A Method for Identifying Abnormal Noises from Automotive Shock Absorbers Based on Multi-Source Signals and Intelligent Algorithms

Abstract

This invention discloses a method for identifying abnormal noises in shock absorbers based on multi-source signal fusion and intelligent algorithms, relating to the field of automotive shock absorber testing technology. The method includes the following steps: building a test system and determining relevant parameters; deploying sensors and debugging data acquisition equipment; acquiring multi-source signals; preprocessing the signals; extracting signal features; dividing the dataset and labeling it; training and optimizing the intelligent algorithm model; identifying the level of abnormal noise; and verifying the identification results. A recognition model is constructed by optimizing a Support Vector Machine (SVM) using a Genetic Algorithm (GA) to achieve accurate determination of the level of abnormal noise in shock absorbers. This invention combines vibration, noise, and excitation signals from vehicle road tests and bench tests to comprehensively capture abnormal noise characteristics. The extracted WPE and WPSE are complementary, reflecting both signal energy distribution and signal complexity, effectively covering low-, medium-, and high-frequency abnormal noise characteristics.

Description

Technical Field

[0001] This invention relates to the field of automotive shock absorber testing technology, and in particular to a shock absorber abnormal noise identification method based on multi-source signal fusion and intelligent algorithm, which is applicable to mass production quality testing of shock absorbers, after-sales fault diagnosis, and overall vehicle NVH performance optimization in the automotive manufacturing industry. Background Technology

[0002] As a core component of the automotive suspension system, the shock absorber's working condition directly affects the vehicle's ride comfort and NVH (noise, vibration, and harshness) performance. With the widespread adoption of electric vehicles, powertrain noise has been significantly reduced, and abnormal noises such as squeaking and clicking from the shock absorber have become major factors affecting the cabin sound quality.

[0003] Currently, the identification and prediction of abnormal noises from vehicle shock absorbers mainly relies on subjective human evaluation, single signal detection, or traditional simulation analysis methods, which have significant technical shortcomings. Existing technologies mostly use a single vehicle road test signal for abnormal noise prediction. The signal source is singular and easily affected by interference from multiple sources such as the road surface, chassis, and body, making it difficult to accurately extract abnormal noise characteristics from the shock absorber. In terms of feature extraction, only conventional time-domain and frequency-domain features are used, without combining energy distribution and signal complexity for joint characterization. This results in insufficient characterization capabilities for weak, intermittent, and high-frequency abnormal noises, and is prone to feature loss and misjudgment.

[0004] In terms of model construction, most existing technologies employ classification models that have not undergone intelligent optimization, relying on manual experience to adjust model parameters and select features. This results in problems such as poor parameter matching, feature redundancy, and weak generalization ability, making it difficult to meet the requirements of mass production testing in terms of accuracy and stability. Furthermore, current technologies lack a standardized and quantitative system for classifying abnormal noise levels, leading to poor consistency in subjective evaluations and an inability to achieve accurate classification. In addition, traditional methods heavily rely on vehicle road testing, resulting in high testing costs, long cycles, and poor repeatability, making them unsuitable for large-scale mass production testing and rapid after-sales diagnostics. In summary, existing shock absorber abnormal noise identification and prediction technologies suffer from shortcomings such as single signal, insufficient feature representation, lack of model optimization, vague classification standards, and poor engineering applicability, hindering shock absorber quality control and NVH performance improvement. Summary of the Invention

[0005] This invention addresses the difficulties and high costs associated with identifying abnormal noises from vehicle shock absorbers by providing a shock absorber noise identification method based on multi-source signal fusion and intelligent algorithms. Based on the acceleration of the top and bottom of the shock absorber piston rod obtained from road tests, as well as the noise from the driver's seat, Discrete Wavelet Transform (DWT) is used to denoise the acquired signals. Wavelet Packet Transform (WPT) is then used to perform layer decomposition on the denoised vibration signals, resulting in 16 sub-frequency bands. The wavelet packet energy and wavelet packet sample entropy of each frequency band are calculated to form a matrix for training and optimizing the intelligent algorithm. The feature vectors processed from bench tests are input into the trained GA-SVM model to evaluate the abnormal noises. The evaluation results are compared with subjective evaluation results to verify the model's accuracy, thus aiding in the identification of abnormal noises from vehicle shock absorbers.

[0006] The objective of this invention is achieved through the following technical solution: A method for identifying abnormal noises from automotive shock absorbers based on multi-source signals and intelligent algorithms includes the following steps: (1) Set up the test system and determine relevant parameters: For road tests: Select a problem vehicle provided by a certain company as the test object, and prepare multiple shock absorber samples with different abnormal noise levels (half front and half rear shock absorbers), covering 5 levels of abnormal noise; build a vehicle road test platform on several different roads, with a test section length ≥2km, an ambient temperature controlled at 16~24℃, and no sound-reflecting objects within 20m of the test area. For bench tests: build a bench test for shock absorbers, simulate the suspension connection stiffness of the whole vehicle through customized fixtures, and set the corresponding preload. (2) Arrange sensors and debug data acquisition equipment: Arrange a microphone sensor above the driver's seat during the vehicle road test, and install an acceleration sensor at the top and bottom of the shock absorber piston rod. The local coordinate system axis of the sensor is parallel to the coordinate system axis of the fixed coordinate system. During the bench test, install the same type of acceleration sensor on the shock absorber piston rod and the point of action, connect the sensor to the data acquisition system, and communicate with the computer. (3) Collecting multi-source signals: During the vehicle road test, the test vehicle speed is controlled at 15~20km / h, and in-vehicle noise signal, piston rod vibration signal and road excitation signal are collected for 10 seconds. One set of data is collected for each shock absorber sample under each road condition. For the bench test, the road excitation signal of the vehicle road test is processed by the linear averaging method and used as the loading signal. The collected parameters are consistent with those of the vehicle road test. Before the test, the shock absorber is preheated for 3 cycles to ensure that the oil temperature is stable within the normal working range. (4) Signal preprocessing: Discrete wavelet transform (DWT) is used to denoise the acquired signal. The wavelet function selected is db5, the decomposition level is 4, and the soft threshold calculation formula is... (x is the original signal), after denoising, remove the transition segments at the beginning and end of the signal, retain the valid signal in the middle 8 seconds, and then apply the formula... The signal is normalized to the [0,1] interval; (5) Extracting signal features: Wavelet packet transform (WPT) is used to decompose the vibration signal after preprocessing in step (4) into 4 levels, resulting in 16 sub-bands (0~80Hz to 1200~1280Hz), which are then processed according to the formula. (i=1~16) Calculate the energy of each sub-band, and then calculate the energy according to the wavelet packet. Normalization (where At the same time, according to the sample entropy formula (m=2, r=0.2×std, where std is the signal standard deviation) Calculate the sample entropy of each sub-band, then calculate the entropy according to the wavelet packet sample entropy. Normalization (where The 16-dimensional WPE and 16-dimensional WPSE are combined into a 32-dimensional feature matrix; (6) Dataset partitioning and labeling: Multiple samples were divided into training, validation, and test sets according to a certain ratio, and a semantic scoring method was used for subjective evaluation. Thirty professional reviewers aged 20-50 (18 men and 12 women) listened to 5-second noise samples through professional headphones and scored them according to a 5-level standard. The scoring results were verified and used as labels. (7) Training and Optimizing the Intelligent Algorithm Model: A model for identifying abnormal sounds is constructed using a genetic algorithm (GA) to optimize a support vector machine (SVM). Parameters such as population size, crossover rate, and mutation rate in the genetic algorithm are set. The SVM uses a Gaussian radial basis kernel function. The feature subset and model parameters are optimized through GA. The classification accuracy of the validation set is used as the fitness function. The GA-SVM model is trained using the training set until the maximum number of iterations is reached or the fitness meets the requirements. (8) Abnormal noise level identification: Input the feature vector after bench test into the trained GA-SVM model, and output the abnormal noise level of level 1 (extremely severe abnormal noise), level 2 (severe abnormal noise), level 3 (slight abnormal noise), level 4 (weak abnormal noise), and level 5 (no abnormal noise); (9) Verification of recognition results: Calculate the matching degree between the model recognition results and the subjective evaluation labels. When the matching degree is ≥96%, the recognition is deemed valid; otherwise, return to step (7) to re-optimize the model parameters.

[0007] Furthermore, the bench loading signal in step (3) is obtained by processing the road excitation signals of all vehicle road tests using the linear averaging method. The excitation frequency is mainly distributed in the range of 0~100Hz, with the peak value concentrated in 20Hz.

[0008] Furthermore, the threshold determination of WPT in step (5) adopts the Daubechies threshold. The 16 sub-bands after decomposition correspond to 0~80Hz, 80~160Hz, ..., 1200~1280Hz respectively, among which the low-mid frequency band (0~640Hz) is the main distribution range of abnormal noise characteristics.

[0009] Furthermore, after GA optimization in step (7), 15 effective WPSE features (excluding the 16th sub-band) and 9 effective WPE features (mainly concentrated in the low-mid frequency band) are selected. The optimal SVM parameters vary depending on the feature.

[0010] Furthermore, the model training process in step (7) adopts an iterative optimization strategy. When the fitness difference between two adjacent iterations is ≤0.5%, the iteration is stopped.

[0011] Furthermore, the five-level standard for judging the level of abnormal noise in step (8) is as follows: Level 1 (extremely serious abnormal noise) is perceived by all passengers and there is a serious quality problem; Level 2 (serious abnormal noise) is perceived by ordinary passengers and the performance is poor; Level 3 (slight abnormal noise) is perceived by picky passengers and there is a defect; Level 4 (weak abnormal noise) is perceived only by professionals and the performance is good; Level 5 (no abnormal noise) is almost no noise and the performance is optimal.

[0012] The present invention also provides an automotive shock absorber abnormal noise identification system based on multi-source signals and intelligent algorithms.

[0013] The present invention also provides a computer device.

[0014] The present invention also provides a computer-readable storage medium.

[0015] Compared with the prior art, the present invention has the following positive effects: 1) Multi-source signal fusion: Combining vibration, noise, and excitation signals from vehicle road tests and bench tests, it comprehensively captures abnormal noise characteristics, improving the accuracy of single signal detection by more than 15%; 2) Dual feature extraction: WPE and WPSE are complementary, reflecting both the signal energy distribution and the signal complexity, effectively covering low-, medium- and high-frequency abnormal noise features; 3) Quantitative level determination: Abnormal noises are divided into multiple levels to avoid individual differences in subjective evaluation and meet the accuracy requirements of mass production testing; 4) Strong engineering applicability: Bench testing replaces part of the vehicle road testing, reducing testing costs and time, and is suitable for large-scale quality control, providing accurate basis for after-sales fault diagnosis. Attached Figure Description

[0016] Figure 1 This is an overall flowchart of the method of this invention.

[0017] Figure 2 This is a schematic diagram of the road test system in an example of the present invention.

[0018] Figure 3 This is a schematic diagram of the abnormal noise test system for the shock absorber bench in an example of the present invention.

[0019] Figure 4 This is a spectrum diagram of the loaded signal in an embodiment of the present invention.

[0020] Figure 5(a) is a comparison of the vibration signal spectrum of different abnormal noise levels under road test in an embodiment of the present invention.

[0021] Figure 5(b) is a comparison of the vibration signal spectrum of different abnormal noise levels under bench test in an embodiment of the present invention.

[0022] Figure 6 This is a schematic diagram of a 4-layer decomposition tree structure for wavelet packet transform in an embodiment of the present invention.

[0023] Figure 7 This is a statistical chart showing the distribution of subjective evaluation scores in an embodiment of the present invention.

[0024] Figure 8 This is a flowchart of the GA-SVM model optimization process in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the present invention clearer and more explicit, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0026] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for identifying abnormal noises from automotive shock absorbers based on multi-source signals and intelligent algorithms, comprising the following steps: Step 1: Set up the test system and determine relevant parameters: For road tests, select the problem vehicle as the test object and prepare multiple shock absorber samples with different levels of abnormal noise (half front and half rear shock absorbers); set up a vehicle road test platform on several different roads, with a test section length ≥2km, an ambient temperature controlled at 16~24℃, and no sound-reflecting objects within the preset test area; For bench tests: set up a bench test for shock absorbers, simulate the suspension connection stiffness of the whole vehicle through fixtures, and set the corresponding preload.

[0027] In one embodiment, a problematic vehicle provided by a certain company was selected as the test vehicle. This ensured the vehicle had no chassis faults, a stable powertrain, and no other factors interfering with the detection of abnormal noises. 150 shock absorber samples of different noise levels were prepared (75 front and 75 rear shock absorbers). Among these, 12 had extremely severe noises, 28 had severe noises, 45 had slight noises, 35 had very faint noises, and 30 had no noises. A vehicle road test platform was constructed, including five typical road conditions: cobblestone road, Belgian cobblestone road, resonance road, tortuous road, and obstacle road. The test section was 2.5 km long, the ambient temperature was controlled at 20℃, there were no sound-reflecting objects within 20m of the test area, and interference such as engine noise was eliminated. The test vehicle was allowed to automatically coast along a slope for testing. The overall layout of the road test was as follows: Figure 2 As shown; simultaneously, a bench test platform was constructed, using an MTS 850 shock absorber test bench. The lower clamp simulated the suspension connection stiffness of the entire vehicle, with a preload set to 500N. Road surface excitation signals from road tests were input to the upper actuator to simulate the vibration experienced by the shock absorber during real-vehicle testing. The overall layout of the bench test system is as follows. Figure 3 As shown.

[0028] Step 2: Deploy sensors and debug data acquisition equipment: Install a microphone sensor above the driver's seat during the vehicle road test, and install acceleration sensors at the top and bottom of the shock absorber piston rod. The local coordinate system axes of the sensors are parallel to the coordinate system axes of the fixed coordinate system. For bench testing, install the same type of acceleration sensor on the shock absorber piston rod and the point of action. Connect each sensor to the data acquisition system, and the data acquisition system communicates with the computer.

[0029] In one embodiment, the sensor placement and device connection relationship are shown in Figure 2. A GRAS40HF microphone is placed above the driver's seat during the vehicle road test to collect in-vehicle noise signals. A PCB triaxial acceleration sensor is installed at the top and bottom of the shock absorber piston rod. For bench testing, the same type of acceleration sensor is installed on the shock absorber piston rod and the point of action. All sensors are connected to the LMS data acquisition system, which communicates with the computer. The vibration signal sampling frequency is set to 2.5kHz and the noise signal sampling frequency is set to 44.1kHz.

[0030] Step 3: Collect multi-source signals: During the vehicle road test, control the test vehicle speed to 15~20km / h and collect in-vehicle noise signal, piston rod vibration signal and road excitation signal for 10 seconds. Collect one set of data for each shock absorber sample under each road condition. For the bench test, the road excitation signal from the vehicle road test is processed by the linear averaging method and used as the loading signal. The acquisition parameters are consistent with those of the vehicle road test. Before the test, the shock absorber is preheated for 3 cycles to ensure that the oil temperature is stable within the normal operating range.

[0031] In this step, the bench loading signal is obtained by processing the road excitation signals of all vehicle road tests using the linear averaging method.

[0032] In one embodiment, during vehicle road testing, the driver maintains a stable speed of 20 km / h using cruise control to avoid sudden acceleration and deceleration. For each shock absorber sample, signals (including in-vehicle noise, piston rod vibration, and road surface excitation signals) are collected for 10 seconds under five different road conditions to ensure the signals contain complete abnormal noise and impact characteristics. For bench testing, the road surface excitation signals from 150 vehicle road tests are processed using a linear averaging method to obtain the loading signals. The processing formula is as follows: (N=150), where, This indicates that the bench test simulates the road surface excitation signal. Number of road tests For the first The road surface excitation signal for this road test. The spectrum of the load signal for the bench test is as follows: Figure 4 As shown, its excitation frequency is mainly distributed in 0~100Hz, with the peak concentrated in 20Hz; before the test, the shock absorber is preheated for 3 cycles (tension-compression reciprocating motion) to ensure that the oil temperature is stable at 25±3℃, and then the data acquisition system is started to collect vibration signals. The collected parameters are completely consistent with the vehicle road test.

[0033] Step 4: Preprocess the vibration and noise signals.

[0034] In one embodiment, Discrete Wavelet Transform (DWT) is used to denoise the acquired vibration and noise signals. The wavelet function selected is db5, the decomposition level is 4, and the soft threshold calculation formula is as follows: In the formula, The threshold for denoising is set; wavelet coefficients with absolute values ​​greater than this threshold are retained. Original vibration signal The total number of sampling points. For the piston rod vibration signal of a certain level 2 abnormal noise damper, the signal-to-noise ratio before denoising was 18dB, and after denoising, it was increased to 32dB; after denoising, the 1s transition segments at the beginning and end of the signal were removed, retaining the effective signal in the middle 8s; then, according to the formula... In the formula, This represents the result after normalization, where the value range is compressed to [0, 1]. The normalized signal effectively preserves the abnormal noise impact characteristics. Figure 5 shows a comparison of the preprocessed spectra of different abnormal noise levels, allowing for a direct observation of the spectral differences between different levels.

[0035] Step 5: Extract signal features: The denoised vibration signal is decomposed into multiple sub-bands using wavelet packet transform (WPT). The wavelet packet energy of each sub-band is calculated, and the sample entropy of each sub-band is calculated and normalized to obtain the feature vector.

[0036] In one embodiment, wavelet packet transform (WPT) is used to decompose the preprocessed vibration signal into a four-level decomposition, with the decomposition tree structure as follows: Figure 6 As shown ( Figure 6 middle The original signal is represented by L, a low-pass filter by H, and decomposition coefficients (1,0)~(4,15). This results in 16 sub-bands (0~80Hz to 1200~1280Hz). According to the formula... ( =1~16) Calculate the energy of each sub-band, then according to Normalization (where The WPE value of the fourth sub-band (240~320Hz) of a certain level 2 abnormal noise sample is 0.5238, which is the maximum value of all sub-bands; at the same time, according to the sample entropy formula... ( =2, =0.2×std, where std is the signal standard deviation) Calculate the sample entropy of each sub-band, then according to Normalization (where The WPSE value of the 11th sub-band (800~880Hz) of this sample is 0.3231, significantly higher than the corresponding value of the level 5 abnormal noise sample; finally, the 16-dimensional WPE and 16-dimensional WPSE are combined into a 32-dimensional feature vector to complete feature extraction. In the formula, Indicates the first Energy of each wavelet packet frequency band Represents the wavelet packet energy; Indicates the first Wavelet packet coefficients corresponding to the frequency bands of wavelet packet decomposition; is the overall normalization coefficient, which is the L2 norm of the energy of the 16 wavelet packet frequency bands; SE is the sample entropy, used to quantify the fault information content of the vibration signal of the vibration damper; The embedding dimension represents the vector length when constructing the phase space vector; The distance threshold used to determine whether two subsequences are "similar" is the similarity tolerance; std represents the standard deviation of the sequence. This represents the total number of data points. Indicates length is The percentage of similar matching pairs in a subsequence; This is the final normalized wavelet packet sample entropy feature vector.

[0037] In this embodiment, the threshold determination of WPT adopts the Daubechies threshold. The 16 sub-bands after decomposition correspond to 0~80Hz, 80~160Hz, ..., 1200~1280Hz respectively, among which the low-mid frequency band (0~640Hz) is the main distribution range of abnormal noise characteristics.

[0038] Step 6: Divide and label the dataset: Divide multiple shock absorber samples into training, validation, and test sets according to a preset ratio, and use semantic scoring for subjective evaluation. Professional reviewers will listen to the noise samples using professional headphones and score them; the scores will be validated and used as labels.

[0039] In one embodiment, 150 shock absorber samples were divided into a training set (70 samples), a validation set (40 samples), and a test set (40 samples) in a 7:4:4 ratio to ensure that samples of each level were evenly distributed across the three datasets. A semantic scoring method was used for subjective evaluation. Thirty professional reviewers aged 20-50 (18 men and 12 women) listened to a 5-second noise sample using Sennheiser HD800 headphones and scored it according to a 5-level standard. The scoring results were verified using the Kendall concordance coefficient (W=0.85≥0.8, where W is the consistency coefficient, used to measure the degree of consistency in scoring the same batch of samples by multiple reviewers). The distribution of subjective evaluation scores is shown in Figure 7. Figure 7 The horizontal axis represents the sample number, the vertical axis represents the average score, and the error bars represent the standard deviation. The scores after successful verification are used as the training labels for the model.

[0040] Step 7: Training and Optimizing the Intelligent Algorithm Model: A GA-SVM model for identifying abnormal sounds is constructed using a Genetic Algorithm (GA) to optimize the Support Vector Machine (SVM). Parameters such as population size, crossover rate, and mutation rate in the Genetic Algorithm are set. The SVM uses a Gaussian radial basis function kernel. The feature subset and model parameters are optimized through GA. The fitness function is set to the classification accuracy on the validation set. The GA-SVM model is trained using the training set until the maximum number of iterations is reached or the fitness requirement is met.

[0041] In one embodiment, the genetic algorithm (GA) parameters are set as follows: population size 100, crossover rate 0.5, mutation rate 0.05, number of iterations 200, chromosome length 36 bits (12-bit encoded C parameter, 8-bit encoded γ parameter, 16-bit encoded feature selection). The model optimization process is as follows: Figure 8As shown, the SVM uses a Gaussian radial basis function kernel and optimizes the feature subset and model parameters through a genetic algorithm (GA). The validation set classification accuracy is used as the fitness function. The GA-SVM model is trained on the training set using an iterative optimization strategy. Iteration stops when the fitness difference between two adjacent iterations is ≤0.5%. Ultimately, 15 effective WPSE features (excluding the 16th sub-band) and 9 effective WPE features (mainly concentrated in the low-to-mid frequency band) are selected. The optimal SVM parameters are C=5.61, γ=0.78 (WPSE features) and C=77.62, γ=0.39 (WPE features). After model training, the validation set accuracy reaches 97.5%, and the running time is 19.68 seconds. It is understandable that the optimal SVM parameters vary depending on the features.

[0042] Step 8: Abnormal noise level identification: Input the feature vector after bench test processing into the trained GA-SVM model, and output the abnormal noise level as follows: Level 1 (extremely severe abnormal noise), Level 2 (severe abnormal noise), Level 3 (slight abnormal noise), Level 4 (weak abnormal noise), and Level 5 (no abnormal noise).

[0043] In one embodiment, the five-level standard for judging the level of abnormal noise is as follows: Level 1 (extremely serious abnormal noise) is perceived by all passengers and there is a serious quality problem; Level 2 (serious abnormal noise) is perceived by ordinary passengers and the performance is poor; Level 3 (slight abnormal noise) is perceived by picky passengers and there is a defect; Level 4 (weak abnormal noise) is perceived only by professionals and the performance is good; Level 5 (no abnormal noise) is almost no noise and the performance is optimal.

[0044] In one embodiment, the 32-dimensional feature vectors (16-dimensional WPE and 16-dimensional WPSE in step 5) collected from bench tests and preprocessed and feature extracted are input into the trained GA-SVM model. The GA-SVM model outputs the abnormal noise levels of Level 1 (extremely severe abnormal noise), Level 2 (severe abnormal noise), Level 3 (slight abnormal noise), Level 4 (weak abnormal noise), and Level 5 (no abnormal noise) based on the matching degree between the feature vectors and the preset level feature templates. During the recognition process, the analysis of the input feature vectors by the GA-SVM model is related to the spectral features in Figure 5. For example, when the Level 1 abnormal noise is output, the corresponding feature vector is highly matched with the Level 1 abnormal noise spectral features in Figure 5(a), ensuring the reliability of the recognition results.

[0045] (9) Verification of recognition results: Calculate the matching degree between the model recognition results and the subjective evaluation labels. When the matching degree reaches the preset value, the recognition is deemed valid; otherwise, return to step (7) to re-optimize the model parameters.

[0046] In one embodiment, 50 test samples were selected for verification. The initial recognition accuracy was 96.8%, and 3 samples were misclassified (2 level 3 samples were misclassified as level 4, and 1 level 4 sample was misclassified as level 3). After a second signal acquisition and feature re-extraction, the corrected matching degree reached 98% ≥ 96%, and the recognition was deemed effective. After the misclassified samples were corrected, all recognition results were completely consistent with the actual abnormal noise level, verifying the accuracy of the method of the present invention.

[0047] In one embodiment, a vehicle shock absorber abnormal noise identification system based on multi-source signals and intelligent algorithms is provided to implement steps 3 to 9 of the aforementioned method. The system includes the following modules: The data acquisition module is used to collect in-vehicle noise signals, piston rod vibration signals and road excitation signals during vehicle road tests, and to collect in-vehicle noise signals and piston rod vibration signals during bench tests with the road excitation signals as loading signals. The preprocessing module is used to preprocess vibration and noise signals; The signal feature module is used to perform layer decomposition on the preprocessed vibration signal using wavelet packet transform to obtain multiple sub-bands, calculate the sub-band energy of each sub-band and normalize it to obtain the wavelet packet energy, calculate the sample entropy of each sub-band and normalize it to obtain the wavelet packet sample entropy, and obtain the feature vector based on the wavelet packet energy and wavelet packet sample entropy. The scoring module uses semantic scoring to subjectively score each shock absorber sample, and the scoring results are verified and used as labels. The training module is used to optimize the support vector machine using a genetic algorithm to obtain a GA-SVM model for identifying abnormal noises; The recognition module is used to input the feature vectors into the trained GA-SVM model and output the abnormal sound level results.

[0048] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform steps 3 to 9.

[0049] In one embodiment, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements steps 3 to 9 of the aforementioned method.

[0050] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined in this invention may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying abnormal sound of a car shock absorber based on multi-source signals and intelligent algorithms, characterized in that, Includes the following steps: For road tests, select the problematic vehicle as the test object and prepare multiple shock absorber samples with different levels of abnormal noise; build a vehicle road test platform for several different roads; for bench tests, build a shock absorber test bench, simulate the suspension connection stiffness of the whole vehicle through fixtures, and set the corresponding preload. A microphone sensor is placed above the driver's seat during the vehicle road test, and an acceleration sensor is installed at the top and bottom of the shock absorber piston rod. The local coordinate system axes of the sensors are parallel to the fixed coordinate system axes. During the bench test, the same type of acceleration sensor is installed on the shock absorber piston rod and the point of action. Each sensor is connected to the data acquisition system, and the data acquisition system communicates with the computer. During the vehicle road test, in-vehicle noise signals, piston rod vibration signals, and road excitation signals are collected for a preset duration at a preset test speed. Data is collected for each shock absorber sample under each road condition. During the bench test, the road excitation signals from the vehicle road test are processed and used as loading signals, and the collected parameters are consistent with those of the vehicle road test. Preprocessing of vibration and noise signals; Wavelet packet transform is used to perform layer decomposition on the preprocessed vibration signal to obtain multiple sub-bands. The energy of each sub-band is calculated and normalized to obtain the wavelet packet energy. The sample entropy of each sub-band is calculated and normalized to obtain the wavelet packet sample entropy. The feature vector is obtained based on the wavelet packet energy and wavelet packet sample entropy. Multiple shock absorber samples were divided into training set, validation set and test set according to a preset ratio. The semantic scoring method was used to subjectively score each shock absorber sample. The scoring results were verified and used as labels. A genetic algorithm was used to optimize the support vector machine, resulting in a GA-SVM model for identifying abnormal noises. Input the feature vectors into the trained GA-SVM model and output the abnormal sound level results.

2. The method according to claim 1, characterized in that, Before the test, the shock absorber was preheated by circulation to ensure that the oil temperature was stable within the normal operating range.

3. The method according to claim 1, characterized in that, The preprocessing of vibration and noise signals includes: Discrete wavelet transform is used to denoise the acquired signal. After denoising, the transition segments at the beginning and end of the signal are removed, and the effective signal of the preset length in the middle is retained. The signal is then standardized.

4. The method according to claim 3, characterized in that, The wavelet function is db5, the decomposition layer is 4 layers, and the soft threshold calculation formula , is the threshold value of denoising, is the total sampling point number of the original vibration signal .

5. The method of claim 1, wherein the method is characterized by, The calculation of each sub-band sub-band energy and normalization to get wavelet packet energy, by formula The calculation of sub-band energy, by formula Normalization, The energy of the first Wavelet packet band; The first Wavelet packet decomposition band corresponding to the wavelet packet coefficient; The overall normalization coefficient, Wavelet packet energy.

6. The method of claim 1, wherein the method is characterized by, In the process of calculating and normalizing the entropy of each sub-band sample to obtain the wavelet packet sample entropy, the formula is used... Calculate the entropy of each sub-band sample, by Normalize, For sample entropy, For the embedding dimension, For similarity tolerance, Total number of data points Indicates length is The percentage of similar matching pairs in a subsequence. This is the final normalized wavelet packet sample entropy feature vector.

7. A method for identifying abnormal noises in automotive shock absorbers based on multi-source signals and intelligent algorithms according to any one of claims 1-6, characterized in that, It also includes the following steps: The matching degree between the model recognition result and the subjective evaluation label is calculated. When the matching degree reaches the preset value, the recognition is deemed valid; otherwise, the model parameters are re-optimized.

8. A vehicle shock absorber abnormal noise identification system based on multi-source signals and intelligent algorithms, characterized in that, Includes the following modules: The data acquisition module is used to collect in-vehicle noise signals, piston rod vibration signals and road excitation signals during vehicle road tests, and to collect in-vehicle noise signals and piston rod vibration signals during bench tests with the road excitation signals as loading signals. The preprocessing module is used to preprocess vibration and noise signals; The signal feature module is used to perform layer decomposition on the preprocessed vibration signal using wavelet packet transform to obtain multiple sub-bands, calculate the sub-band energy of each sub-band and normalize it to obtain the wavelet packet energy, calculate the sample entropy of each sub-band and normalize it to obtain the wavelet packet sample entropy, and obtain the feature vector based on the wavelet packet energy and wavelet packet sample entropy. The scoring module uses semantic scoring to subjectively score each shock absorber sample, and the scoring results are verified and used as labels. The training module is used to optimize the support vector machine using a genetic algorithm to obtain a GA-SVM model for identifying abnormal noises; The recognition module is used to input the feature vectors into the trained GA-SVM model and output the abnormal sound level results.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method steps of each module in claim 8.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. When the computer program is executed by the processor, it implements the method steps of each module in claim 8.

Images