Chassis detection method and detection system of new energy automobile

Abstract

The invention discloses a chassis detection method and device for a new energy automobile. The chassis detection method of the new energy automobile comprises the following steps: acquiring spectral image data and voiceprint data of an automobile chassis; performing space matching processing on the visible light image, the infrared image and the ultraviolet image; performing feature extraction processing on the visible light image, the infrared image and the ultraviolet image which are subjected to space matching processing, and determining an image anomaly detection result; determining voiceprint features according to the voiceprint data; wherein the voiceprint features comprise voiceprint transient features and voiceprint steady-state features; determining a voiceprint anomaly detection result according to the voiceprint transient state feature and the voiceprint steady state feature; and determining a defect position and a defect type of the automobile chassis according to the image anomaly detection result and the voiceprint anomaly detection result. According to the invention, the safety detection technology capability and detection efficiency of the new energy automobile chassis are improved, and the defect detection precision of the new energy automobile chassis is improved.

Description

Technical Field

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

[0002] With the large-scale development of the new energy vehicle industry, the safety performance of power batteries, as the core component of new energy vehicle power, is directly related to vehicle operation safety and the safety of users' lives and property. Power batteries experience performance degradation after multiple charge-discharge cycles, and overcharging, over-discharging, and external high and low temperature environments accelerate this degradation, further leading to problems such as reduced battery capacity and decreased charge-discharge efficiency. Furthermore, current new energy vehicles generally place power battery packs in the vehicle chassis, making them susceptible to damage during driving due to collisions, scratches, or other external forces, which can cause damage to the battery pack or internal cells, becoming a hidden danger of rapid performance degradation and escalating failures. In severe cases, this can even lead to thermal runaway, causing fires and explosions.

[0003] Traditional testing methods typically rely on manual visual inspection or close-range contact testing, which suffers from low efficiency. With the rapid growth of the electric vehicle market, the limitations of existing testing methods have gradually become apparent, and they can no longer meet the high-precision testing needs in complex environments. Summary of the Invention

[0004] This invention provides a chassis inspection method and system for new energy vehicles, in order to improve the safety inspection technology and efficiency of new energy vehicle chassis, as well as the accuracy of defect detection of new energy vehicle chassis.

[0005] According to one aspect of the present invention, a chassis testing method for a new energy vehicle is provided, the chassis testing method comprising:

[0006] Acquire spectral image data and acoustic signature data of the vehicle chassis; the spectral image data includes visible light images, infrared images, and ultraviolet images;

[0007] Spatial matching processing is performed on visible light images, infrared images, and ultraviolet images;

[0008] Feature extraction is performed on visible light, infrared, and ultraviolet images that have undergone spatial matching to determine the image anomaly detection results.

[0009] Voiceprint features are determined based on voiceprint data; among which, voiceprint features include transient voiceprint features and steady-state voiceprint features.

[0010] The results of voiceprint anomaly detection are determined based on the transient and steady-state characteristics of voiceprint.

[0011] The location and type of defects in the vehicle chassis are determined based on the results of image anomaly detection and voiceprint anomaly detection.

[0012] Furthermore, spatial matching processing is performed on the visible light image, infrared image, and ultraviolet image to determine the spectral matching image. This process also includes:

[0013] Preprocessing is performed on visible light images, infrared images, and ultraviolet images.

[0014] Furthermore, feature extraction processing is performed on the spectral matching image to determine the image anomaly detection results, including:

[0015] Based on Hough line transform, spatial matching processing is performed on visible light images, infrared images, and ultraviolet images.

[0016] Furthermore, feature extraction processing is performed on the visible light, infrared, and ultraviolet images that have undergone spatial matching to determine the image anomaly detection results, including:

[0017] Feature extraction processing is performed on visible light images, infrared images, and ultraviolet images that have undergone spatial matching to determine the feature data map of each type of image;

[0018] Each feature data map is fused to form a spectral fusion feature map;

[0019] The location and type of defects in the vehicle chassis are determined based on the spectral fusion feature map.

[0020] Furthermore, feature extraction processing is performed on the visible light, infrared, and ultraviolet images that have undergone spatial matching to determine the feature data map for each type of image, including:

[0021] Based on the visible light image that has undergone spatial matching processing, visible light feature information is extracted, and a visible light feature map is determined based on the visible light feature information; wherein, the visible light feature information includes visible light morphological features, texture features, and edge features;

[0022] Infrared feature information is extracted from the infrared image after spatial matching processing, and an infrared feature map is determined based on infrared temperature features; among which, infrared temperature features include temperature gradient features and temperature statistical features.

[0023] Based on the ultraviolet images that have undergone spatial matching processing, ultraviolet feature information is extracted, and an ultraviolet feature map is determined based on the ultraviolet feature information; among which, the ultraviolet feature information includes discharge intensity features, ultraviolet morphological features, and spatial features.

[0024] Furthermore, voiceprint features are determined based on the voiceprint data, including:

[0025] Preprocess the voiceprint data to determine the discrete frame voiceprint signal;

[0026] Signal extraction processing is performed on discrete frame voiceprint signals to determine transient and steady-state voiceprint features.

[0027] Furthermore, the results of voiceprint anomaly detection are determined based on the transient and steady-state characteristics of the voiceprint, including:

[0028] Defect categories are determined by combining transient and steady-state characteristics of voiceprints with a voiceprint fault diagnosis model.

[0029] The coordinates of the sound source of the defect were determined using the TDOA algorithm;

[0030] The results of acoustic anomaly detection are determined based on the defect category and the coordinates of the sound source.

[0031] Furthermore, based on the image anomaly detection results and voiceprint anomaly detection results, the location and type of defects in the vehicle chassis are determined, including:

[0032] The image anomaly detection results and voiceprint anomaly detection results are processed by rule association to determine the location and type of defects in the car chassis.

[0033] Furthermore, spectral image data and voiceprint data of the vehicle chassis are acquired, including:

[0034] Spectral image data is acquired using a visible light camera, an infrared thermal imager, and an ultraviolet imager, and voiceprint data is acquired using a voiceprint sensor.

[0035] According to another aspect of the present invention, a chassis testing system for new energy vehicles is provided, which is used in any of the chassis testing methods for new energy vehicles described in the above embodiments.

[0036] The chassis inspection method for new energy vehicles provided in this invention achieves simultaneous acquisition of multi-source sensor data of the vehicle chassis by acquiring spectral image data and acoustic fingerprint data. Compared with the low efficiency and limited coverage of existing technologies that rely on manual inspection or contact sensor inspection, this method enables real-time non-contact inspection of the vehicle chassis, improving the safety inspection capabilities and efficiency of new energy vehicle chassis. Furthermore, by performing spatial matching processing on visible light, infrared, and ultraviolet images in the spectral image data, and then extracting features from these images, the method determines image anomaly detection results. Simultaneously, it determines acoustic fingerprint features based on acoustic fingerprint data, and further determines acoustic fingerprint anomaly detection results based on transient and steady-state acoustic fingerprint features. Finally, it determines the defect location and type of the vehicle chassis based on both image and acoustic fingerprint anomaly detection results. This method achieves the determination of defect location and type of the vehicle chassis through the complementarity and fusion of multi-dimensional information, improving the defect detection accuracy of new energy vehicle chassis.

[0037] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a flowchart of a chassis testing method for a new energy vehicle according to an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of chassis testing for a new energy vehicle according to an embodiment of the present invention. Detailed Implementation

[0041] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0042] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0043] This invention provides a chassis testing method for new energy vehicles. Figure 1 This is a flowchart of a chassis testing method for new energy vehicles according to an embodiment of the present invention. Figure 2 This is a schematic diagram of chassis testing for a new energy vehicle according to an embodiment of the present invention, for reference. Figure 1 and Figure 2 The chassis testing methods for new energy vehicles include:

[0044] S110. Acquire spectral image data and acoustic data of the vehicle chassis; wherein, the spectral image data includes visible light image, infrared image and ultraviolet image.

[0045] Specifically, a non-contact detection system can be constructed using a visible light camera 1, an infrared thermal imager 2, an ultraviolet imager 3, and a voiceprint sensor 4 to acquire spectral image data and voiceprint data. That is, the visible light camera 1 acquires visible light images, the infrared thermal imager 2 acquires infrared images, the ultraviolet imager 3 acquires ultraviolet images, and the voiceprint sensor 4 acquires voiceprint data. In this embodiment, the visible light camera 1 can be a high-resolution industrial camera with high dynamic range and low-illuminance performance to acquire visible light images of the chassis surface for detecting surface damage, corrosion, deformation, and other defects. The infrared thermal imager 2 can be an uncooled infrared thermal imager with high sensitivity and high resolution to acquire infrared thermal images of the chassis surface for detecting temperature anomalies and identifying faults such as overheating and overcooling. The ultraviolet imager 3 can be a solar-blind ultraviolet imager with high sensitivity and anti-interference capabilities to acquire ultraviolet images of the chassis surface for detecting corona discharge and identifying faults such as insulation degradation and partial discharge.

[0046] Furthermore, when mechanical and electrical structures of new energy vehicles experience malfunctions and defects, they will generate weak abnormal signals. By using the voiceprint sensor 4 to collect voiceprint data, the source of the fault can be located, and potential defects inside the equipment can be explored in depth. The voiceprint sensor 4 can use an 8-channel high-sensitivity microphone array with wide bandwidth and low noise performance. The microphone array is fixed in the detection area of ​​the new energy vehicle chassis in a regular hexagonal distribution. With the center of the front axle of the chassis as the coordinate origin, the three-dimensional coordinates of each microphone are pre-calibrated. The voiceprint sensor 4 collects mechanical waves at a specific period and converts them into electrical signals to obtain the time-domain signal of the voiceprint.

[0047] S120. Perform spatial matching processing on visible light images, infrared images, and ultraviolet images.

[0048] Specifically, since different cameras and sensors are at different distances and angles from the car chassis, tilting and deformation may occur. Therefore, the three types of images can be geometrically corrected based on the Hough line transform to achieve spatial registration.

[0049] S130. Perform feature extraction processing on the visible light image, infrared image, and ultraviolet image that have undergone spatial matching processing to determine the image anomaly detection results.

[0050] Specifically, feature extraction is performed on the visible light, infrared, and ultraviolet images that undergo spatial matching to determine the feature data map for each type of image; each feature data map is fused to form a spectral fusion feature map; and the location and type of defects in the vehicle chassis are determined based on the spectral fusion feature map.

[0051] S140. Determine voiceprint features based on voiceprint data; among which, voiceprint features include transient voiceprint features and steady-state voiceprint features.

[0052] Specifically, the voiceprint data can be preprocessed first to determine the discrete frame voiceprint signal, and then the signal extraction processing can be performed based on the discrete frame voiceprint signal to determine the transient and steady-state characteristics of the voiceprint.

[0053] S150. Determine the voiceprint anomaly detection results based on the transient and steady-state characteristics of the voiceprint.

[0054] Specifically, the defect category can be determined by combining the transient and steady-state characteristics of the voiceprint with the voiceprint fault diagnosis model, and the sound source coordinates of the defect can be determined by the TDOA algorithm. Finally, the voiceprint anomaly detection result can be determined based on the defect category and the sound source coordinates.

[0055] S160. Determine the location and type of defects in the vehicle chassis based on the image anomaly detection results and the voiceprint anomaly detection results.

[0056] Specifically, it is necessary to perform rule-based correlation processing on the image anomaly detection results and the voiceprint anomaly detection results to determine the location and type of defects in the car chassis.

[0057] The chassis inspection method for new energy vehicles provided in this invention achieves simultaneous acquisition of multi-source sensor data of the vehicle chassis by acquiring spectral image data and acoustic fingerprint data. Compared with the low efficiency and limited coverage of existing technologies that rely on manual inspection or contact sensor inspection, this method enables real-time non-contact inspection of the vehicle chassis, improving the safety inspection capabilities and efficiency of new energy vehicle chassis. Furthermore, by performing spatial matching processing on visible light, infrared, and ultraviolet images in the spectral image data, and then extracting features from these images, the method determines image anomaly detection results. Simultaneously, it determines acoustic fingerprint features based on acoustic fingerprint data, and further determines acoustic fingerprint anomaly detection results based on transient and steady-state acoustic fingerprint features. Finally, it determines the defect location and type of the vehicle chassis based on both image and acoustic fingerprint anomaly detection results. This method achieves the determination of defect location and type of the vehicle chassis through the complementarity and fusion of multi-dimensional information, improving the defect detection accuracy of new energy vehicle chassis.

[0058] Furthermore, spatial matching processing is performed on the visible light image, infrared image, and ultraviolet image to determine the spectral matching image. This process also includes:

[0059] Preprocessing is performed on visible light images, infrared images, and ultraviolet images.

[0060] Specifically, since different sensors have different spectral response curves, the acquired visible light images, infrared images and ultraviolet images need to be spectral response corrected. At the same time, they also need to be preprocessed by denoising, feature enhancement and pseudo color mapping respectively. For example, for visible light images, Gaussian filtering can be used to remove minor noise while preserving details such as surface cracks and corrosion. The CLAHE algorithm is then used to enhance local contrast, highlighting the difference between defects such as cracks and corrosion and normal areas. Finally, white balance calibration is performed using a standard white board as a reference to obtain a calibrated visible light image, eliminating color deviations caused by ambient light and ensuring accurate color reproduction. For infrared images, wavelet denoising is first used to eliminate thermal noise interference, and then grayscale stretching is used to enhance the temperature gradient, making hot spots more prominent. The correspondence between grayscale and temperature is calibrated based on a standard black body, and finally, pseudo-color mapping is performed to obtain a visualized image of temperature distribution, intuitively presenting the temperature distribution. For ultraviolet images, morphological opening operations are first used to remove false discharge points, and then grayscale stretching is performed on the discharge signal to enhance the brightness of the real discharge area. The discharge intensity is calibrated based on a standard discharge power source, and finally, pseudo-color mapping is performed to obtain a visualized image of discharge intensity, intuitively presenting the distribution of discharge strength.

[0061] Furthermore, feature extraction processing is performed on the spectral matching image to determine the image anomaly detection results, including:

[0062] Based on Hough line transform, spatial matching processing is performed on visible light images, infrared images, and ultraviolet images.

[0063] Specifically, spatial registration of visible light, infrared, and ultraviolet images is performed based on the Hough line transform. First, the Canny algorithm is used to extract the chassis edges to enhance the straight-line features of the car chassis's rigid structure. Then, the Hough line transform is applied to the edge-enhanced image, and a minimum voting threshold is set to extract reference lines. Mutually perpendicular orthogonal feature lines are selected as geometric references, and their angles, distances, and other parameters are recorded. Geometric correction is performed on individual images, calculating the angles between the reference lines and the horizontal and vertical directions of the image. Overall tilt is eliminated through rotation transformation, and a scaling factor is calculated based on known physical dimensions. Affine transformation is then used to correct perspective distortion, ensuring that the spacing between lines in the image matches reality. Using the visible light image as a reference, the above steps are repeated in the infrared and ultraviolet images, extracting the same orthogonal feature lines and performing geometric correction. The positional deviations between corresponding feature lines in the infrared and ultraviolet images and the visible light image are calculated. Translation transformation is used to align the feature lines of the three types of images, ensuring that the pixel coordinate overlap of the same physical line reaches 95% in all three types of images, thus completing spatial registration.

[0064] Furthermore, feature extraction processing is performed on the visible light, infrared, and ultraviolet images that have undergone spatial matching to determine the image anomaly detection results, including:

[0065] Feature extraction processing is performed on visible light images, infrared images, and ultraviolet images that have undergone spatial matching to determine the feature data map of each type of image;

[0066] Each feature data map is fused to form a spectral fusion feature map;

[0067] The location and type of defects in the vehicle chassis are determined based on the spectral fusion feature map.

[0068] Specifically, visible light feature information can be extracted from spatially matched visible light images, and a visible light feature map can be determined based on this information. This visible light feature information includes visible light morphological features, texture features, and edge features. Similarly, infrared feature information can be extracted from spatially matched infrared images, and an infrared feature map can be determined based on infrared temperature features. Infrared temperature features include temperature gradient features and temperature statistical features. Ultraviolet (UV) feature information can be extracted from spatially matched UV images, and a UV feature map can be determined based on this information. UV feature information includes discharge intensity features, UV morphological features, and spatial features. The visible light, infrared, and UV feature maps are then normalized, and channel-level stitching is performed based on the spatially registered pixel coordinates to form a spectral fusion feature map. This spectral fusion feature map is then input into a fault diagnosis model for defect identification, thereby determining the location and type of defects in the vehicle chassis.

[0069] Furthermore, feature extraction processing is performed on the visible light, infrared, and ultraviolet images that have undergone spatial matching to determine the feature data map for each type of image, including:

[0070] Based on the visible light image that has undergone spatial matching processing, visible light feature information is extracted, and a visible light feature map is determined based on the visible light feature information; wherein, the visible light feature information includes visible light morphological features, texture features, and edge features;

[0071] Infrared feature information is extracted from the infrared image after spatial matching processing, and an infrared feature map is determined based on infrared temperature features; among which, infrared temperature features include temperature gradient features and temperature statistical features.

[0072] Based on the ultraviolet images that have undergone spatial matching processing, ultraviolet feature information is extracted, and an ultraviolet feature map is determined based on the ultraviolet feature information; among which, the ultraviolet feature information includes discharge intensity features, ultraviolet morphological features, and spatial features.

[0073] Specifically, for visible light images, an adaptive thresholding segmentation method is used to binarize the calibrated visible light images, and candidate regions for defects and anomalies are segmented based on gray-level differences. For candidate regions for defects and anomalies, contours are extracted using a contour detection algorithm, morphological parameters are calculated to form contour feature maps, texture parameters are calculated using a gray-level co-occurrence matrix to obtain texture feature maps, edge features are extracted using an improved Cann operator, and edge contours are refined using a morphological dilation operation. The magnitude and direction distribution of edge gradients are calculated to obtain edge feature maps. The contour feature maps, texture feature maps, and edge feature maps are then concatenated by channel to obtain visible light feature maps, which are used to characterize structural defects on the chassis surface.

[0074] For infrared images, the Sobel operator is used to calculate the temperature distribution visualization image, i.e., the gradient of the temperature field in the horizontal and vertical directions in the infrared image. Regions with gradient amplitudes greater than a preset threshold are extracted as candidate regions for thermal anomalies, and the gradient direction histogram is calculated to obtain a temperature gradient feature map. Temperature statistical parameters are calculated for the candidate regions for thermal anomalies to generate a temperature statistical feature map, which includes the average temperature, temperature standard deviation, and temperature difference. The temperature gradient feature map and the temperature statistical feature map are stitched together according to channels to obtain an infrared feature map, which is used to characterize thermal anomaly defects on the chassis surface.

[0075] For ultraviolet images, the discharge intensity visualization image, i.e., the ultraviolet image, is segmented using an adaptive threshold based on grayscale differences to obtain candidate regions for discharge anomalies. The discharge intensity gradient, average discharge intensity, intensity standard deviation, and maximum intensity of the candidate regions for discharge anomalies are extracted to generate a discharge intensity feature map. In addition, morphological features such as area, perimeter, and extension are extracted from the candidate regions for discharge anomalies to obtain a discharge morphology feature map. The discharge intensity feature map and the discharge morphology feature map are then stitched together according to channels to obtain an ultraviolet feature map.

[0076] Based on the pixel coordinates of the spatially registered visible light, infrared, and ultraviolet images, the intersection of the candidate regions for anomalies in the three types of images is calculated. If the overlap is greater than 70%, it is identified as a high-confidence candidate region for anomalies, and a candidate region appearing alone is marked as a low-confidence candidate region for anomalies. Combining the prior knowledge base of the vehicle chassis, the candidate regions for anomalies are matched. If the overlap between the candidate region for anomalies and the contour of normal chassis components is greater than 85%, it is identified as a false defect and removed, resulting in the final set of candidate regions for anomalies. For each candidate region for anomalies, the corresponding spectral fusion feature map is extracted and input into the fault diagnosis model, which then outputs the defect type. For each identified candidate region for anomalies, the geometric center coordinates of its minimum bounding rectangle are calculated as the initial pixel position of the defect, and the width and height dimensions of the minimum bounding rectangle are recorded. Based on the camera calibration parameters and sensor installation position, the pixel coordinates are converted into chassis physical coordinates through perspective projection transformation to obtain the defect location. Among them, a lightweight convolutional neural network can be selected as the backbone of the fault diagnosis model. The fault diagnosis model is trained using historical data, and the model parameters are dynamically updated by combining the cross-entropy function and the Adam optimizer to obtain a trained fault diagnosis model.

[0077] Furthermore, determining voiceprint features based on voiceprint data also includes:

[0078] Preprocess the voiceprint data to determine the discrete frame voiceprint signal;

[0079] Signal extraction processing is performed on discrete frame voiceprint signals to determine transient and steady-state voiceprint features.

[0080] Specifically, the voiceprint data undergoes preprocessing, which includes pre-emphasis, denoising, filtering, and framing. For example, firstly, a first-order linear high-pass filter is used for the original voiceprint data. High-frequency enhancement is achieved through a pre-emphasis formula to avoid mistakenly filtering out effective high-frequency components as noise during noise removal. The enhanced partial discharge pulse signal is more conducive to the denoising algorithm distinguishing noise from real discharge. The pre-emphasis formula is as follows:

[0081] y[n]=x[n]-αx[n-1], α∈[0.9,0.97];

[0082] Where y[n] is the pre-emphasized voiceprint data, x[n] represents the nth sampling point of the original voiceprint data, α is the pre-emphasis coefficient, α∈[0.9,0.97], preferably α=0.95.

[0083] To eliminate sporadic interference, the pre-emphasized voiceprint data is decomposed into 5 layers using the db6 wavelet basis. Noisy high-frequency wavelet coefficients are then soft-thresholded before reconstructing the voiceprint data to eliminate sporadic interference. To eliminate periodic interference, the SVD denoising algorithm is used to arrange the voiceprint data into a matrix frame by frame. Singular value decomposition is performed on the matrix, discarding small singular value components, and the matrix is ​​then reconstructed to eliminate periodic interference, resulting in denoised voiceprint data. Bandpass filtering is used to remove environmental noise outside the frequency band, such as low-frequency background vibration and high-frequency electromagnetic interference. A Mel filter is then introduced to convert the filtered voiceprint data to the Mel frequency domain. The Mel spectral capability of each frame is calculated using the Mel filter bank, and an energy threshold is set to suppress residual noise within the band. The data is then converted to a voiceprint time-domain signal using inverse Mel transform. The Mel-enhanced voiceprint data is then framed with a frame length of 20ms and a frame shift of 50%. A Hanning window is applied to each frame to divide the continuous voiceprint data into several frames, resulting in discrete frame voiceprint signals.

[0084] After determining the discrete frame acoustic signature signal, based on the discrete frame acoustic signature signal, the transient frame is first located by detecting short-time energy and zero-crossing rate, and the peak amplitude, energy rise rate, high frequency ratio and instantaneous frequency in the frequency domain of the frame are extracted; for the steady-state frame, the Mel-Frequency Cepstral Coefficients (MFCC) and the corresponding first-order and second-order differences, the Gamma-frequency Cepstral Coefficients (GFCC) and the corresponding difference values ​​are calculated, and combined with the spectral centroid, bandwidth and autocorrelation coefficient, the transient and steady-state characteristics of the acoustic signature are determined.

[0085] Furthermore, the results of voiceprint anomaly detection are determined based on the transient and steady-state characteristics of the voiceprint, including:

[0086] Defect categories are determined by combining transient and steady-state characteristics of voiceprints with a voiceprint fault diagnosis model.

[0087] The coordinates of the sound source of the defect were determined using the TDOA algorithm;

[0088] The results of acoustic anomaly detection are determined based on the defect category and the coordinates of the sound source.

[0089] The training process of the voiceprint fault diagnosis model is described as follows: First, a large amount of historical voiceprint data from new energy vehicles under normal operating conditions is collected as positive samples. After pre-emphasis, denoising, filtering, and framing operations, voiceprint features are extracted to construct a positive sample feature library, which serves as the benchmark for anomaly detection. Then, an autoencoder is used to construct a positive sample model, which is trained using the positive sample feature library to learn the feature distribution patterns of normal voiceprint data. Samples judged as "normal" by the positive sample model are automatically included in the positive sample library, while samples judged as "abnormal" by the positive sample model are automatically triggered in a high-precision acquisition mode to record the voiceprint features for that period. Next, samples judged as abnormal are labeled with fault types through manual annotation, and the labeled samples are included in the negative sample library. For recurring abnormal samples, their core features are extracted to establish a fault feature template library, which is then used for subsequent positive... When the sample model collects samples with similar features, it automatically matches templates and pre-labels them, which are then manually confirmed before being added to the database. Next, a generative adversarial network (GAN) is used to enhance scarce negative samples. Based on the feature distribution of real samples, five times the number of synthetic samples are generated, and different degrees of anomaly are simulated by adjusting parameters such as frequency. Based on the positive sample model, positive and negative samples are continuously collected and added to the full sample database. A classification head is added to the positive sample autoencoder to construct the full sample model. Using the positive sample model as a foundation, the full sample model is trained through transfer learning. The weights of the first three layers of the autoencoder are frozen, and the parameters of the last two layers and the classification head are fine-tuned based on the full sample database. The loss function is a weighted sum of reconstruction loss and cross-entropy loss. When the accuracy of the full sample model on the test set is higher than that of the positive sample model, it is automatically switched to the main model as the voiceprint fault diagnosis model.

[0090] Specifically, the TDOA algorithm can be used to calculate the sound source coordinates of the defect location. For multi-channel acoustic time-domain signals during the fault period, the GCC-PHAT algorithm is used to calculate the signal arrival time difference (TDOA) between each pair of microphones. Based on the TDOA value and the microphone array coordinates, a set of distance difference equations is constructed. The least squares method is used to solve the sound source coordinates, thereby realizing the sound source localization of the defect location. Then, the acoustic anomaly detection result is determined according to the defect type and the sound source coordinates.

[0091] Furthermore, based on the image anomaly detection results and voiceprint anomaly detection results, the location and type of defects in the vehicle chassis are determined, including:

[0092] The image anomaly detection results and voiceprint anomaly detection results are processed by rule association to determine the location and type of defects in the car chassis.

[0093] Specifically, based on the timestamp synchronized by the Picture Transfer Protocol (PTP), image and voiceprint detection results within the same detection period are matched, with the time error controlled within 10ms. Then, the Euclidean distance between the physical coordinates of the two types of results is calculated based on the image and voiceprint anomaly detection results. Distances ≤50mm are classified as the same defect, outputting a "double confirmation" result, along with the defect location and type. When spatially related but with inconsistent defect categories, the high-confidence modality result takes precedence, and the other defect category is marked as a defect requiring further review. When only one modality detects a defect location, the defect result is retained and marked as "single-modal confirmation." Finally, the multimodal fusion result is output in structured data form, including defect ID, defect category, physical coordinates, confidence level, fusion marker, and defect size, providing standardized input for subsequent risk grading decisions. Risk grading decisions are made based on the multimodal fusion results. According to the severity of the fused fault characteristics, the vehicle chassis status is divided into four levels: normal, Level 1 warning, Level 2 warning, and Level 3 warning, with different decision outputs for each level.

[0094] Furthermore, spectral image data and voiceprint data of the vehicle chassis are acquired, including:

[0095] The spectral image data is acquired using a visible light camera, an infrared thermal imager, and an ultraviolet imager, and the voiceprint data is acquired using a voiceprint sensor.

[0096] This invention provides a chassis testing system for new energy vehicles. The chassis testing system is used to execute the chassis testing method for new energy vehicles described in any of the above embodiments, and therefore has the beneficial effects of the chassis testing method for new energy vehicles, which will not be repeated here.

[0097] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0098] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for testing the chassis of a new energy vehicle, characterized in that, include: Acquire spectral image data and voiceprint data of the vehicle chassis; wherein, the spectral image data includes visible light images, infrared images, and ultraviolet images; Spatial matching processing is performed on the visible light image, the infrared image, and the ultraviolet image; Feature extraction processing is performed on the visible light image, the infrared image, and the ultraviolet image that have undergone the spatial matching process to determine the image anomaly detection result; Voiceprint features are determined based on the voiceprint data; wherein, the voiceprint features include transient voiceprint features and steady-state voiceprint features; The voiceprint anomaly detection result is determined based on the transient and steady-state characteristics of the voiceprint. The location and type of defects in the vehicle chassis are determined based on the image anomaly detection results and the voiceprint anomaly detection results.

2. The chassis testing method for new energy vehicles according to claim 1, characterized in that, Spatial matching processing is performed on the visible light image, the infrared image, and the ultraviolet image to determine the spectral matching image. Prior to this, the process also includes: The visible light image, the infrared image, and the ultraviolet image are preprocessed.

3. The chassis testing method for new energy vehicles according to claim 1, characterized in that, The spectral matching image is subjected to feature extraction processing to determine the image anomaly detection result, including: Based on Hough line transform, spatial matching processing is performed on the visible light image, the infrared image, and the ultraviolet image.

4. The chassis testing method for new energy vehicles according to claim 1, characterized in that, Feature extraction processing is performed on the visible light image, the infrared image, and the ultraviolet image that have undergone the spatial matching process to determine the image anomaly detection result, including: Feature extraction processing is performed on the visible light image, the infrared image, and the ultraviolet image that have undergone the spatial matching process to determine the feature data map of each type of image; Each of the aforementioned feature data maps is fused to form a spectral fusion feature map; The location and type of defects in the vehicle chassis are determined based on the spectral fusion feature map.

5. The chassis testing method for new energy vehicles according to claim 4, characterized in that, Feature extraction processing is performed on the visible light image, the infrared image, and the ultraviolet image that undergo the spatial matching process to determine the feature data map of each type of image, including: Based on the visible light image undergoing the spatial matching process, visible light feature information is extracted, and a visible light feature map is determined based on the visible light feature information; wherein, the visible light feature information includes visible light morphological features, texture features, and edge features; Based on the infrared image undergoing the spatial matching process, infrared feature information is extracted, and an infrared feature map is determined based on infrared temperature features; wherein, the infrared temperature features include temperature gradient features and temperature statistical features; Based on the ultraviolet image undergoing the spatial matching process, ultraviolet feature information is extracted, and an ultraviolet feature map is determined based on the ultraviolet feature information; wherein, the ultraviolet feature information includes discharge intensity features, ultraviolet morphological features, and spatial features.

6. The chassis testing method for new energy vehicles according to claim 1, characterized in that, Determining voiceprint features based on the voiceprint data includes: The voiceprint data is preprocessed to determine discrete frame voiceprint signals; Based on the discrete frame voiceprint signal, signal extraction processing is performed to determine the transient and steady-state characteristics of the voiceprint.

7. The chassis testing method for new energy vehicles according to claim 6, characterized in that, The voiceprint anomaly detection result is determined based on the transient and steady-state characteristics of the voiceprint, including: The defect category is determined by combining the transient and steady-state characteristics of the voiceprint with the voiceprint fault diagnosis model. The coordinates of the sound source of the defect were determined using the TDOA algorithm; The voiceprint anomaly detection result is determined based on the defect category and the sound source coordinates.

8. The chassis testing method for new energy vehicles according to claim 1, characterized in that, The location and type of defects in the vehicle chassis are determined based on the image anomaly detection results and the voiceprint anomaly detection results, including: The image anomaly detection results and the voiceprint anomaly detection results are subjected to rule-based correlation processing to determine the location and type of defects in the vehicle chassis.

9. The chassis testing method for new energy vehicles according to claim 1, characterized in that, Acquire spectral image data and voiceprint data of the vehicle chassis, including: The spectral image data is acquired using a visible light camera, an infrared thermal imager, and an ultraviolet imager, and the voiceprint data is acquired using a voiceprint sensor.

10. A chassis testing system for new energy vehicles, characterized in that, The chassis testing method for performing any one of the new energy vehicles as described in claims 1-9.

Images