Electric vehicle component performance testing method based on multi-source sensing data

Abstract

This invention relates to the field of component performance testing technology, specifically a method for testing the performance of electric vehicle components based on multi-source sensor data. The method includes: collecting time-series data of structural vibration, temperature distribution field data, and electromagnetic interference spectrum data of the component under test under preset test conditions. An improved multi-modal feature fusion algorithm is employed, adjusting the fusion weights based on the physical coupling relationship between vibration signals and temperature fields to generate corresponding fused performance features. These features are input into a cloud-based performance evaluation model to obtain a comprehensive evaluation result and potential failure mode prediction. An enhanced test excitation sequence is generated, and response data is collected to update the sensor dataset and conduct iterative testing. This method optimizes the multi-source data fusion processing logic, improves the dynamic testing process, comprehensively covers the operating state of components under multi-physical field coupling, and enhances the comprehensiveness and adaptability of component performance testing.

Description

Technical Field

[0001] This invention relates to the field of component performance testing technology, and in particular to a method for testing the performance of electric vehicle components based on multi-source sensor data. Background Technology

[0002] In routine performance testing of electric vehicle components, a single-dimensional sensor data collection and analysis approach is often used. Existing multimodal data fusion processing methods mostly employ uniform and fixed weight allocation rules for calculations, without incorporating the physical correlation between different sensor signals to adjust the mechanism. Routine component performance testing procedures maintain a uniform standard, with pre-set test stimuli and the overall testing process executed only once according to the predetermined procedure, maintaining a static operation mode for data collection and analysis.

[0003] Single-dimensional data detection modes cannot adapt to the working environment of multiple physical fields superimposed during actual component operation. Fixed-weight feature fusion modes cannot adapt to the interconnected changes in multi-source data under different operating conditions, making the extracted performance features unable to match the actual operating state. Standardized fixed testing modes lack dynamic adjustment capabilities, cannot adjust the testing direction based on potential operational anomalies of components, and limited data samples are insufficient to comprehensively reflect the overall performance status of components. To address these technical shortcomings, it is necessary to optimize the weight allocation logic of multimodal feature fusion, dynamically adjust the testing content based on operational state prediction, and improve the full-dimensional performance testing process of components through data iterative updates. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a performance testing method for electric vehicle components based on multi-source sensor data.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for testing the performance of electric vehicle components based on multi-source sensor data, comprising: Acquire a set of multi-source sensor data of the electric vehicle component under test under preset test conditions. The set of multi-source sensor data includes structural vibration time series data, temperature distribution field data and electromagnetic interference spectrum data. An improved multimodal feature fusion algorithm is used to collaboratively process the multi-source sensor data set to generate fused performance characteristics of the electric vehicle components under test. The fused performance characteristics include structural dynamic stress distribution, thermal load coupling factor, and electromagnetic compatibility index. The improved multimodal feature fusion algorithm adjusts the weight allocation mechanism of feature fusion based on the physical coupling relationship between vibration signal and temperature field. The fused performance characteristics are input into the component performance evaluation model deployed in the cloud to obtain the comprehensive performance evaluation results and potential failure mode predictions of the electric vehicle component under test. Based on the predicted potential failure modes, a targeted enhanced test stimulus sequence is generated by matching from a preset test case library; The enhanced test stimulus sequence is executed, and corresponding response data is collected to update the multi-source sensor data set for iterative testing.

[0006] As a further aspect of the present invention, the step of using an improved multimodal feature fusion algorithm to collaboratively process the multi-source sensor data set to generate fused performance features of the electric vehicle component under test includes: The vibration time series data of the structure are subjected to joint time-frequency domain analysis to extract the vibration dominant frequency, harmonic components and damping characteristic parameters, and to form a vibration feature vector. Heat source identification and heat flow path analysis are performed on the temperature distribution field data to extract the location of the highest temperature point, temperature gradient and thermal equilibrium time constant, and to form a thermal feature vector. The electromagnetic interference spectrum data is subjected to frequency band energy and noise feature analysis to extract the characteristic frequency amplitude, broadband noise floor and pulse interference density, and to form an electromagnetic feature vector. The improved multimodal feature fusion algorithm is invoked to fuse the vibration feature vector, thermal feature vector, and electromagnetic feature vector; wherein... The improved multimodal feature fusion algorithm first dynamically calculates the coupling weight between vibration and thermal features based on real-time acquired temperature gradient data, then adaptively adjusts the contribution of vibration features in the fusion process based on the pulse interference density in the electromagnetic interference spectrum, and finally generates a unified fusion feature representation through a multi-layer feature cross-aggregation network. Quantitative indicators that simultaneously reflect mechanical, thermal, and electrical states are extracted from the unified fusion feature representation and used as the dynamic stress distribution, thermal load coupling factor, and electromagnetic compatibility index of the structure, respectively.

[0007] As a further aspect of the present invention, the time-frequency domain joint analysis of the structural vibration time series data is performed to extract the dominant vibration frequency, harmonic components, and damping characteristic parameters, thereby constructing a vibration feature vector, including: Empirical mode decomposition (EMD) is performed on the vibration time series data of the structure to obtain a series of intrinsic mode function components. Calculate the instantaneous frequency and instantaneous amplitude of each intrinsic mode function component, and plot the Hilbert spectrum; Identify the frequency bands with concentrated energy from the Hilbert spectrum, take the frequency point with the highest energy as the dominant vibration frequency, and take the energy characteristics at its integer multiples as harmonic components; The Hilbert transform is performed on the components corresponding to the dominant vibration frequency in the intrinsic mode function components, the phase angle of its analytic signal is calculated, and the damping ratio of the system is obtained by fitting the phase angle change rate as a damping characteristic parameter. The vibration characteristic vector is formed by arranging the dominant vibration frequency, the amplitude of each harmonic component, and the damping ratio in a preset order.

[0008] As a further aspect of the present invention, the improved multimodal feature fusion algorithm adjusts the weight allocation mechanism for feature fusion based on the physical coupling relationship between vibration signals and temperature fields. Its working principle includes: Real-time monitoring of the temperature distribution field data, and calculation of the average temperature change rate and local maximum temperature gradient of the component surface; Establish a mapping relationship between the rate of temperature change and the correction coefficient of the material's elastic modulus, and adjust the weights of the stiffness-related characteristic components in the vibration characteristic vector synchronously based on the correction coefficient. A mapping relationship between local temperature gradient and thermal stress concentration factor is established, and the contribution of stress amplitude-related characteristic components in vibration characteristic vector in fusion calculation is dynamically adjusted based on thermal stress concentration factor. In the multi-layered feature cross-aggregation network, a cross-modal attention mechanism is introduced for the feature representations derived from vibration and temperature, enabling the network to focus on the deep fusion of physically highly coupled feature channels; In the output layer of the algorithm, a normalization operation based on the thermal load coupling factor is introduced to ensure that the final generated fused feature representation is comparable under different thermal environment conditions.

[0009] As a further aspect of the present invention, the fused performance characteristics are input into a component performance evaluation model deployed in the cloud to obtain the comprehensive performance evaluation results and potential failure mode predictions of the electric vehicle component under test, including: The dynamic stress distribution of the structure is compared with the material fatigue strength threshold of the electric vehicle component under test stored in the performance evaluation model to identify the stress exceeding the standard area and calculate its over-limit accumulation time, thereby generating a mechanical fatigue risk index. The thermal load coupling factor is compared with the material thermal deformation threshold and insulation level threshold of the electric vehicle component under test stored in the performance evaluation model to evaluate the degree of thermal performance degradation and insulation aging risk, and generate a thermal failure risk index. The electromagnetic compatibility index is compared with the electromagnetic susceptibility threshold and external emission limit of the electric vehicle component under test stored in the performance evaluation model to assess the risk of internal and external electromagnetic interference and generate an electromagnetic interference risk index. The comprehensive performance evaluation result is obtained by integrating the mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index through weighted decision-making. Based on the relative magnitude and spatial distribution correlation of each risk index, the failure types and their locations are matched from the failure mode library of the performance evaluation model and used as the prediction of the potential failure modes.

[0010] As a further aspect of the present invention, based on the predicted potential failure modes, a targeted enhanced test stimulus sequence is generated from a preset test case library, including: The potential failure mode prediction is analyzed to extract the predicted failure type, location of occurrence, and dominant risk factors; Using the aforementioned dominant risk factors as search keywords, a set of basic test excitation templates are selected from the test case library. The basic test excitation templates include specific vibration patterns, temperature change curves, or electromagnetic interference waveforms. Based on the location of occurrence, the selected basic test excitation template is modified for spatial load distribution so that the application location and intensity distribution of the excitation are focused on the location of occurrence. Based on the severity of each risk index in the comprehensive performance evaluation results, the amplitude scaling factor and the number of iterations of the enhanced test stimulus sequence are determined. The modified and scaled basic test stimulus templates are arranged sequentially according to the coupling relationship of risk factors to generate the enhanced test stimulus sequence.

[0011] As a further aspect of the present invention, the enhanced test stimulus sequence is executed, and corresponding response data is collected to update the multi-source sensing data set for iterative testing, including: The enhanced test stimulus sequence is executed on the electric vehicle component under test, and its response is acquired at a higher sampling frequency to obtain high-resolution response data under enhanced testing. The high-resolution response data is processed using the same procedure as the initial test to generate fusion performance characteristics under enhanced testing. The fusion performance characteristics under the enhanced test are compared with the fusion performance characteristics obtained in the initial test to obtain the change in performance characteristics; The changes in the performance characteristics are fed back to the component performance evaluation model deployed in the cloud to update the calculation of the mechanical fatigue risk index, thermal failure risk index and electromagnetic interference risk index; Based on the updated risk index, an updated potential failure mode prediction is generated, and it is determined whether the test termination condition is met. If not, a new enhanced test stimulus sequence is generated again based on the updated potential failure mode prediction, and the test process continues iteratively.

[0012] As a further aspect of the present invention, the change in performance characteristics is fed back to the component performance evaluation model deployed in the cloud to update the calculation of the mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index, including: The change in the dynamic stress distribution of the structure is extracted from the change in the performance characteristics, and then superimposed on the initial stress distribution to obtain the cumulative stress distribution; Based on the accumulated stress distribution and combined with the nonlinear fatigue damage accumulation model of the material, the mechanical fatigue damage degree is recalculated and the mechanical fatigue risk index is updated. The change in thermal load coupling factor is extracted from the change in the performance characteristics, and combined with the real-time temperature field of the component. Through thermal-structural coupling simulation, the thermally induced deformation and stress field are updated, and the thermal failure risk index is reassessed. The changes in electromagnetic compatibility index are extracted from the changes in the performance characteristics, the evolution of the electromagnetic susceptibility of components under enhanced electromagnetic excitation is analyzed, and the electromagnetic interference risk index is updated. The updated mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index will serve as the basis for a new round of comprehensive performance evaluation and failure mode prediction.

[0013] As a further aspect of the present invention, in the multi-layered feature cross-aggregation network, a cross-modal attention mechanism is introduced for the feature representations derived from vibration and temperature, enabling the network to focus on physically highly coupled feature channels for deep fusion. The implementation steps include: From the pre-processed vibration feature vector and thermal feature vector, corresponding intermediate feature representation matrices are constructed respectively, and each column of the intermediate feature representation matrix represents a feature channel; The intermediate feature representation matrix derived from the vibration feature vector is denoted as the query matrix, and the intermediate feature representation matrix derived from the thermal feature vector is denoted as the bond matrix and the value matrix. Calculate the dot product between the query matrix and the transpose of the key matrix to obtain the attention score matrix; The attention score matrix is ​​scaled and normalized to obtain the cross-modal attention weight matrix; The value matrix is ​​weighted and summed using the cross-modal attention weight matrix to output the attention-weighted pre-fusion thermal feature representation; The attention-weighted pre-fusion thermal feature representation is fused with the original vibration intermediate feature representation matrix through element-wise addition to generate cross-modal feature interaction results. The cross-modal feature interaction results are input into the subsequent fully connected layers of the multi-layer feature cross-aggregation network for nonlinear transformation, and finally a unified fused feature representation is output.

[0014] As a further aspect of the present invention, the potential failure mode prediction is analyzed, and the predicted failure type, location of occurrence, and dominant risk factors are extracted. The implementation steps include: Receive the potential failure mode prediction, which is structured data or text string containing a failure type description, location coordinate information and risk factor identifiers; Natural language processing is performed on the text strings or structured data of the potential failure mode prediction to identify and extract keywords describing the failure mode, including fracture, overheating, insulation breakdown, and electromagnetic leakage. The identified keywords are then mapped to standardized failure type codes. The location coordinate information is parsed, and the location coordinate information is represented by a three-dimensional coordinate system or a two-dimensional grid number. The parsed coordinates are transformed into a coordinate system consistent with the digital model of the electric vehicle parts under test, and used as the occurrence location. The specific values ​​of the mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index are read from the potential failure mode prediction. By comparing the values ​​of the mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index, the risk category corresponding to the risk index with the largest value is determined as the dominant risk factor. The standardized fault type code, the converted location coordinates, and the identified dominant risk factor category are packaged and output for subsequent enhanced test stimulus sequence matching.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: By adjusting the feature fusion weight allocation mechanism based on the physical coupling relationship between vibration signals and temperature fields, the conventional calculation method of fixed weights is broken, and data processing is completed by following the inherent correlation between multiple types of sensor data. The proportion of different modal sensor data in the fusion calculation is balanced, reducing interference from invalid information when heterogeneous data is mixed, and standardizing the feature extraction logic of multi-source sensor data. The generation process of structural dynamic stress distribution, thermal load coupling factor, and electromagnetic compatibility index closely reflects the multi-physics coupling state of actual component operation, improving the operational logic of multi-modal data collaborative analysis, fully restoring the original characteristics of various performance dimensions of components, and optimizing the operational form of multi-source data integration processing.

[0016] By matching corresponding test cases with the predicted results of potential failure modes, differentiated enhanced test stimulus sequences are generated, changing the single, fixed test execution mode. Response data collected from newly added test stages are continuously entered into the sensor data set, continuously expanding the data dimensions and coverage. Iterative test processes are continuously implemented to broaden the coverage boundaries of test conditions and accumulate component operation data under different operating conditions. The data samples for performance evaluation are continuously enriched, gradually covering diverse operating states of components, uncovering performance characteristics that are difficult to cover with conventional testing, enriching the reference dimensions for performance evaluation, forming a dynamic and cyclical test operation mode, and improving the operational system for comprehensive component performance testing. Attached Figure Description

[0017] Figure 1 This is a flowchart of the electric vehicle component performance testing method based on multi-source sensor data described in this invention; Figure 2 A flowchart illustrating the process of generating fusion performance features using an improved multimodal feature fusion algorithm; Figure 3 A flowchart illustrating the process of adjusting the weight allocation mechanism for the improved multimodal feature fusion algorithm. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] See Figure 1 This invention provides a method for testing the performance of electric vehicle components based on multi-source sensor data, the implementation process of which includes: A multi-source sensor dataset containing structural vibration time-series data, temperature distribution field data, and electromagnetic interference spectrum data is acquired for the electric vehicle components under test under preset test conditions. An improved multi-modal feature fusion algorithm is used to collaboratively process this dataset, generating fused performance characteristics of the electric vehicle components. These characteristics include structural dynamic stress distribution, thermal load coupling factor, and electromagnetic compatibility index. The improved algorithm adjusts the weight allocation mechanism for feature fusion based on the physical coupling relationship between vibration signals and temperature fields. The fused performance characteristics are input into a component performance evaluation model deployed in the cloud to obtain a comprehensive performance evaluation result and potential failure mode prediction for the electric vehicle components. Based on the potential failure mode prediction, a targeted enhanced test stimulus sequence is generated from a preset test case library. This enhanced test stimulus sequence is executed, and corresponding response data is collected to update the multi-source sensor dataset for iterative testing.

[0021] In one embodiment of the present invention, after obtaining the multi-source sensor data set, refer to... Figure 2 The process involves joint time-frequency domain analysis of structural vibration time-series data to extract the dominant frequency, harmonic components, and damping characteristic parameters, forming a vibration feature vector. This includes empirical mode decomposition (EMD) of the structural vibration time-series data to obtain a series of intrinsic mode function (IMF) components. The instantaneous frequency and amplitude of each IMF component are calculated, and a Hilbert spectrum is plotted. Frequency bands with concentrated energy are identified from the Hilbert spectrum, with the highest energy frequency identified as the dominant frequency, and the energy characteristics at its integer multiples identified as harmonic components. A Hilbert transform is performed on the IMF components corresponding to the dominant frequency, and the phase angle of the analytic signal is calculated. The damping ratio of the system is obtained by fitting the phase angle change rate, serving as a damping characteristic parameter. The dominant frequency, the amplitudes of each harmonic component, and the damping ratio are arranged in a predetermined order to form the vibration feature vector. Finally, heat source identification and heat flow path analysis are performed on the temperature distribution field data to extract the location of the highest temperature point, temperature gradient, and thermal equilibrium time constant, forming a thermal feature vector. Frequency band energy and noise characteristics of electromagnetic interference (EMI) spectrum data are analyzed to extract characteristic frequency amplitudes, broadband noise floor, and impulse interference density, forming an electromagnetic feature vector. An improved multimodal feature fusion algorithm is then used to fuse the vibration, thermal, and electromagnetic feature vectors. This algorithm first dynamically calculates the coupling weights between vibration and thermal features based on real-time acquired temperature gradient data. Secondly, it adaptively adjusts the contribution of vibration features to the fusion process based on the impulse interference density in the EMI spectrum. Finally, a unified fused feature representation is generated through a multi-layered feature cross-aggregation network. Quantitative indicators reflecting mechanical, thermal, and electrical states are extracted from this unified fused feature representation and used as the structural dynamic stress distribution, thermal load coupling factor, and electromagnetic compatibility index, respectively.

[0022] In practice, after acquiring the multi-source sensor data set of the electric vehicle components under preset test conditions, an improved multi-modal feature fusion algorithm is used to collaboratively process the multi-source sensor data set. Time-frequency domain joint analysis is performed on the structural vibration time-series data to extract the dominant vibration frequency, harmonic components, and damping characteristic parameters, forming a vibration feature vector. Specifically, the time-frequency domain joint analysis is completed by performing empirical mode decomposition (EMD) on the structural vibration time-series data. EMD decomposes the non-stationary structural vibration time-series data into a series of intrinsic mode function (IMF) components arranged from high to low frequencies. The instantaneous frequency and instantaneous amplitude of each IMF component are calculated, and a Hilbert spectrum is plotted based on the calculation results. Frequency bands with concentrated energy are identified from the Hilbert spectrum, with the highest energy frequency being taken as the dominant vibration frequency, and the energy characteristics at integer multiples of the dominant vibration frequency being taken as harmonic components. Hilbert transform is performed on the components corresponding to the dominant vibration frequency in the intrinsic mode function components obtained from empirical mode decomposition, and the phase angle of their analytic signals is calculated. The damping ratio of the system is obtained by fitting the phase angle change rate, and the damping ratio is used as a damping characteristic parameter. The dominant vibration frequency, the amplitude of each harmonic component, and the damping ratio are arranged in a preset order to form a vibration characteristic vector.

[0023] In some embodiments, heat source identification and heat flow path analysis are performed on the temperature distribution field data to extract the location of the highest temperature point, temperature gradient, and thermal equilibrium time constant, forming a thermal feature vector. Frequency band energy and noise feature analysis is performed on the electromagnetic interference spectrum data to extract the characteristic frequency amplitude, broadband noise floor, and pulse interference density, forming an electromagnetic feature vector. An improved multimodal feature fusion algorithm is then used to fuse the vibration feature vector, thermal feature vector, and electromagnetic feature vector. In a specific implementation, the improved multimodal feature fusion algorithm dynamically calculates the coupling weight between the vibration feature vector and the thermal feature vector based on the real-time acquired temperature gradient data. In some embodiments, the calculation of the coupling weight has a functional relationship with the temperature gradient data; one optional functional relationship is expressed as: ; in: This represents the coupling weighting coefficient between vibrational and thermal characteristics. It is the weighting benchmark coefficient. Temperature affects the attenuation coefficient. This is the magnitude of the temperature gradient data calculated in real time. The contribution of the vibration feature vector in the fusion process is adaptively adjusted based on the pulse interference density in the electromagnetic interference spectrum; the higher the pulse interference density, the greater the reduction in the contribution of the vibration feature vector. A unified fused feature representation is generated through a multi-layer feature cross-aggregation network, which consists of alternating fully connected layers and nonlinear activation functions. Quantitative indicators reflecting the mechanical, thermal, and electrical states are extracted from the unified fused feature representation. The analysis process is completed through a linear mapping layer, outputting as the structural dynamic stress distribution, thermal load coupling factor, and electromagnetic compatibility index, respectively. The structural dynamic stress distribution can be understood as a sequence of stress values ​​associated with spatial location, the thermal load coupling factor is a scalar characterizing the intensity of thermo-mechanical interaction, and the electromagnetic compatibility index includes interference level data across multiple frequency bands.

[0024] In one embodiment of the present invention, the improved multimodal feature fusion algorithm adjusts the weight allocation mechanism for feature fusion based on the physical coupling relationship between vibration signals and temperature fields. See also... Figure 3The system monitors temperature distribution field data in real time and calculates the average temperature change rate and local maximum temperature gradient of component surfaces. A mapping relationship is established between the temperature change rate and the material elastic modulus correction coefficient, and the weights of stiffness-related feature components in the vibration feature vector are adjusted synchronously based on the correction coefficient. A mapping relationship is also established between the local temperature gradient and the thermal stress concentration factor, and the contribution of stress amplitude-related feature components in the vibration feature vector to the fusion calculation is dynamically adjusted based on the thermal stress concentration factor. In a multi-layered feature cross-aggregation network, a cross-modal attention mechanism is introduced for the feature representations derived from vibration and temperature, enabling the network to focus on deeply fused feature channels that are physically highly coupled. The implementation steps of this cross-modal attention mechanism include: constructing corresponding intermediate feature representation matrices from the pre-processed vibration and thermal feature vectors, where each column of the intermediate feature representation matrix represents a feature channel. The intermediate feature representation matrix derived from the vibration feature vector is denoted as the query matrix, and the intermediate feature representation matrix derived from the thermal feature vector is denoted as the key matrix and value matrix. The dot product between the query matrix and the transpose of the key matrix is ​​calculated to obtain the attention score matrix. The attention score matrix is ​​scaled and normalized to obtain the cross-modal attention weight matrix. The value matrix is ​​then weighted and summed using the cross-modal attention weight matrix to output the attention-weighted pre-fusion thermal feature representation. This attention-weighted pre-fusion thermal feature representation is then fused element-wise with the original vibration intermediate feature representation matrix to generate the cross-modal feature interaction result. This cross-modal feature interaction result is input into the subsequent fully connected layers of a multi-layer feature cross-aggregation network for nonlinear transformation, outputting a unified fused feature representation. At the algorithm's output layer, a normalization operation based on the thermal load coupling factor is introduced to ensure the comparability of the final fused feature representation under different thermal environmental conditions.

[0025] In practical implementation, the improved multimodal feature fusion algorithm adjusts the weight allocation mechanism of feature fusion based on the physical coupling relationship between vibration signals and temperature fields. It monitors temperature distribution field data in real time and calculates the average temperature change rate and local maximum temperature gradient on the surface of components. A mapping relationship is established between the temperature change rate and the material elastic modulus correction coefficient, and the weights of stiffness-related feature components in the vibration feature vector are adjusted synchronously based on the elastic modulus correction coefficient. In some embodiments, the relationship between the elastic modulus correction coefficient and the average temperature change rate can be defined by a piecewise function or lookup table. The elastic modulus correction coefficient directly affects the scaling ratio of stiffness-related components such as natural frequencies in the vibration feature vector. A mapping relationship is established between the local temperature gradient and the thermal stress concentration factor, and the contribution of stress amplitude-related feature components in the vibration feature vector to the fusion calculation is dynamically adjusted based on the thermal stress concentration factor. Optionally, the thermal stress concentration factor is proportional to the local maximum temperature gradient, so that vibration stress features corresponding to high temperature gradient regions receive higher weights during fusion.

[0026] In a multi-layered feature cross-aggregation network, a cross-modal attention mechanism is introduced to the feature representations derived from vibration and temperature, enabling the network to focus on deep fusion of physically highly coupled feature channels. Specifically, the implementation steps of the cross-modal attention mechanism include: constructing corresponding intermediate feature representation matrices from the pre-processed vibration and thermal feature vectors, where each column represents a feature channel. The intermediate feature representation matrix derived from the vibration feature vector is denoted as the query matrix Q, and the intermediate feature representation matrices derived from the thermal feature vector are denoted as the key matrix K and value matrix V. It can be understood that the query matrix Q, key matrix K, and value matrix V have compatible dimensions for matrix operations. The dot product between the query matrix Q and the transpose of the key matrix K is calculated to obtain the attention score matrix. The attention score matrix is ​​then scaled and normalized; scaling is achieved by dividing by the square root of the key vector dimension, and normalization uses the softmax function to obtain the cross-modal attention weight matrix A. The value matrix V is then weighted and summed using the cross-modal attention weight matrix A to output the attention-weighted, pre-fusion thermal feature representation. The attention-weighted pre-fusion thermal feature representation is fused element-wise with the original vibrational intermediate feature representation matrix to generate a cross-modal feature interaction result. This cross-modal feature interaction result is then input into subsequent fully connected layers of a multi-layer feature cross-aggregation network for nonlinear transformation, ultimately outputting a unified fused feature representation. In some embodiments, the nonlinear transformation may include a ReLU activation function.

[0027] In practical implementation, the mapping relationship between the temperature change rate and the material elastic modulus correction coefficient can be expressed as a function. One possible mapping relationship formula is: ; in: This represents the adjusted material elastic modulus correction factor. This represents the reference elastic modulus of the material at a reference temperature. It is the temperature sensitivity coefficient. It is the scaling factor for the rate of change. It is the calculated average temperature change rate of the component surface. It is a hyperbolic tangent function. Based on the elastic modulus correction factor... The weights of stiffness-related feature components in the vibration eigenvector are adjusted, for example, by multiplying the frequency-related feature components by... In the output layer of the improved multimodal feature fusion algorithm, a normalization operation based on the thermal load coupling factor is introduced. This normalization operation divides each element in the unified fused feature representation by the thermal load coupling factor of the current computation cycle, ensuring the comparability of the generated fused feature representations under different thermal environmental conditions. It can be understood that the thermal load coupling factor is a scalar value in each computation.

[0028] In one embodiment of the present invention, the fused performance characteristics are input into a component performance evaluation model deployed in the cloud to obtain a comprehensive performance evaluation result and potential failure mode prediction for the electric vehicle component under test. This process includes comparing the dynamic stress distribution of the structure with the material fatigue strength thresholds of the electric vehicle component under test stored in the performance evaluation model, identifying stress exceeding the limit area and calculating its over-limit accumulation time to generate a mechanical fatigue risk index. The thermal load coupling factor is compared with the material thermal deformation thresholds and insulation level thresholds of the electric vehicle component under test stored in the performance evaluation model to assess the degree of thermal performance degradation and insulation aging risk, generating a thermal failure risk index. Electromagnetic compatibility indicators are compared with the electromagnetic susceptibility thresholds and external emission limits of the electric vehicle component under test stored in the performance evaluation model to assess internal and external electromagnetic interference risks, generating an electromagnetic interference risk index. The mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index are fused, and a comprehensive performance evaluation result is obtained through weighted decision-making. Based on the relative magnitude and spatial distribution correlation of each risk index, the most likely failure type and its location are matched from the failure mode library of the performance evaluation model as potential failure mode predictions.

[0029] In practical implementation, the integrated performance characteristics are input into a component performance evaluation model deployed in the cloud to obtain the comprehensive performance evaluation results and potential failure mode predictions for the electric vehicle components under test. Specifically, the dynamic stress distribution of the structure is compared with the material fatigue strength thresholds of the electric vehicle components under test stored in the performance evaluation model to identify stress-exceeding areas and calculate their over-limit accumulation time, generating a mechanical fatigue risk index. The stress-exceeding area refers to the spatial location point or grid cell in the dynamic stress distribution of the structure where the stress value exceeds the material fatigue strength threshold. The over-limit accumulation time refers to the total time the stress value exceeds the threshold during the entire duration of the preset test condition. The calculation of the mechanical fatigue risk index comprehensively considers the area of ​​the over-limit area, the over-limit stress amplitude, and the over-limit accumulation time. In some embodiments, the thermal load coupling factor is compared with the material thermal deformation threshold and insulation level threshold of the electric vehicle components under test stored in the performance evaluation model to assess the degree of thermally induced performance degradation and insulation aging risk, generating a thermal failure risk index. The material thermal deformation threshold is used to determine whether the component will undergo excessive thermal deformation, and the insulation level threshold is used to determine whether the insulation performance of the material is within the allowable range. The calculation of the thermal failure risk index incorporates the degree of deviation between the thermal load coupling factor and the aforementioned threshold, as well as the duration of high temperature.

[0030] In practical implementation, the electromagnetic compatibility index is compared with the electromagnetic susceptibility thresholds and external emission limits of the electric vehicle components under test stored in the performance evaluation model to assess the risks of internal and external electromagnetic interference and generate an electromagnetic interference risk index. The electromagnetic susceptibility threshold is used to assess the component's tolerance to external electromagnetic interference, while the external emission limit is used to assess whether the level of electromagnetic interference emitted by the component during operation complies with regulations. The calculation of the electromagnetic interference risk index considers both the severity of internal and external interference risks. The mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index are integrated, and a comprehensive performance evaluation result is obtained through weighted decision-making. Weighted decision-making involves assigning a preset weight coefficient to each risk index. An optional weighted decision-making formula is expressed as follows: ; in: This indicates the overall performance evaluation result. This indicates the mechanical fatigue risk index. Indicates the thermal failure risk index. Indicates the electromagnetic interference risk index. , , These are the corresponding preset weight coefficients, and they satisfy... It is understandable that the weighting coefficients can be adjusted based on the type and importance of the component being measured.

[0031] Based on the relative magnitude and spatial distribution correlation of each risk index, the most likely failure type and its location are matched from the failure mode library of the performance evaluation model as potential failure mode predictions. In some embodiments, the failure mode library of the performance evaluation model is a predefined data table that stores the mapping relationship between different combinations of risk indices, spatial distribution characteristics, and known failure modes. The matching process compares the currently calculated mechanical fatigue risk index, thermal failure risk index, electromagnetic interference risk index, and spatial information of structural dynamic stress distribution and temperature field distribution with the entries in the failure mode library to find the best matching failure mode description and its typical location. See Table 1 for a partial mapping logic.

[0032] Table 1 Failure Mode Matching Logic Table: In one embodiment of the present invention, a targeted enhanced test stimulus sequence is generated by matching from a preset test case library based on potential failure mode prediction. This step includes parsing the potential failure mode prediction and extracting the predicted fault type, location, and dominant risk factor. Specifically, the steps are as follows: 1. Receiving the potential failure mode prediction, which is structured data or a text string containing a fault type description, location coordinate information, and risk factor identifiers. 2. Performing natural language processing on the text string or structured data of the potential failure mode prediction to identify and extract keywords describing the fault morphology, including fracture, overheating, insulation breakdown, and electromagnetic leakage. Mapping the identified keywords to standardized fault type codes. 3. Parsing the location coordinate information, which is represented in a three-dimensional coordinate system or two-dimensional grid numbering. Transforming the parsed coordinates to a coordinate system consistent with the digital model of the electric vehicle component under test, as the location of occurrence. 4. Reading the specific values ​​of the mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index from the potential failure mode prediction. 5. Comparing the values ​​of the mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index, and determining the risk category corresponding to the risk index with the largest value as the dominant risk factor. The standardized fault type codes, converted location coordinates, and identified dominant risk factor categories are packaged and output. Then, using the dominant risk factor as the search keyword, a set of basic test excitation templates is selected from the test case library. These templates include specific vibration spectra, temperature change curves, or electromagnetic interference waveforms. Based on the location of occurrence, the selected basic test excitation templates undergo spatial load distribution correction to focus the excitation application location and intensity distribution on the occurrence location. Based on the severity of various risk indices in the comprehensive performance evaluation results, the amplitude scaling factor and number of iterations for the enhanced test excitation sequence are determined. The corrected and scaled basic test excitation templates are then sequentially arranged according to the coupling relationship of risk factors to generate the enhanced test excitation sequence.

[0033] In practical implementation, the integrated performance characteristics are input into a component performance evaluation model deployed in the cloud to obtain the comprehensive performance evaluation results and potential failure mode predictions for the electric vehicle components under test. Specifically, the dynamic stress distribution of the structure is compared with the material fatigue strength thresholds of the electric vehicle components under test stored in the performance evaluation model to identify stress-exceeding areas and calculate their over-limit accumulation time, generating a mechanical fatigue risk index. The stress-exceeding area refers to the spatial location point or grid cell in the dynamic stress distribution of the structure where the stress value exceeds the material fatigue strength threshold. The over-limit accumulation time refers to the total time the stress value exceeds the threshold during the entire duration of the preset test condition. The calculation of the mechanical fatigue risk index comprehensively considers the area of ​​the over-limit area, the over-limit stress amplitude, and the over-limit accumulation time. In some embodiments, the thermal load coupling factor is compared with the material thermal deformation threshold and insulation level threshold of the electric vehicle components under test stored in the performance evaluation model to assess the degree of thermally induced performance degradation and insulation aging risk, generating a thermal failure risk index. The material thermal deformation threshold is used to determine whether the component will undergo excessive thermal deformation, and the insulation level threshold is used to determine whether the insulation performance of the material is within the allowable range. The calculation of the thermal failure risk index incorporates the degree of deviation between the thermal load coupling factor and the aforementioned threshold, as well as the duration of high temperature.

[0034] In practical implementation, the electromagnetic compatibility index is compared with the electromagnetic susceptibility thresholds and external emission limits of the electric vehicle components under test stored in the performance evaluation model to assess the risks of internal and external electromagnetic interference and generate an electromagnetic interference risk index. The electromagnetic susceptibility threshold is used to assess the component's tolerance to external electromagnetic interference, while the external emission limit is used to assess whether the level of electromagnetic interference emitted by the component during operation complies with regulations. The calculation of the electromagnetic interference risk index considers both the severity of internal and external interference risks. The mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index are integrated, and a comprehensive performance evaluation result is obtained through weighted decision-making. Weighted decision-making involves assigning a preset weight coefficient to each risk index. An optional weighted decision-making formula is expressed as follows: ; in: This indicates the overall performance evaluation result. This indicates the mechanical fatigue risk index. Indicates the thermal failure risk index. Indicates the electromagnetic interference risk index. , , These are the corresponding preset weight coefficients, and they satisfy... It is understandable that the weighting coefficients can be adjusted based on the type and importance of the component being measured.

[0035] Based on the relative magnitude and spatial distribution correlation of each risk index, the most likely failure type and its location are matched from the failure mode library of the performance evaluation model as potential failure mode predictions. In some embodiments, the failure mode library of the performance evaluation model is a predefined data table that stores the mapping relationship between different combinations of risk indices, spatial distribution characteristics, and known failure modes. The matching process compares the currently calculated mechanical fatigue risk index, thermal failure risk index, electromagnetic interference risk index, and spatial information of structural dynamic stress distribution and temperature field distribution with the entries in the failure mode library to find the best matching failure mode description and its typical location. See Table 2 for a partial mapping logic.

[0036] Table 2 Failure Mode Matching Logic Table: In one embodiment of the present invention, an enhanced test stimulus sequence is executed, and corresponding response data is collected to update the multi-source sensor dataset for iterative testing. The enhanced test stimulus sequence is executed on the electric vehicle component under test, and its response is collected at a higher sampling frequency to obtain high-resolution response data under enhanced testing. This high-resolution response data undergoes the same processing procedure as the initial test to generate fused performance characteristics under enhanced testing. The fused performance characteristics under enhanced testing are compared differentially with the fused performance characteristics obtained from the initial test to obtain the performance characteristic change. This performance characteristic change is fed back to the component performance evaluation model deployed in the cloud to update the calculation of the mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index. Specifically, the change in structural dynamic stress distribution is extracted from the performance characteristic change and superimposed with the initial stress distribution to obtain the cumulative stress distribution. Based on the cumulative stress distribution and combined with the material's nonlinear fatigue damage accumulation model, the mechanical fatigue damage degree is recalculated, and the mechanical fatigue risk index is updated. The change in thermal load coupling factor is extracted from the performance characteristic change and combined with the component's real-time temperature field. Through thermal-structural coupling simulation, the thermally induced deformation and stress field are updated, thereby re-evaluating the thermal failure risk index. The changes in electromagnetic compatibility (EMC) indicators are extracted from the changes in performance characteristics. The evolution of the electromagnetic susceptibility of components under enhanced electromagnetic excitation is analyzed, and the EMC risk index is updated. The updated mechanical fatigue risk index, thermal failure risk index, and EMC risk index are used as the basis for a new round of comprehensive performance evaluation and failure mode prediction. Based on the updated risk indices, updated potential failure mode predictions are generated, and it is determined whether the test termination conditions are met. If not, a new enhanced test excitation sequence is generated again based on the updated potential failure mode predictions, and the testing process continues iteratively.

[0037] In practice, an enhanced test stimulus sequence is executed and corresponding response data is collected to update the multi-source sensor dataset for iterative testing. The enhanced test stimulus sequence is executed on the components of the electric vehicle under test, and its response is collected at a higher sampling frequency to obtain high-resolution response data under enhanced testing. The higher sampling frequency is a significant improvement over the initial testing sampling frequency. The high-resolution response data under enhanced testing undergoes the same processing procedure as the initial testing, including joint time-frequency domain analysis, heat source identification and frequency band energy analysis, and an improved multimodal feature fusion algorithm to generate fused performance features under enhanced testing. The fused performance features under enhanced testing are then compared with the fused performance features obtained from the initial testing to obtain the change in performance features. The difference comparison involves calculating the difference between the corresponding values ​​of the two fused performance features element by element.

[0038] In practical implementation, the changes in performance characteristics are fed back to the component performance evaluation model deployed in the cloud to update the calculations of the mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index. The changes in the dynamic stress distribution of the structure are extracted from the performance characteristic changes and superimposed on the initial stress distribution to obtain the cumulative stress distribution. Superposition refers to the algebraic addition of stress values ​​at corresponding spatial locations. Based on the cumulative stress distribution and combined with the material's nonlinear fatigue damage accumulation model, the mechanical fatigue damage degree is recalculated, and the mechanical fatigue risk index is updated. One optional nonlinear fatigue damage accumulation model adopts a modified form of the Miner criterion, whose damage degree... The calculation formula is: ; in: This represents the mechanical fatigue damage degree calculated so far. This represents the total number of stress level levels into which the stress spectrum is divided. Indicates the first Level of stress The actual number of cycles experienced is obtained from historical data on cumulative stress distribution. Indicates the material's stress level The number of cycles required to reach failure is determined by the material's SN curve. It is a damage nonlinearity index greater than 1. The updated mechanical fatigue risk index and the calculated mechanical fatigue damage degree... Proportional.

[0039] In some embodiments, the change in thermal load coupling factor is extracted from the changes in performance characteristics and combined with the real-time temperature field of the component. Through thermo-structural coupling simulation, the thermally induced deformation and stress field are updated, thereby reassessing the thermal failure risk index. The thermo-structural coupling simulation uses the updated temperature field and thermal load coupling factor as input to calculate the new thermal deformation and thermal stress distribution. The change in electromagnetic compatibility index is extracted from the changes in performance characteristics to analyze the evolution of the component's electromagnetic susceptibility under enhanced electromagnetic excitation, updating the electromagnetic interference risk index. The evolution analysis focuses on the growth trend of characteristic frequency amplitude and the rise of the broadband noise floor. The updated mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index serve as the basis for a new round of comprehensive performance evaluation and failure mode prediction. It can be understood that the updating process enables the performance evaluation model to reflect the degradation of the component's state under enhanced testing excitation.

[0040] Based on the updated risk index, an updated potential failure mode prediction is generated, and it is determined whether the test termination condition is met. The test termination condition may be that the risk index exceeds a preset limit threshold, the number of iterations reaches a preset upper limit, or the change in performance characteristics is less than a preset convergence threshold. In some embodiments, if the test termination condition is not met, a new enhanced test stimulus sequence is generated again based on the updated potential failure mode prediction, and the iterative test process continues.

[0041] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for testing the performance of electric vehicle components based on multi-source sensor data, characterized in that, include: Acquire a set of multi-source sensor data of the electric vehicle component under test under preset test conditions. The set of multi-source sensor data includes structural vibration time series data, temperature distribution field data and electromagnetic interference spectrum data. An improved multimodal feature fusion algorithm is used to collaboratively process the multi-source sensor data set to generate fused performance characteristics of the electric vehicle components under test. The fused performance characteristics include structural dynamic stress distribution, thermal load coupling factor, and electromagnetic compatibility index. The improved multimodal feature fusion algorithm adjusts the weight allocation mechanism of feature fusion based on the physical coupling relationship between vibration signal and temperature field. The fused performance characteristics are input into the component performance evaluation model deployed in the cloud to obtain the comprehensive performance evaluation results and potential failure mode predictions of the electric vehicle component under test. Based on the predicted potential failure modes, a targeted enhanced test stimulus sequence is generated by matching from a preset test case library; The enhanced test stimulus sequence is executed, and corresponding response data is collected to update the multi-source sensor data set for iterative testing.

2. The method for testing the performance of electric vehicle components based on multi-source sensor data according to claim 1, characterized in that, The improved multimodal feature fusion algorithm is used to collaboratively process the multi-source sensor data set to generate fused performance features of the electric vehicle component under test, including: The vibration time series data of the structure are subjected to joint time-frequency domain analysis to extract the vibration dominant frequency, harmonic components and damping characteristic parameters, and to form a vibration feature vector. Heat source identification and heat flow path analysis are performed on the temperature distribution field data to extract the location of the highest temperature point, temperature gradient and thermal equilibrium time constant, and to form a thermal feature vector. The electromagnetic interference spectrum data is subjected to frequency band energy and noise feature analysis to extract the characteristic frequency amplitude, broadband noise floor and pulse interference density, and to form an electromagnetic feature vector. The improved multimodal feature fusion algorithm is invoked to fuse the vibration feature vector, thermal feature vector, and electromagnetic feature vector; wherein... The improved multimodal feature fusion algorithm first dynamically calculates the coupling weight between vibration and thermal features based on real-time acquired temperature gradient data, then adaptively adjusts the contribution of vibration features in the fusion process based on the pulse interference density in the electromagnetic interference spectrum, and finally generates a unified fusion feature representation through a multi-layer feature cross-aggregation network. Quantitative indicators that simultaneously reflect mechanical, thermal, and electrical states are extracted from the unified fusion feature representation and used as the dynamic stress distribution, thermal load coupling factor, and electromagnetic compatibility index of the structure, respectively.

3. The method for testing the performance of electric vehicle components based on multi-source sensor data according to claim 2, characterized in that, The vibration time-series data of the structure are subjected to joint time-frequency domain analysis to extract the dominant vibration frequency, harmonic components, and damping characteristic parameters, forming a vibration feature vector, including: Empirical mode decomposition (EMD) is performed on the vibration time series data of the structure to obtain a series of intrinsic mode function components. Calculate the instantaneous frequency and instantaneous amplitude of each intrinsic mode function component, and plot the Hilbert spectrum; Identify the frequency bands with concentrated energy from the Hilbert spectrum, take the frequency point with the highest energy as the dominant vibration frequency, and take the energy characteristics at its integer multiples as harmonic components; The Hilbert transform is performed on the components corresponding to the dominant vibration frequency in the intrinsic mode function components, the phase angle of its analytic signal is calculated, and the damping ratio of the system is obtained by fitting the phase angle change rate as a damping characteristic parameter. The vibration characteristic vector is formed by arranging the dominant vibration frequency, the amplitude of each harmonic component, and the damping ratio in a preset order.

4. The method for testing the performance of electric vehicle components based on multi-source sensor data according to claim 2, characterized in that, The improved multimodal feature fusion algorithm adjusts the weight allocation mechanism for feature fusion based on the physical coupling relationship between vibration signals and temperature fields. Its working principle includes: Real-time monitoring of the temperature distribution field data, and calculation of the average temperature change rate and local maximum temperature gradient of the component surface; Establish a mapping relationship between the rate of temperature change and the correction coefficient of the material's elastic modulus, and adjust the weights of the stiffness-related characteristic components in the vibration characteristic vector synchronously based on the correction coefficient. A mapping relationship between local temperature gradient and thermal stress concentration factor is established, and the contribution of stress amplitude-related characteristic components in vibration characteristic vector in fusion calculation is dynamically adjusted based on thermal stress concentration factor. In the multi-layered feature cross-aggregation network, a cross-modal attention mechanism is introduced for the feature representations derived from vibration and temperature, enabling the network to focus on the deep fusion of physically highly coupled feature channels; In the output layer of the algorithm, a normalization operation based on the thermal load coupling factor is introduced to ensure that the final generated fused feature representation is comparable under different thermal environment conditions.

5. The method for testing the performance of electric vehicle components based on multi-source sensor data according to claim 1, characterized in that, The fused performance characteristics are input into a component performance evaluation model deployed in the cloud to obtain the comprehensive performance evaluation results and potential failure mode predictions of the electric vehicle component under test, including: The dynamic stress distribution of the structure is compared with the material fatigue strength threshold of the electric vehicle component under test stored in the performance evaluation model to identify the stress exceeding the standard area and calculate its over-limit accumulation time, thereby generating a mechanical fatigue risk index. The thermal load coupling factor is compared with the material thermal deformation threshold and insulation level threshold of the electric vehicle component under test stored in the performance evaluation model to evaluate the degree of thermal performance degradation and insulation aging risk, and generate a thermal failure risk index. The electromagnetic compatibility index is compared with the electromagnetic susceptibility threshold and external emission limit of the electric vehicle component under test stored in the performance evaluation model to assess the risk of internal and external electromagnetic interference and generate an electromagnetic interference risk index. The comprehensive performance evaluation result is obtained by integrating the mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index through weighted decision-making. Based on the relative magnitude and spatial distribution correlation of each risk index, the failure types and their locations are matched from the failure mode library of the performance evaluation model and used as the prediction of the potential failure modes.

6. The method for testing the performance of electric vehicle components based on multi-source sensor data according to claim 5, characterized in that, Based on the predicted potential failure modes, targeted enhanced test stimulus sequences are generated from a pre-defined test case library, including: The potential failure mode prediction is analyzed to extract the predicted failure type, location of occurrence, and dominant risk factors; Using the aforementioned dominant risk factors as search keywords, a set of basic test excitation templates are selected from the test case library. The basic test excitation templates include specific vibration patterns, temperature change curves, or electromagnetic interference waveforms. Based on the location of occurrence, the selected basic test excitation template is modified for spatial load distribution so that the application location and intensity distribution of the excitation are focused on the location of occurrence. Based on the severity of each risk index in the comprehensive performance evaluation results, the amplitude scaling factor and the number of iterations of the enhanced test stimulus sequence are determined. The modified and scaled basic test stimulus templates are arranged sequentially according to the coupling relationship of risk factors to generate the enhanced test stimulus sequence.

7. The method for testing the performance of electric vehicle components based on multi-source sensor data according to claim 6, characterized in that, Execute the enhanced test stimulus sequence and collect corresponding response data to update the multi-source sensor data set, and perform iterative testing, including: The enhanced test stimulus sequence is executed on the electric vehicle component under test, and its response is acquired at a higher sampling frequency to obtain high-resolution response data under enhanced testing. The high-resolution response data is processed using the same procedure as the initial test to generate fusion performance characteristics under enhanced testing. The fusion performance characteristics under the enhanced test are compared with the fusion performance characteristics obtained in the initial test to obtain the change in performance characteristics; The changes in the performance characteristics are fed back to the component performance evaluation model deployed in the cloud to update the calculation of the mechanical fatigue risk index, thermal failure risk index and electromagnetic interference risk index; Based on the updated risk index, an updated potential failure mode prediction is generated, and it is determined whether the test termination condition is met. If not, a new enhanced test stimulus sequence is generated again based on the updated potential failure mode prediction, and the test process continues iteratively.

8. The method for testing the performance of electric vehicle components based on multi-source sensor data according to claim 7, characterized in that, The changes in the performance characteristics are fed back to the component performance evaluation model deployed in the cloud to update the calculations of the mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index, including: The change in the dynamic stress distribution of the structure is extracted from the change in the performance characteristics, and then superimposed on the initial stress distribution to obtain the cumulative stress distribution; Based on the accumulated stress distribution and combined with the nonlinear fatigue damage accumulation model of the material, the mechanical fatigue damage degree is recalculated and the mechanical fatigue risk index is updated. The change in thermal load coupling factor is extracted from the change in the performance characteristics, and combined with the real-time temperature field of the component. Through thermal-structural coupling simulation, the thermally induced deformation and stress field are updated, and the thermal failure risk index is reassessed. The changes in electromagnetic compatibility index are extracted from the changes in the performance characteristics, the evolution of the electromagnetic susceptibility of components under enhanced electromagnetic excitation is analyzed, and the electromagnetic interference risk index is updated. The updated mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index will serve as the basis for a new round of comprehensive performance evaluation and failure mode prediction.

9. The method for testing the performance of electric vehicle components based on multi-source sensor data according to claim 4, characterized in that, In the multi-layered feature cross-aggregation network, a cross-modal attention mechanism is introduced for the feature representations derived from vibration and temperature, enabling the network to focus on physically highly coupled feature channels for deep fusion. The implementation steps include: From the pre-processed vibration feature vector and thermal feature vector, corresponding intermediate feature representation matrices are constructed respectively, and each column of the intermediate feature representation matrix represents a feature channel; The intermediate feature representation matrix derived from the vibration feature vector is denoted as the query matrix, and the intermediate feature representation matrix derived from the thermal feature vector is denoted as the bond matrix and the value matrix. Calculate the dot product between the query matrix and the transpose of the key matrix to obtain the attention score matrix; The attention score matrix is ​​scaled and normalized to obtain the cross-modal attention weight matrix; The value matrix is ​​weighted and summed using the cross-modal attention weight matrix to output the attention-weighted pre-fusion thermal feature representation; The attention-weighted pre-fusion thermal feature representation is fused with the original vibration intermediate feature representation matrix through element-wise addition to generate cross-modal feature interaction results. The cross-modal feature interaction results are input into the subsequent fully connected layers of the multi-layer feature cross-aggregation network for nonlinear transformation, and finally a unified fused feature representation is output.

10. The method for testing the performance of electric vehicle components based on multi-source sensor data according to claim 6, characterized in that, The process of analyzing the potential failure mode prediction, extracting the predicted failure type, location, and dominant risk factors includes the following steps: Receive the potential failure mode prediction, which is structured data or text string containing a failure type description, location coordinate information and risk factor identifiers; Natural language processing is performed on the text strings or structured data of the potential failure mode prediction to identify and extract keywords describing the failure mode, including fracture, overheating, insulation breakdown, and electromagnetic leakage. The identified keywords are then mapped to standardized failure type codes. The location coordinate information is parsed, and the location coordinate information is represented by a three-dimensional coordinate system or a two-dimensional grid number. The parsed coordinates are transformed into a coordinate system consistent with the digital model of the electric vehicle parts under test, and used as the occurrence location. The specific values ​​of the mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index are read from the potential failure mode prediction. By comparing the values ​​of the mechanical fatigue risk index, thermal failure risk index, and electromagnetic interference risk index, the risk category corresponding to the risk index with the largest value is determined as the dominant risk factor. The standardized fault type code, the converted location coordinates, and the identified dominant risk factor category are packaged and output for subsequent enhanced test stimulus sequence matching.

Images