Detection method of driving motor for new energy automobile

Abstract

The invention relates to a detection method of a driving motor for a new energy automobile in the technical field of detection of the driving motor for the new energy automobile. The detection method comprises the following steps: a dynamic excitation step: applying composite dynamic excitation to a detected driving motor; and a follow-up synchronous acquisition step: under the action of the composite dynamic excitation, controlling a detection execution mechanism to move relative to a shell of the detected driving motor according to a preset track, and synchronously acquiring a vibration signal, an acoustic signal and an electrical signal of the detected driving motor. According to the method, through sequential composite excitation and benefit from high efficiency of follow-up synchronous acquisition and rapid reasoning of a fusion diagnosis algorithm, the whole process from excitation to diagnosis report output can be strictly controlled within 90 seconds, and the problem that traditional subentry detection is low in efficiency is fundamentally solved.

Description

Technical Field

[0001] This invention relates to the field of drive motor technology for new energy vehicles, and in particular to a testing method for drive motors used in new energy vehicles. Background Technology

[0002] As a core power component of new energy electric vehicles, the final quality inspection of the drive motor before it leaves the factory is crucial, directly affecting the long-term reliability and safety of the product. With the industry's ever-increasing demands for production efficiency and zero defect rates, traditional off-line inspection methods for drive motors are no longer sufficient to meet the needs of modern intelligent manufacturing. Current mainstream production line inspection solutions mainly face the following technical bottlenecks: 1. Difficulty in balancing testing efficiency and production line cycle time: Existing production lines typically employ multiple independent workstations connected in series or parallel to comprehensively evaluate drive motor performance, performing electrical safety, mechanical performance, and functional tests separately. For example, while solutions offered by mainstream suppliers such as Marposs can perform comprehensive functional testing, they are mostly conducted at separate workstations and for specific items. This discrete testing process results in excessively long total testing time for a single drive motor, severely restricting production line cycle time and making it difficult to meet the demands of high-efficiency production.

[0003] 2. Static / fixed-point testing is disconnected from actual operating conditions: Existing testing methods are mostly conducted under simple steady-state conditions such as when the drive motor is stationary or under uniform speed and no load (e.g., a method and device for early warning of faults in a permanent magnet synchronous motor disclosed in Chinese invention patent application No. 202311043304.1). For example, conventional vibration and noise (NVH) testing or back EMF analysis is usually performed at a fixed speed point. However, the drive motor will undergo complex dynamic processes (such as acceleration, loading, etc.) in actual operation. Testing under static or simple operating conditions cannot effectively stimulate certain potential faults that only appear under dynamic stress, such as early damage to bearings under variable load, resonance of the rotor when accelerating through the critical speed, or defects in the dynamic response of the controller, resulting in incomplete detection coverage and potential quality and safety hazards.

[0004] 3. Insufficient Single-Dimensional Diagnosis and Complex Fault Identification Capabilities: While existing intelligent diagnostic research has incorporated artificial intelligence algorithms, it largely focuses on in-depth analysis of single data sources or laboratory scenarios. For example, some existing technologies disclose fault diagnosis methods based on multimodal sensors and hybrid deep networks (such as CNN+LSTM), but their core lies in high-precision, non-real-time in-depth analysis of offline collected data, failing to address the problem of high-speed synchronous acquisition and real-time fusion of multi-source signals on the production line (e.g., a method and device for fault diagnosis of drive motors based on artificial intelligence technology disclosed in Chinese invention patent application number 202510897151.X). Other technical solutions attempt to conduct tests in dynamic environments (such as combined temperature, humidity, and vibration environments). These solutions aim to verify the reliability of products throughout their entire lifecycle, with test cycles lasting several hours or even longer, making them completely unsuitable for production line testing cycles measured in seconds (e.g., a test method and system for a motor driver of a new energy vehicle's on-board motor disclosed in Chinese invention patent application number 202511758065.7). Therefore, existing technologies lack a production line-level solution that can simultaneously complete the excitation, acquisition, and integrated intelligent diagnosis of multiple physical quantities in a very short time.

[0005] In summary, the key technical challenge in this field is how to achieve automated, synchronized, and intelligent comprehensive testing and accurate fault diagnosis of the mechanical, electrical, and acoustic properties of drive motors within a strict time constraint (e.g., 60-90 seconds) during the production line assembly process, so as to simultaneously meet the triple requirements of high efficiency, high coverage, and high precision. Summary of the Invention

[0006] To overcome the shortcomings of the prior art, the present invention discloses a testing method for drive motors used in new energy vehicles.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A testing method for drive motors used in new energy vehicles includes the following steps: Dynamic excitation steps: Apply composite dynamic excitation to the drive motor under test; Follow-up synchronous acquisition step: Under the action of the composite dynamic excitation, the detection actuator is controlled to move relative to the housing of the drive motor under test along a preset trajectory, and the vibration signal, acoustic signal and electrical signal of the drive motor under test are acquired synchronously. Fusion diagnostic steps: The vibration signal, the acoustic signal and the electrical signal are fused to extract fusion features to obtain a fusion feature vector, and the fusion feature vector is input into a pre-trained fault classification model to output the diagnostic results of the tested drive motor.

[0008] Furthermore, the composite dynamic excitation sequentially includes the following stages: During the no-load frequency sweep phase, the speed of the drive motor under test is linearly increased from zero to the first preset speed at a preset frequency sweep rate. During the steady-state load phase, a steady-state load is applied at the first preset speed and maintained within the first time window; During the dynamic load phase, a periodic fluctuating load is superimposed on the steady-state load and continues for a second time window. During the reverse-drive power generation phase, the rotor of the tested drive motor is driven by a constant-speed power source, and the back electromotive force signal of the tested drive motor when it is in an open-circuit state is collected.

[0009] Preferably, the preset sweep frequency rate is not less than 300 rpm / s, and the sweep frequency range of the no-load sweep frequency stage covers the first critical speed of the spindle of the drive motor under test, so as to effectively excite mechanical resonance; the first critical speed is determined by finite element analysis or experimental modal analysis based on the structural parameters of the drive motor under test.

[0010] Furthermore, in the follow-up synchronous acquisition step, a unified high-precision clock source is configured for all data channels that acquire the vibration signal, the acoustic signal, and the electrical signal; the real-time pose data of the detection actuator is time-domain aligned with the vibration signal, the acoustic signal, and the electrical signal using the same timestamp to ensure strict synchronization of the multi-source signals on the time axis.

[0011] Furthermore, the fusion feature extraction includes: From the vibration signal, the characteristic frequency amplitudes related to bearing failure are extracted as vibration signal features; From the acoustic signal, the sound pressure level features within a specific frequency band are extracted as acoustic signal features; The effective value of the quadrature-axis current ripple and the total harmonic distortion rate of the back electromotive force are extracted from the electrical signal as electrical signal characteristics.

[0012] Furthermore, the fusion diagnostic step also includes feature dimensionality reduction and enhancement: the vibration signal features, the acoustic signal features, and the electrical signal features are normalized and spliced ​​to form an initial feature vector, and then the initial feature vector is dimensionality reduced using a sparse autoencoder to obtain a more compact and more representative fusion feature vector.

[0013] Furthermore, the fault classification model is a multi-branch neural network model based on an attention mechanism. The multi-branch neural network model includes a first branch for processing the vibration signal features and the acoustic signal features, and a second branch for processing the electrical signal features. The attention mechanism is used to perform weighted fusion of the output features of the first branch and the second branch.

[0014] Furthermore, the detection method also includes a model optimization step: when the output confidence of the fault classification model for the current diagnostic result is lower than a preset threshold, the currently collected vibration signal, acoustic signal and electrical signal are stored as sample data in the updated sample library, and the fault classification model is incrementally trained using the sample data in the updated sample library.

[0015] Furthermore, the preset trajectory is generated offline based on the three-dimensional model of the test drive motor housing to ensure that the sensor on the detection actuator maintains a constant optimal measurement distance and angle with the housing surface during the movement.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention integrates traditionally separate, time-based testing items into a sequential, continuous, automated process through composite dynamic excitation. Thanks to the high efficiency of synchronous data acquisition and the rapid inference of the fusion diagnostic algorithm, the entire process from excitation to outputting a diagnostic report can be strictly controlled within 90 seconds (up to 85 seconds in a preferred embodiment). This fundamentally solves the pain points of low efficiency and long production line time associated with traditional separate testing, achieving an order-of-magnitude improvement in testing efficiency without reducing the number of testing items.

[0017] 2. The composite dynamic excitation of this invention can simulate the dynamic stress experienced by the drive motor during actual operation, effectively stimulating potential faults that are difficult to detect under static or simple operating conditions. For example, high-speed frequency sweep can expose rotor dynamic imbalance problems, and dynamic fluctuating loads can stimulate loosening of mechanical connections or early bearing damage. Simultaneously, the laser vibrometer and microphone array provide non-contact measurement, avoiding installation errors and acquiring global information from multiple key locations on the drive motor housing, which is more comprehensive than fixed-point measurements. Therefore, defects can be detected earlier and more comprehensively, intercepting faults that might be missed by traditional methods before leaving the factory, significantly improving product quality and long-term reliability.

[0018] 3. This invention ensures the inherent correlation of multi-source signals in the time domain through synchronous acquisition, laying a solid foundation for subsequent fusion analysis. Through a multi-branch neural network model based on an attention mechanism, the system can automatically learn the importance weights of various features under different fault modes, achieving deep information fusion and complementarity. This makes the diagnostic results not only a "qualified / unqualified" judgment, but also outputs specific fault types (such as "bearing outer ring wear") and high confidence levels, and even provides judgment criteria through attention weights, achieving accurate and interpretable intelligent diagnosis and solving the problem of inaccurate identification of complex faults by traditional threshold methods or single-model methods.

[0019] 4. The model in the diagnostic system of this invention has online optimization capabilities. When the model's diagnostic confidence for certain "marginal samples" is low, the system automatically stores these "marginal samples" in a sample library, which can then be used for incremental training in conjunction with manual review results. This allows the diagnostic system to continuously self-optimize as production data accumulates, new drive motor models are introduced, or new fault modes emerge, resulting in a continuous improvement in diagnostic accuracy. This endows the production line with long-term adaptability and vitality, solving the problem of technical rigidity where fixed algorithm models struggle to adapt to product updates and process changes. Attached Figure Description

[0020] Figure 1 The overall flowchart of the detection method provided in the embodiments of the present invention is shown.

[0021] Figure 2 This is a block diagram illustrating the principle of synchronous acquisition and time-domain alignment of multi-source signals in an embodiment of the present invention.

[0022] Figure 3 This is a schematic diagram of the structure of a multi-branch neural network model based on the attention mechanism in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining this invention and are not intended to limit it. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0024] Please refer to the instruction manual appendix. Figure 1-3 The present invention provides the following technical solutions: This invention provides a testing method for drive motors used in new energy vehicles, aiming to meet the needs of high-speed production lines for rapid, comprehensive, and intelligent quality testing of drive motors. The following detailed explanation of each step of this method, using a preferred embodiment, will illustrate this method in detail.

[0025] In a typical implementation scenario, the system implementing this method may include an excitation control unit, a multi-axis motion control unit (i.e., the control system for the detection actuator), a multi-source signal synchronous acquisition unit, a central processing unit, and a human-machine interface unit. The excitation control unit drives a high-dynamic-performance servo motor as a load, coaxially connected to the tested drive motor via a coupling. The multi-axis motion control unit controls a six-degree-of-freedom industrial robot. The end flange of the six-degree-of-freedom industrial robot integrates vibration sensors, acoustic sensors, and electrical sensors, collectively constituting the sensing part of the detection actuator. The vibration sensor can be a laser vibrometer, the acoustic sensor can be a miniature microphone array, and the electrical sensor can be a differential voltage probe. The central processing unit is equipped with an industrial switch and a high-performance industrial computer with IEEE 1588 (PTP) precision clock protocol functionality. The central processing unit is responsible for the synchronization coordination, data processing, and intelligent diagnostics of the entire process.

[0026] This method includes the following steps: S1. Dynamic Excitation Step: Apply composite dynamic excitation to the drive motor under test. The composite dynamic excitation is not a single operating condition, but a program executed according to a strict timing sequence, with the aim of efficiently stimulating potential mechanical and electrical fault characteristics.

[0027] Furthermore, step S1 includes the following stages in sequence: S1.1, No-load frequency sweep stage: The excitation control unit controls the load servo motor to drive the rotor of the test drive motor from rest, linearly accelerating at a preset frequency sweep rate (set to 500 rpm / s in this embodiment) until it reaches the first preset speed (e.g., 120% of the rated speed, i.e., 3600 rpm). The high-speed frequency sweep process aims to cover the first critical speed of the spindle of the test drive motor. If dynamic imbalance or structural resonance exists, it will be significantly excited during the no-load frequency sweep stage. The preset frequency sweep rate can be adjusted according to the inertia and test intensity requirements of different types of drive motors, but it should generally not be lower than 300 rpm / s to achieve effective excitation.

[0028] The setting of the preset sweep rate and the first preset speed ensures that the excitation energy effectively covers the first two natural frequencies (critical speeds) of the rotor-bearing system of the tested drive motor, thereby efficiently exciting any potential dynamic imbalances and structural resonance faults. As a non-limiting example, for a permanent magnet synchronous drive motor with a rated speed of 3000 rpm, the first critical speed is typically in the range of 1800-2200 rpm. Therefore, setting the sweep endpoint (i.e., the first preset speed) to 3600 rpm and the sweep rate to 500 rpm / s ensures that the excitation scan of the critical frequency band is completed within approximately 4-5 seconds.

[0029] S1.2 Steady-state load stage: Once the rotational speed reaches the first preset speed and stabilizes, the excitation control unit immediately applies a steady-state load. For example, 50% of the rated torque of the tested drive motor is applied as the steady-state load, and maintained for a first time window (e.g., 10-30 seconds). The steady-state load stage is used to simulate the smooth operation of the drive motor under load, to observe winding heating, bearing temperature rise, and operational stability under constant stress.

[0030] S1.3 Dynamic Load Phase: While maintaining the steady-state load, a periodic fluctuating load is superimposed. For example, a sinusoidal torque disturbance with an amplitude of 10% of the rated torque and a frequency of 5Hz is superimposed as the periodic fluctuating load and sustained for a second time window (e.g., 2-15 seconds). This dynamic load phase is used to simulate vehicle acceleration, deceleration, or road condition fluctuations to expose dynamic problems such as loose connections, abnormal gaps, or controller response lag.

[0031] The combination of the steady-state load phase and the dynamic load phase is used to simulate the real-world operating conditions of the drive motor as it transitions from smooth operation to being subjected to periodic disturbances. The frequency of the periodic fluctuating load (e.g., 5Hz) is selected to match the typical fluctuation frequency of common road surface excitations or loaded machinery, in order to detect abnormal responses of the system under dynamic stress.

[0032] S1.4, Back-Drag Generation Stage: The drive power supply to the tested drive motor is disconnected, leaving the windings in an open-circuit state. The excitation control unit switches to speed control mode, controlling the load servo motor to precisely drag the rotor of the tested drive motor at a constant, low speed (e.g., 1000 rpm). During the back-draft generation stage, the voltage at the three-phase output terminals of the tested drive motor is directly measured using the differential voltage probe to obtain a pure back electromotive force (EMF) signal. This back EMF signal is used to analyze electrical performance such as the symmetry of the permanent magnet magnetic field and the consistency of the stator windings.

[0033] S2, Follow-up Synchronous Acquisition Step: The follow-up synchronous acquisition step is performed simultaneously with the dynamic excitation step. Trajectory Planning: Before the testing begins, the motion trajectory of the six-degree-of-freedom industrial robot (i.e., the preset trajectory) is pre-planned based on the three-dimensional model of the drive motor under test. This trajectory planning ensures that when the laser vibrometer and the micro-microphone array move with the testing actuator to multiple pre-set measurement points on the drive motor housing (such as directly above the front and rear bearing seats, or in the middle of the housing), the sensing axes of the laser vibrometer and the micro-microphone array are perpendicular to the surface of the drive motor housing at those measurement points, and that the laser vibrometer and the micro-microphone array maintain a constant optimal measurement distance (e.g., 50mm) from the housing surface. The preset trajectory is generated offline, can be stored, and can be used for testing all drive motors of the same model.

[0034] Synchronous data acquisition is achieved by the central processing unit (CPU) providing a unified high-precision clock source for all data acquisition cards, robot controllers, and excitation control units via the IEEE 1588 protocol. When the composite dynamic excitation begins, the CPU issues a global trigger signal. The six-degree-of-freedom (6DOF) industrial robot moves along the planned trajectory and continuously sends its real-time pose data (including 3D coordinates and attitude) to the data acquisition system. This real-time pose data is timestamped with a timestamp identical to the sensor data. The laser vibrometer, the miniature microphone array, and the differential voltage probe synchronously acquire vibration, acoustic, and electrical signals. The sampling clock deviation of all signal channels is controlled within 1 microsecond, ensuring that the vibration, acoustic, and electrical signals are strictly aligned in the time domain with the real-time pose data of the six-DOF industrial robot, forming a spatiotemporally correlated "data cube."

[0035] S3. Fusion Diagnostic Steps: The fusion diagnostic steps are the core of intelligent diagnosis, involving signal processing, feature engineering, and artificial intelligence model decision-making.

[0036] Signal preprocessing and feature extraction: First, the synchronously acquired raw signals are preprocessed. The vibration signal is bandwidth filtered (e.g., 10Hz-10kHz) to remove irrelevant frequency band interference; beamforming and spectral subtraction algorithms are applied to the acoustic signal to enhance the sound source from the direction of the driven motor under test and suppress environmental noise; Park transform is performed on the electrical signal to transform it from a three-phase stationary coordinate system to a synchronous rotating coordinate system.

[0037] The synchronous rotating coordinate system includes a direct axis (d-axis) and a quadrature axis (q-axis). The d-axis direction is aligned with the direction of the rotor permanent magnet magnetic field and is used to control the excitation component. The q-axis direction leads the d-axis by 90 electrical degrees and is used to control the torque component. The "quadrature axis current ripple" refers to the fluctuation component in the q-axis current signal, and the effective value of the quadrature axis current ripple characterizes the pulsation amplitude of the motor output torque.

[0038] Subsequently, feature extraction is performed. From the preprocessed vibration signal, through Fast Fourier Transform (FFT) and envelope demodulation analysis, the theoretical values ​​of the bearing's inner ring passing frequency (BPFI), outer ring passing frequency (BPFO), roller passing frequency (BSF), and cage failure frequency (FTF) are calculated according to the bearing's model parameters (such as the number of rollers, pitch diameter, and contact angle) in the tested drive motor. Then, the actual peak values ​​near the theoretical frequencies are identified in the signal spectrum, and the amplitude of these actual peak values ​​is precisely extracted as vibration signal features.

[0039] For example, bearing fault characteristic frequencies (such as inner ring passing frequency BPFI, outer ring passing frequency BPFO, etc.) are calculated based on bearing geometric parameters (number of rollers Nb, pitch circle diameter Dm, roller diameter d, contact angle α) and current rotational speed fr, using formulas known in the field of mechanical fault diagnosis. For example, outer ring passing frequency BPFO = (Nb / 2)fr(1-(d / Dm)). cosα).

[0040] From the preprocessed acoustic signal, the sound pressure level characteristics (A-weighted) within a specific frequency band sensitive to mechanical noise, from 500Hz to 2000Hz, are calculated as acoustic signal characteristics.

[0041] From the transformed electrical signal, the effective value of the quadrature-axis current ripple (reflecting torque pulsation) is calculated, and FFT analysis is performed on the back electromotive force signal to calculate the total harmonic distortion (THD) of the back electromotive force signal. The effective value of the quadrature-axis current ripple and the total harmonic distortion are used as electrical signal characteristics.

[0042] Feature fusion and dimensionality reduction: The extracted vibration signal features, acoustic signal features, and electrical signal features are normalized to eliminate the influence of dimensions. The normalized features are then concatenated into an initial feature vector. This initial feature vector typically has a high dimensionality. Subsequently, a pre-trained sparse autoencoder is used to reduce the dimensionality of the initial feature vector. This sparse autoencoder learns the intrinsic structure of the data, retaining the most critical information while reducing dimensionality, and outputting a more compact and representative fused feature vector.

[0043] The sparse autoencoder requires pre-training before deployment. During training, initial feature vectors extracted from a large amount of historical qualified drive motor detection data and various typical faulty drive motors are used as the training set. The training objective is to minimize the error between the input feature vector and the decoder's reconstructed output. Simultaneously, a sparsity penalty term (such as KL divergence) is added to the loss function, causing most neurons in the hidden layer to be in an inhibited state most of the time, thereby learning the inherent sparse representation of the data. Through this training, the sparse autoencoder can learn the most discriminative low-dimensional projection of the drive motor's healthy and faulty states in the feature space.

[0044] Intelligent model diagnosis: The fused feature vector is input into a pre-trained fault classification model. The fault classification model employs a multi-branch neural network model based on an attention mechanism.

[0045] The pre-training process of the fault classification model includes: collecting drive motor samples containing normal states and various known fault states; acquiring multi-source signals of the drive motor samples under the composite dynamic excitation; generating sample feature vectors through the fusion feature extraction step; and labeling them with corresponding category labels. Using the sample feature vectors as input and the category labels as output, the parameters of the multi-branch neural network model are iteratively optimized using a gradient descent algorithm (such as the Adam optimizer) with the cross-entropy loss function until the classification accuracy of the model on the independent validation set tends to stabilize.

[0046] Model Structure: The multi-branch neural network model comprises two main branches. The first branch (e.g., a multilayer perceptron MLP) is dedicated to processing the vibration signal features and the acoustic signal features, i.e., the feature subset related to the mechanical state; the second branch (another MLP) is dedicated to processing the electrical signal features.

[0047] Attention Mechanism: After the first branch and the second branch output their respective high-level feature representations, an attention mechanism layer is introduced. This layer automatically learns and calculates the importance (i.e., weights α1 and α2) of the high-level feature representations of the first and second branches to the current diagnostic task. For example, if the current symptoms are more strongly associated with electrical faults, the attention mechanism layer will assign a higher weight to the high-level feature representation of the second branch.

[0048] The attention mechanism layer receives a first feature vector F1 output from the first branch and a second feature vector F2 output from the second branch. The attention mechanism layer calculates an attention score vector using a small feedforward neural network. This attention score vector reflects the importance of each dimension of the features in F1 and F2 to the final diagnostic task under the input of the current fused feature vector. Subsequently, the attention score is normalized to weight coefficients α1 and α2 (α1 + α2 = 1) using the Softmax function, and finally, weighted fusion is performed: fused feature = α1F1 + α2F2.

[0049] Weighted fusion and classification: The first feature vector F1 and the second feature vector F2 are weighted and summed (α1F1 + α2F1) to achieve adaptive weighted fusion. The fused features are then passed through a fully connected classification layer, ultimately outputting a probability distribution vector, i.e., the diagnostic result. The diagnostic result indicates the probability that the tested drive motor belongs to a predefined category such as "qualified", "bearing failure", "rotor imbalance", or "winding asymmetry", and provides the overall confidence score (the highest category probability value output by the model).

[0050] Model optimization mechanism: To maintain the long-term effectiveness of the model, the system includes a model optimization step. A preset threshold is set (e.g., 0.7-0.9). When the confidence level of the fault classification model's output for a certain diagnosis falls below the preset threshold, the system automatically stores all raw signals collected for that diagnosis, extracted features, and labels that have been manually verified as sample data into the updated sample library. Weekly or monthly, the system uses the sample data in the updated sample library to incrementally train the original model, thereby achieving continuous model evolution.

[0051] The incremental training is specifically carried out in one of the following ways: (1) mixing the new data in the updated sample library with some historical data and fine-tuning the model with all parameters; or (2) using the elastic weight consolidation algorithm to minimize the loss on the new data while imposing constraints on the parameters that are important to the model on the old task to prevent catastrophic forgetting. After training is completed, the model performance is evaluated using an independent validation set. After confirming the performance improvement, the new model is deployed online to replace the old model.

[0052] The entire process, from the initial excitation to the output of the diagnostic report, can be strictly controlled within 90 seconds through the tight integration and automated control of the above steps, with pure data processing and diagnostic calculation time being less than 10 seconds. The diagnostic report, along with excitation parameters, key signal segments, characteristic values, and other data, are all bound to the unique serial number of the tested drive motor to form a complete digital file, which is then uploaded to the factory's MES system to achieve full quality traceability.

[0053] To verify the effectiveness of this method, a certain type of drive motor (rated speed 3000 rpm, rated torque 200 Nm) was tested using the above parameters. The model was trained using a historical dataset containing 1000 qualified drive motors and 200 drive motors with known faults (bearings, rotors, windings). In continuous testing of 500 drive motors on an actual production line, the average detection time of this method was 85 seconds, the accuracy rate for identifying known fault types reached 99.2%, and the false positive rate for qualified drive motors was less than 0.5%. All diagnostic results generated structured reports and were successfully uploaded to the MES (Manufacturing Execution System).

[0054] The above embodiments are merely illustrative of the technical solutions of the present invention and are not intended to limit the scope of protection. Those skilled in the art should understand that various changes and modifications can be made to the technical parameters (such as sweep rate, load percentage, and fluctuation frequency), specific algorithm selection (such as using other types of autoencoders or neural network variants), or hardware implementation (such as using other multi-axis motion platforms as detection actuators) in the above embodiments without departing from the principles of the present invention. Any other algorithm or model variation capable of achieving multi-source signal feature fusion and fault classification falls within the scope of the present invention.

Claims

1. A method for testing drive motors used in new energy vehicles, characterized in that, Includes the following steps: Dynamic excitation steps: Apply composite dynamic excitation to the drive motor under test; Follow-up synchronous acquisition step: Under the action of the composite dynamic excitation, the detection actuator is controlled to move relative to the housing of the drive motor under test along a preset trajectory, and the vibration signal, acoustic signal and electrical signal of the drive motor under test are acquired synchronously. Fusion diagnostic steps: The vibration signal, the acoustic signal and the electrical signal are fused to extract fusion features to obtain a fusion feature vector, and the fusion feature vector is input into a pre-trained fault classification model to output the diagnostic results of the tested drive motor.

2. The testing method for drive motors for new energy vehicles according to claim 1, characterized in that: The composite dynamic excitation includes the following stages in sequence: During the no-load frequency sweep phase, the speed of the drive motor under test is linearly increased from zero to the first preset speed at a preset frequency sweep rate. During the steady-state load phase, a steady-state load is applied at the first preset speed and maintained within the first time window; During the dynamic load phase, a periodic fluctuating load is superimposed on the steady-state load and continues for a second time window. During the reverse-drive power generation phase, the rotor of the tested drive motor is driven by a constant-speed power source, and the back electromotive force signal of the tested drive motor when it is in an open-circuit state is collected.

3. The testing method for drive motors for new energy vehicles according to claim 2, characterized in that: The preset sweep frequency rate is not less than 300 rpm / s, and the sweep frequency range of the no-load sweep frequency stage covers the first critical speed of the spindle of the drive motor under test. The first critical speed is determined by finite element analysis or experimental modal analysis based on the structural parameters of the drive motor under test.

4. The testing method for drive motors for new energy vehicles according to claim 1, characterized in that: In the follow-up synchronous acquisition step, a unified high-precision clock source is configured for all data channels that acquire the vibration signal, the acoustic signal, and the electrical signal; the real-time pose data of the detection actuator is time-domain aligned with the vibration signal, the acoustic signal, and the electrical signal using the same timestamp.

5. The testing method for drive motors for new energy vehicles according to claim 1, characterized in that: The fusion feature extraction includes: From the vibration signal, the characteristic frequency amplitudes related to bearing failure are extracted as vibration signal features; From the acoustic signal, the sound pressure level features within a specific frequency band are extracted as acoustic signal features; The effective value of the quadrature-axis current ripple and the total harmonic distortion rate of the back electromotive force are extracted from the electrical signal as electrical signal characteristics.

6. The testing method for drive motors for new energy vehicles according to claim 5, characterized in that: The fusion diagnostic step further includes: normalizing and splicing the vibration signal features, acoustic signal features and electrical signal features to form an initial feature vector, and then using a sparse autoencoder to perform dimensionality reduction processing on the initial feature vector to obtain the fused feature vector.

7. The testing method for drive motors for new energy vehicles according to claim 5 or 6, characterized in that: The fault classification model is a multi-branch neural network model based on an attention mechanism. The multi-branch neural network model includes a first branch for processing the vibration signal features and the acoustic signal features, and a second branch for processing the electrical signal features. The attention mechanism is used to perform weighted fusion of the output features of the first branch and the second branch.

8. The testing method for drive motors for new energy vehicles according to claim 7, characterized in that: It also includes a model optimization step: when the confidence level of the fault classification model's output of the current diagnostic result is lower than a preset threshold, the currently collected vibration signal, acoustic signal and electrical signal are stored as sample data in the updated sample library, and the fault classification model is incrementally trained using the sample data in the updated sample library.

9. The testing method for drive motors for new energy vehicles according to claim 1, characterized in that: The preset trajectory is generated offline based on the three-dimensional model of the housing of the drive motor under test.

Images