 230 up and down to adjust the motor along the Z-axis. The adjustment chamber 220 can slide along the guide rail 211 to move the motor along the X-axis, which facilitates the adjustment of the motor position. An adjustable bracket 320 is set on the cast iron platform 300 to fix the dynamometer. The dynamometer is connected to the motor under test. After the connection is completed, the test can be carried out.

[0143] Measurement Point Selection: For speed measurement, at least 10 speed points should be selected within the powertrain's operating speed range. The lowest speed point should ideally not exceed 10% of the highest operating speed, and the interval between adjacent speed points should not exceed 10% of the highest operating speed. The selection of measurement points should include necessary characteristic points: rated operating speed point, highest operating speed point, lowest speed point corresponding to continuous power, and other specially defined operating points. For torque measurement, under powered-off conditions, at least 10 torque points should generally be selected at each speed point. For high-speed operation, the number of torque points selected at each speed point can be appropriately reduced, but should not be less than 5. The selection of measurement points should include necessary characteristic points: points with continuous torque values, points with peak torque values, points on the continuous power curve, points on the peak power curve, and other specially defined operating points.

[0144] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of protection claimed by the present invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A test system for the power supply characteristics of a motor used in new energy vehicles, characterized in that: It includes a physical environment simulation unit, which is used to provide the tested motor with a composite environmental field including temperature, humidity and mechanical vibration; The integrated load simulation unit includes a mechanical inertia module, a road surface characteristic simulation module, and a braking load simulation module, which are used to apply a dynamic load coupled with vehicle inertia, road surface resistance characteristics, and braking intensity to the tested motor in the composite environment field. A multi-dimensional data sensing and fusion unit is used to synchronously and in real time collect the electrical, mechanical, thermal and vibration parameters of the tested motor under the composite environmental field and dynamic load. The intelligent power supply performance diagnosis and closed-loop optimization unit is communicatively connected to the multi-dimensional data sensing and fusion unit and the controller of the tested motor, and is used for: Based on the fused multi-dimensional data, the energy flow distribution during the power feeding process is analyzed in real time and key loss sources are located. Based on the characteristics of the key loss sources, optimized parameters for the regenerative braking control strategy of the tested motor are dynamically generated. The optimized parameters are sent to the controller of the motor under test, and the integrated load simulation unit is driven to perform a new round of testing to verify the optimization effect.

2. The test system for the power supply characteristics of a motor for new energy vehicles according to claim 1, characterized in that: The road surface characteristic simulation module is implemented through a programmable current harmonic injection unit to simulate the wheel slip ratio changes caused by different road surface adhesion coefficients. This change is converted into harmonic interference signals with specific frequencies and amplitudes to the motor current loop.

3. The test system for the power supply characteristics of a motor for new energy vehicles according to claim 2, characterized in that: The road surface characteristic simulation module includes a programmable harmonic current injector. It is configured to: calculate and generate corresponding current harmonic disturbance signals in real time based on the adhesion coefficient slip ratio curve corresponding to the input target road surface type, and superimpose them on the current reference signal of the motor controller to simulate the impact of the road condition on the electromagnetic torque and energy recovery stability of the motor.

4. The test system for the power supply characteristics of a motor for new energy vehicles according to claim 3, characterized in that: The multi-dimensional data perception and fusion unit includes: The electrical sensing subunit is used to collect motor terminal voltage, current, DC bus voltage, and high-frequency switching harmonics. The mechanical sensing subunit is used to collect torque, rotational speed, and multi-axial vibration acceleration. The thermal sensing subunit uses a combination of an infrared thermal imager and an embedded thermocouple to acquire the spatial temperature field distribution of the motor stator winding, permanent magnet and controller; The core of the data fusion is to align multi-source data based on timestamps and utilize the correlation between the vibration parameters and high-frequency current harmonics to diagnose specific electromagnetic losses caused by mechanical vibration.

5. The test system for the power supply characteristics of a motor for new energy vehicles according to claim 4, characterized in that: The power supply performance closed-loop optimization unit includes: The loss tracing model takes multi-dimensional data as input and outputs a quantitative proportion of total loss, which is decomposed into copper loss, iron loss, permanent magnet eddy current loss, mechanical loss and controller switching loss. The strategy optimization engine, based on the output of the loss tracing model, generates optimized parameters for adjusting the switching frequency or PWM modulation strategy if the iron loss ratio is high; and generates optimized parameters for adjusting the field weakening control strategy or current phase if the permanent magnet eddy current loss is abnormal. The strategy optimization engine integrates a reinforcement learning algorithm, which treats the motor controller parameters as actions of the intelligent agent and uses the reduction of the total power supply efficiency and specific loss components as rewards. Through continuous interaction with the environment, it autonomously seeks to obtain a more robust set of control strategy parameters.

6. The test system for the power supply characteristics of a motor for new energy vehicles according to claim 1, characterized in that: The system also includes a cloud-based collaborative analysis platform, which is used to collect historical test data and optimized parameter sets of motors from different locations and models; When the power supply performance closed-loop optimization unit starts optimization, it can call the optimal initial parameter set of similar motor models from the cloud collaborative analysis platform and accelerate the local optimization process based on transfer learning technology.

7. The test system for the power supply characteristics of a motor for new energy vehicles according to claim 1, characterized in that: The physical environment simulation unit includes: A temperature and humidity-controlled sealed chamber is used to enclose the tested motor and its controller. A multi-exciter vibration table, installed in the temperature and humidity controllable sealed chamber, is used to apply six-degree-of-freedom road spectrum vibration to the motor mounting base.

8. The test system for the power supply characteristics of a motor for new energy vehicles according to claim 7, the system further includes a digital twin acceleration optimization engine; the engine contains a high-fidelity simulation model of the test motor and its controller, which can perform a large number of simulation tests of control parameter combinations in parallel in virtual space before the intelligent diagnosis and closed-loop optimization unit for power supply performance starts physical iterative optimization, quickly pre-screen a subset of potential parameters, thereby significantly reducing the number of subsequent physical iterative optimizations and time; The simulation model of the digital twin acceleration optimization engine integrates digital fingerprint parameters that characterize the individual characteristics of the tested motor, learned from the historical data of the multi-source heterogeneous data synchronous sensing and fusion unit, so that the virtual simulation results are highly consistent with the physical measurement results.

9. A test method for the power supply characteristics of a motor in a new energy vehicle using the system described in any one of claims 1 to 9, characterized in that: Includes the following steps: S1: Environmental Coupling Settings: Set the parameters of the physical environment simulation unit and set the target working conditions in the integrated load simulation unit to form a composite test scenario; S2: Multidimensional data synchronous acquisition: Drive the system to operate and synchronously acquire all-dimensional parameters of the tested motor under the composite test scenario through the multidimensional data sensing and fusion unit; S3: Loss Source Tracing and Problem Localization: Based on fused data, the loss source tracing model is used to quantitatively analyze the composition and proportion of various losses in real time, and to locate the dominant loss source; S4: Closed-loop optimization of control strategy: Based on the characteristics of the dominant loss source, the strategy optimization engine dynamically generates control parameter adjustment suggestions and sends them to the motor controller; S5: Optimization effect verification and iteration: Under the premise of keeping the environment and load conditions unchanged, retest using new control parameters, compare the performance indicators before and after optimization, and repeat steps S3-S5 if the target is not met until the optimal result is achieved.

10. The test method for the power supply characteristics of a motor for new energy vehicles according to claim 9, characterized in that: In step S3, the specific method for locating the dominant loss source includes: analyzing the coherence function of the vibration acceleration amplitude at a specific frequency and the corresponding frequency harmonic component in the current; if the coherence coefficient is higher than the threshold, it is determined that there is additional iron loss caused by mechanical vibration.

11. The test method for the power supply characteristics of a motor for a new energy vehicle according to claim 9, characterized in that: In step S1, when the target operating condition has similar historical optimization records in the cloud knowledge base, the system automatically recommends using the historical optimal parameters adjusted by transfer learning as the initial strategy, thereby skipping unnecessary benchmark degradation tests and directly entering the fine optimization stage.

Images