New energy automobile motor abnormal sound detection method and system based on decision level fusion
By adopting a decision-making fusion method in new energy vehicle motors, combining multiple audio features and support vector machine classifiers, and using the fusion results of D-S evidence theory, the problem of insufficient accuracy caused by a single feature and classifier is solved, and abnormal fault detection is improved accuracy and robustness.
Patent Information
- Application Number
- CN202411592229.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-08
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the prior art, the dependence of audio classification on single features and classifiers leads to insufficient classification accuracy and robustness of motor abnormality detection of new energy vehicle motors, and mainly rely on manual methods and low accuracy.
Using a decision-level fusion method, a variety of audio features (such as STFT, MFCC and Mel spectrum) are extracted by collecting audio signal data of new energy vehicle motors, and inputting these features into multiple support vector machine classifiers for training, using D-S Evidence theory fusion classification results are used to determine whether the sound signal is normal.
It improves the recognition accuracy and reliability of abnormal sounds of motors of new energy vehicles, enhances the detection and recognition capabilities of the system in complex environments, and ensures the comprehensiveness and accuracy of abnormal sound detection.
Smart Images

Figure CN119091910A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of industrial intelligent diagnosis and quality control, and in particular to a method and system for detecting abnormal noise of a new energy vehicle motor based on decision-level fusion. Background Art
[0002] In recent years, with the development of artificial intelligence, industrial abnormal noise fault detection and diagnosis methods based on artificial intelligence algorithms have become a research hotspot in the manufacturing production process. With the continuous evolution of machine learning and deep learning technologies, their popularity in industrial applications has continued to increase, greatly promoting the dual improvement of production efficiency and product quality. In the manufacturing scenario, machine learning technology provides a more accurate solution for quality control by virtue of its ability to efficiently extract key information from complex data.
[0003] Abnormal sound, that is, the abnormal sound made by the equipment during normal operation, is usually regarded as an early warning signal of potential equipment problems, which may involve various fault conditions such as part wear, poor lubrication, loose components, etc. Given the potential impact of abnormal sound on equipment safety, cost-effectiveness and operating efficiency, abnormal sound detection occupies a pivotal position in the quality control process.
[0004] In the field of audio classification, existing technologies mainly rely on single features and classifiers, which not only limits the improvement of classification accuracy and robustness, but also lacks an effective information fusion mechanism, resulting in unsatisfactory detection results. Especially in the detection of abnormal faults in the motor system of new energy vehicles, since the motor system is composed of many complex components and subsystems, the complex interactions between them make fault detection particularly difficult. At present, the detection of abnormal motor faults mainly relies on manual methods, and faults are located through means such as hearing, vision and touch, but this method has limited detection accuracy and is difficult to meet the needs of the rapid development of modern automotive technology. Summary of the invention
[0005] To this end, an embodiment of the present invention provides a method and system for detecting abnormal noise of new energy vehicle motors based on decision-level fusion, which is used to solve the problems in the prior art that audio classification relies on a single feature and classifier, resulting in insufficient classification accuracy and robustness, and that abnormal fault detection of new energy vehicle motors mainly relies on manual methods and has low accuracy.
[0006] In order to solve the above problems, an embodiment of the present invention provides a new energy vehicle motor abnormal noise detection method based on decision-level fusion, the method comprising: Step S1: collecting audio signal data of a new energy vehicle motor in various working states, annotating the audio signal data to distinguish normal sounds from abnormal sounds, and constructing a data set based on the annotated data; Step S2: preprocessing the audio signals in the data set, and extracting multiple audio features from the preprocessed audio signals; Step S3: inputting the extracted multiple audio features into multiple independent support vector machine classifiers for training, thereby obtaining multiple trained support vector machine classifiers; Step S4: multiple trained support vector machine classifiers independently identify and classify new audio signals based on different audio features to generate classification results, and convert the classification results of each support vector machine classifier into a basic probability distribution evidence body; Step S5: Fusing the basic probability distribution evidence bodies from different support vector machine classifiers to obtain a fusion result; Step S6: judging whether the sound signal of the new energy vehicle motor is normal according to the fusion result.
[0007] Preferably, the method for preprocessing the audio signal in the data set includes removing background noise and normalizing processing.
[0008] Preferably, the method for extracting multiple audio features from the preprocessed audio signal is: The audio features of the preprocessed audio signal are extracted using short-time Fourier transform, Mel-frequency cepstrum coefficient and Mel-spectrum respectively, including STFT features, MFCC features and Mel-spectrum features, where STFT represents short-time Fourier transform, MFCC represents Mel-frequency cepstrum coefficient and Mel represents Mel.
[0009] Preferably, the method of extracting the audio features of the preprocessed audio signal using short-time Fourier transform is: Given a discrete-time signal , split it into a series of length There is a certain overlap between each frame to improve the time resolution; a window function is applied to each frame signal , to reduce the edge effect; the signal after windowing is expressed as: ; In the formula, is the mth frame signal; It is frame shift; is the index of the time domain sample; For each windowed frame signal Perform a fast Fourier transform and calculate its frequency components: ; In the formula, It is The frequency content of the frame, indicating the frequency The complex value at contains amplitude and phase information; is an imaginary unit; Calculate the power spectral density of each frame or Amplitude Spectrum , the calculation formula is: ; ; Arrange the spectrum results of each frame in time order to form a two-dimensional matrix , where each element represents the and frequency Signal strength on: .
[0010] Preferably, the method of extracting audio features of the preprocessed audio signal using Mel-frequency cepstral coefficients is: First, the original audio signal is preprocessed, including frame segmentation and windowing. Then, the preprocessed data is subjected to fast Fourier transform to obtain the energy spectrum. Then, the energy spectrum is filtered using multiple Mel filters, and the logarithm of the filtered energy is taken to calculate the logarithmic energy. Finally, the logarithmic energy is subjected to discrete cosine transform to extract the MFCC features.
[0011] Preferably, the method of extracting audio features of the preprocessed audio signal using Mel spectrum is: Divide the continuous audio signal into multiple short-time frames and perform windowing on each frame signal; Perform fast Fourier transform on each frame of windowed signal to convert the time domain signal into a frequency domain signal to obtain the spectrum representation of the frame; Square the spectrum obtained by fast Fourier transform and calculate the power spectrum of each frequency component; Process the power spectrum through a set of Mel filters; Perform logarithmic operation on the energy output by the Mel filter bank to obtain the logarithmic energy spectrum.
[0012] Preferably, the method of inputting the extracted multiple audio features into multiple independent support vector machine classifiers for training to obtain multiple trained support vector machine classifiers is: The extracted audio features are input into multiple independent support vector machine classifiers for training. During the training process, the performance of each support vector machine classifier is evaluated using five-fold cross validation to optimize its parameters, and finally multiple trained support vector machine classifiers are obtained.
[0013] Preferably, the method for fusing basic probability distribution evidence bodies from different support vector machine classifiers is: Using DS evidence theory, the basic probability distribution evidence bodies from different support vector machine classifiers are fused; For two support vector machine classifiers and BPA, where BPA represents the basic probability assignment, the fusion The definition is as follows: ; In the formula, Indicates fusion ; It's the complete set Any subset of ; and Support vector machine classifier and BPA in the denominator is the conflict coefficient, which indicates the degree to which two SVM classifiers support completely different categories. This term is used for normalization to ensure that the sum of BPA is 1; For the fusion of multiple support vector machine classifiers, the recursive characteristics of the DS evidence theory rules are used; for three support vector machine classifiers , first merge get Then with Fusion; and so on, to achieve the fusion of any number of support vector machine classifiers.
[0014] Preferably, the basic probability distribution is used to represent the support degree of a certain support vector machine classifier for different categories, and the support quantity of each category is quantified as a probability value; it reflects the trust of the support vector machine classifier in the category, and is mapped from the power set of the entire set to a numerical value between [0,1]; probability distribution is performed for each possible category and its combination, and the undistributed part is used to represent uncertainty, and the sum of the basic probability distribution is 1.
[0015] The embodiment of the present invention further provides a new energy vehicle motor abnormal sound detection system based on decision-level fusion, which is used to implement the above-mentioned new energy vehicle motor abnormal sound detection method based on decision-level fusion, and specifically includes: The audio signal acquisition and data set construction module is used to collect audio signal data of new energy vehicle motors in various working states, annotate the audio signal data to distinguish normal sounds from abnormal sounds, and construct a data set based on the annotated data; The data preprocessing and feature extraction module is used to preprocess the audio signals in the data set and extract multiple audio features from the preprocessed audio signals; A model training module is used to input the extracted multiple audio features into multiple independent support vector machine classifiers for training, thereby obtaining multiple trained support vector machine classifiers; The multi-classifier independent identification and BPA calculation module is used to independently identify and classify new audio signals based on different audio features using multiple trained support vector machine classifiers, generate classification results, and convert the classification results of each support vector machine classifier into a basic probability distribution evidence body; The BPA fusion module is used to fuse the basic probability allocation evidence bodies from different support vector machine classifiers to obtain the fusion result; The detection result judgment module is used to judge whether the sound signal of the new energy vehicle motor is normal based on the fusion result.
[0016] It can be seen from the above technical solutions that the present invention has the following beneficial effects: (1) The present invention overcomes the problem of insufficient recognition accuracy of a single feature in a complex motor sound environment by combining multiple audio features and independent classifiers, thereby ensuring accurate classification of normal and abnormal motor sounds.
[0017] (2) The present invention introduces the DS evidence theory and integrates the classification results from different features to improve the reliability of abnormal sound detection and accurately identify abnormal sounds during motor operation.
[0018] (3) The present invention enhances the system's ability to detect and identify new or unseen sound signals in the complex environment of new energy vehicle motors through comprehensive analysis of multiple features and multiple classifiers, thereby ensuring the comprehensiveness and accuracy of abnormal sound detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] In order to more clearly illustrate the implementation cases of the present invention or the technical solutions in the prior art, the following is a brief description of the drawings required for use in the embodiments. By referring to the drawings, the features and advantages of the present invention will be more clearly understood. The drawings are schematic and should not be understood as limiting the present invention in any way. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them: Figure 1 A flowchart of a method for detecting abnormal noise of a new energy vehicle motor based on decision-level fusion provided in an embodiment; Figure 2 It is an architecture diagram of abnormal noise detection of new energy vehicle motor based on decision-level fusion in the embodiment; Figure 3 It is a schematic diagram of multi-feature extraction spectrum in the embodiment; Figure 4 A block diagram of a new energy vehicle motor abnormal noise detection system based on decision-level fusion provided in an embodiment. DETAILED DESCRIPTION
[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Embodiment 1
[0021] In order to solve the problem that the audio classification in the existing technology relies on a single feature and classifier, resulting in insufficient classification accuracy and robustness, and the abnormal fault detection of new energy vehicle motors mainly relies on manual methods with low accuracy. Figure 1 As shown, an embodiment of the present invention proposes a new energy vehicle motor abnormal noise detection method based on decision-level fusion, the method comprising: Step S1: collecting audio signal data of a new energy vehicle motor in various working states, annotating the audio signal data to distinguish normal sounds from abnormal sounds, and constructing a data set based on the annotated data; Step S2: preprocessing the audio signals in the data set, and extracting multiple audio features from the preprocessed audio signals; Step S3: inputting the extracted multiple audio features into multiple independent support vector machine classifiers for training, thereby obtaining multiple trained support vector machine classifiers; Step S4: multiple trained support vector machine classifiers independently identify and classify new audio signals based on different audio features to generate classification results, and convert the classification results of each support vector machine classifier into a basic probability distribution evidence body; Step S5: Fusing the basic probability distribution evidence bodies from different support vector machine classifiers to obtain a fusion result; Step S6: judging whether the sound signal of the new energy vehicle motor is normal according to the fusion result.
[0022] It can be seen from the above technical scheme that the present invention provides a new energy vehicle motor abnormal sound detection method based on decision-level fusion, by collecting and labeling the audio signal data of the new energy vehicle motor to build a data set, extracting multiple audio features such as STFT, MFCC and Mel spectrum after pre-processing, respectively inputting multiple independent SVM classifiers for training, and fusion of the classification results of each classifier based on DS evidence theory. This method effectively overcomes the problem of insufficient classification accuracy and robustness caused by single features and classifiers, improves the recognition accuracy and reliability of abnormal sounds of new energy vehicle motors, enhances the detection and recognition capabilities of the system in complex environments, and is particularly suitable for processing complex or high uncertainty working conditions.
[0023] like Figure 2 The architecture diagram of abnormal noise detection of new energy vehicle motors based on decision-level fusion is shown, which is used to identify whether the sound of the motor is normal under different working conditions.
[0024] In step S1, professional audio acquisition equipment is used to collect audio signal data of new energy vehicle motors in various working states, and the audio signal data is annotated to distinguish normal sounds (OK) from abnormal sounds (NG). A data set is constructed based on the annotated data, and the constructed data set is divided into a training set, a validation set, and a test set to ensure that subsequent models can be effectively trained, validated, and tested on different data sets.
[0025] In step S2, the audio signals in the data set are preprocessed, including background noise removal and normalization, to improve data quality. A variety of audio features are extracted from the preprocessed audio signals.
[0026] Specifically, Figure 3 As shown, in the multi-feature extraction process, the present invention simultaneously uses three different methods to extract useful feature information from the preprocessed audio signal: short-time Fourier transform (STFT), Mel-frequency cepstral coefficients (MFCC) and Mel-spectrum (Mel-spectrum). Among them, short-time Fourier transform (STFT) is used to obtain the time-frequency domain representation of the audio, showing the frequency changes of the signal in different time periods. Mel-frequency cepstral coefficients (MFCC) extract more compact audio features, reflecting the spectral shape of the audio signal. Mel Spectrogram extracts frequency-related features by simulating the auditory characteristics of the human ear. These feature information will serve as the basis for subsequent classifier training and provide rich feature data for the model.
[0027] Furthermore, the Short-Time Fourier Transform (STFT) is a variant of the Fourier transform, which is used to analyze the time-frequency characteristics of the signal. The STFT slides a window over the signal and performs a Fourier transform on the signal within the window to obtain the spectrum information of the signal changing over time. The process of extracting the Short-Time Fourier Transform (STFT) features can be summarized as the following steps: (1) Given a discrete-time signal , split it into a series of length There is a certain overlap between each frame to improve the time resolution; a window function is applied to each frame signal (such as Hanning window or Hamming window) to reduce edge effects; the signal after windowing is expressed as: ; In the formula, is the mth frame signal; is the frame shift (i.e., the amount of overlap between adjacent frames); is the index of the time domain samples.
[0028] (2) For each windowed frame signal Perform a fast Fourier transform (FFT) to calculate its frequency components: ; In the formula, It is The frequency content of the frame, indicating the frequency The complex value at contains amplitude and phase information; Is an imaginary unit.
[0029] (3) Calculate the power spectral density of each frame or Amplitude Spectrum , the calculation formula is: ; .
[0030] (4) Arrange the spectrum results of each frame in chronological order to form a two-dimensional matrix , to represent the time-frequency characteristics of the signal, where each element represents the and frequency Signal strength on: .
[0031] Furthermore, Mel-Frequency Cepstral Coefficients (MFCC) is a method for audio feature extraction, which obtains feature coefficients by framing and windowing the audio signal, performing a fast Fourier transform (FFT), filtering it with a Mel filter bank, taking the logarithmic energy, and applying a discrete cosine transform (DCT).
[0032] Mel-frequency cepstral coefficients (MFCC) convert frequency domain information into cepstral domain information after a series of signal processing steps on the audio signal: ; In the formula, It is cepstral coefficients; It is through The power spectrum of a Mel filter; is the number of Mel filters; is the index of the MFCC coefficients.
[0033] Specifically, first, the original audio signal is preprocessed, including framing and windowing; then, the preprocessed data is subjected to a fast Fourier transform to obtain an energy spectrum; then, the energy spectrum is filtered using multiple Mel filters, and the logarithm of the filtered energy is taken to calculate the logarithmic energy; finally, the logarithmic energy is subjected to a discrete cosine transform to extract the MFCC features. This method effectively simulates the auditory perception of the human ear by capturing the frequency distribution characteristics, and ultimately obtains a set of characteristic coefficients to represent the overall characteristics of the audio signal. MFCC features are widely used in speech and audio signal processing because they can effectively reflect the acoustic characteristics of audio.
[0034] Furthermore, Mel Spectrogram (Mel spectrum) is an audio signal processing method whose main purpose is to imitate the human ear's perception characteristics of sounds of different frequencies. Human hearing is more sensitive to low-frequency sounds and less sensitive to high-frequency sounds. In order to reflect this perception characteristic, the Mel scale uses a nonlinear frequency scale, which has a higher resolution in the low-frequency range and a relatively low resolution in the high-frequency range. This frequency conversion is closer to the auditory characteristics of the human ear. The conversion relationship between Mel frequency and traditional Hertz (Hz) frequency can be expressed by the following formula: ; In the formula, represents the Mel frequency; is the frequency in Hertz (Hz). This formula demonstrates the nonlinear conversion characteristics of frequency, which enables the Mel spectrum to more accurately capture the audio characteristics perceived by the human auditory system.
[0035] Specifically, using Mel spectrum to extract audio features of the preprocessed audio signal includes the following steps: (1) Split the continuous audio signal into multiple short-time frames, with each frame usually 20-40 milliseconds long; perform windowing on each frame. Common window functions include Hanning window or Hamming window to reduce the boundary effect generated during the frequency conversion process.
[0036] (2) Perform a fast Fourier transform (FFT) on the windowed signal of each frame. FFT converts the time domain signal into a frequency domain signal, thereby obtaining the spectrum representation of the frame.
[0037] (3) Square the spectrum obtained by FFT and calculate the power spectrum of each frequency component. The purpose of this step is to quantify the energy distribution of the signal at different frequencies.
[0038] (4) The power spectrum is processed through a set of Mel filters. The Mel filter bank is usually composed of a series of triangular filters whose center frequencies and bandwidths are distributed according to the Mel scale. The Mel scale is a frequency scale that simulates the auditory perception of the human ear, with the low frequency part being finer and the high frequency part being coarser.
[0039] (5) Perform a logarithmic operation on the energy output by the Mel filter bank to obtain a logarithmic energy spectrum. The logarithmic operation simulates the human ear's perception characteristics of sounds of different intensities.
[0040] In step S3, the extracted audio features are respectively input into multiple independent support vector machine (SVM) classifiers for training, corresponding to SVM1, SVM2 and SVM3 respectively. During the training process, the performance of each SVM classifier is evaluated using five-fold cross validation to optimize its parameters, and finally multiple trained SVM classifiers are obtained.
[0041] Each SVM classifier is specially trained to identify the normal or abnormal state of the audio signal based on its own feature extraction method. After training, the output of each classifier is converted into a basic probability assignment (BPA) evidence body, corresponding to BPA evidence body 1, BPA evidence body 2, and BPA evidence body 3. These BPA evidence bodies are used to quantify the SVM classifier's confidence that a specific audio signal belongs to the normal or abnormal category, thereby providing a basis for further decision fusion.
[0042] In step S4, multiple trained support vector machine classifiers independently identify and classify new audio signals based on different audio features to generate classification results. The classification results of each SVM classifier are converted into a basic probability distribution evidence body to quantify the trust of each SVM classifier for different categories, providing a basis for subsequent fusion.
[0043] In step S5, the BPA evidence bodies from different SVM classifiers are fused to obtain a fusion result.
[0044] Specifically, the present invention utilizes the DS evidence theory to fuse the BPA evidence bodies from different SVM classifiers, thereby improving the accuracy and reliability of classification decisions and ensuring that the final fusion result can comprehensively reflect the judgment of each SVM classifier.
[0045] Basic probability assignment (BPA) is a core concept in Dempster-Shafer (DS) evidence theory. It is used to represent the support degree of a support vector machine classifier for different categories. It quantifies the support of each category into a probability value, reflecting the trust of the support vector machine classifier in the category. The BPA function maps from the power set of the entire set to a value between [0,1]; it assigns probabilities to each possible category and its combination. The unassigned part is used to represent uncertainty, and the sum of the basic probability assignments is 1. This representation can effectively handle classification uncertainty and provide a basis for subsequent evidence fusion.
[0046] DS evidence theory is a mathematical method for dealing with uncertainty and conflicting information. It generates a more accurate and comprehensive judgment result by integrating the judgment information of multiple classifiers. Through this method, the system can effectively utilize the advantages of each classifier and make up for the limitations of a single classifier, thereby improving the classification performance of complex audio signals.
[0047] The core of DS evidence theory is to fuse the BPA evidence of multiple classifiers to obtain more reliable judgments. The fusion rule is implemented by Dempster's Rule of Combination. For example, for two SVM classifiers and BPA, where BPA represents the basic probability assignment, the fusion The definition is as follows: ; In the formula, Indicates fusion ; It's the complete set Any subset of ; and Support vector machine classifier and BPA in the denominator is the conflict coefficient, which indicates the degree to which two SVM classifiers support completely different categories. This term is used for normalization to ensure that the sum of BPA is 1; For the fusion of multiple SVM classifiers, the recursive characteristics of the DS evidence theory rules are used; for three SVM classifiers , first merge get Then with Fusion; and so on, to achieve the fusion of any number of support vector machine classifiers.
[0048] In step S6, whether the sound signal of the new energy vehicle motor is normal is determined based on the fusion result.
[0049] In order to further illustrate the advantages of the present invention, it is described below in conjunction with specific experiments.
[0050] In the experiment, a method based on multi-feature extraction and multi-classifier fusion (the method of the present invention) was used to detect abnormal sounds of new energy vehicle motors. Through three different feature extraction methods, STFT, Mel spectrum and MFCC, the multi-dimensional feature information of the audio signal is captured, so that the classifier can more comprehensively identify normal and abnormal sounds. The classification result of each feature is generated by an independent SVM classifier and fused using DS theory. In order to verify the effect of multi-feature fusion detection and single feature, an ablation experiment was carried out to test the detection performance of using only a single feature (STFT, MFCC or Mel spectrum) and using all feature fusion strategies. The test results are shown in Table 1. When any one feature is used alone, the accuracy of the model is lower than that of the multi-feature fusion strategy, and the classification accuracy of the multi-feature fusion strategy reaches 95.6%. The experimental results show that multi-feature fusion can more comprehensively capture abnormal features in audio signals and significantly improve the accuracy and reliability of motor audio anomaly detection.
[0051] Table 1 Ablation experiment Embodiment 2
[0052] like Figure 4 As shown, the present invention provides a new energy vehicle motor abnormal sound detection system based on decision-level fusion, which is used to implement the new energy vehicle motor abnormal sound detection method based on decision-level fusion of the above-mentioned embodiment 1, and specifically includes: The audio signal acquisition and data set construction module 100 is used to collect audio signal data of the new energy vehicle motor in various working states, and to mark the audio signal data to distinguish normal sounds from abnormal sounds, and to construct a data set based on the marked data; The data preprocessing and feature extraction module 200 is used to preprocess the audio signal in the data set and extract multiple audio features from the preprocessed audio signal; The model training module 300 is used to input the extracted multiple audio features into multiple independent support vector machine classifiers for training, thereby obtaining multiple trained support vector machine classifiers; The multi-classifier independent identification and BPA calculation module 400 is used for independently identifying and classifying new audio signals based on different audio features using multiple trained support vector machine classifiers, generating classification results, and converting the classification results of each support vector machine classifier into basic probability distribution; The BPA fusion module 500 is used to fuse the basic probability distributions from different support vector machine classifiers to obtain a fusion result; The detection result judgment module 600 is used to judge whether the sound signal of the new energy vehicle motor is normal according to the fusion result.
[0053] A new energy vehicle motor abnormal sound detection system based on decision-level fusion in this embodiment is used to implement the above-mentioned new energy vehicle motor abnormal sound detection method based on decision-level fusion. Therefore, the specific implementation method of the new energy vehicle motor abnormal sound detection system based on decision-level fusion can be seen in the embodiment part of the new energy vehicle motor abnormal sound detection method based on decision-level fusion in the previous text, for example, the audio signal acquisition and data set construction module 100, the data preprocessing and feature extraction module 200, the model training module 300, the multi-classifier independent recognition and BPA calculation module 400, the BPA fusion module 500, and the detection result judgment module 600 are respectively used to implement the above-mentioned new energy vehicle motor abnormal sound detection method based on decision-level fusion Steps S1, S2, S3, S4, S5, and S6. Therefore, its specific implementation method can refer to the description of the corresponding embodiments of each part. In order to avoid redundancy, it will not be repeated here.
[0054] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
[0055] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0056] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0057] Obviously, the above embodiments are merely examples for clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from them are still within the protection scope of the invention.
Claims
1. A method for detecting abnormal noise of a new energy vehicle motor based on decision-level fusion, characterized in that: include: Step S1: collecting audio signal data of a new energy vehicle motor in various working states, annotating the audio signal data to distinguish normal sounds from abnormal sounds, and constructing a data set based on the annotated data; Step S2: preprocessing the audio signals in the data set, and extracting multiple audio features from the preprocessed audio signals; Step S3: inputting the extracted multiple audio features into multiple independent support vector machine classifiers for training, thereby obtaining multiple trained support vector machine classifiers; Step S4: multiple trained support vector machine classifiers independently identify and classify new audio signals based on different audio features to generate classification results, and convert the classification results of each support vector machine classifier into a basic probability distribution evidence body; Step S5: Fusing the basic probability distribution evidence bodies from different support vector machine classifiers to obtain a fusion result; Step S6: judging whether the sound signal of the new energy vehicle motor is normal according to the fusion result.
2. The new energy vehicle motor abnormal noise detection method based on decision-level fusion according to claim 1 is characterized in that: The method for preprocessing the audio signal in the data set includes removing background noise and normalizing processing.
3. The new energy vehicle motor abnormal noise detection method based on decision-level fusion according to claim 1 is characterized in that: The method for extracting multiple audio features from the preprocessed audio signal is: The audio features of the preprocessed audio signal are extracted using short-time Fourier transform, Mel-frequency cepstrum coefficient and Mel-spectrum respectively, including STFT features, MFCC features and Mel-spectrum features, where STFT represents short-time Fourier transform, MFCC represents Mel-frequency cepstrum coefficient and Mel represents Mel.
4. The new energy vehicle motor abnormal noise detection method based on decision-level fusion according to claim 3 is characterized in that: The method of extracting the audio features of the preprocessed audio signal using short-time Fourier transform is: Given a discrete-time signal , split it into a series of length Short time frame; There is a certain overlap between each frame to improve the time resolution; a window function is applied to each frame signal , to reduce the edge effect; the signal after windowing is expressed as: ; In the formula, is the mth frame signal; It is frame shift; is the index of the time domain sample; For each windowed frame signal Perform a fast Fourier transform and calculate its frequency components: ; In the formula, It is The frequency content of the frame, indicating the frequency The complex value at contains amplitude and phase information; is an imaginary unit; Calculate the power spectral density of each frame or Amplitude Spectrum , the calculation formula is: ; ; Arrange the spectrum results of each frame in time order to form a two-dimensional matrix , where each element represents the and frequency Signal strength on: 。 5. The new energy vehicle motor abnormal noise detection method based on decision-level fusion according to claim 3 is characterized in that: The method of extracting the audio features of the preprocessed audio signal using Mel frequency cepstral coefficients is: First, the original audio signal is preprocessed, including frame segmentation and windowing. Then, the preprocessed data is subjected to fast Fourier transform to obtain the energy spectrum. Then, the energy spectrum is filtered using multiple Mel filters, and the logarithm of the filtered energy is taken to calculate the logarithmic energy. Finally, the logarithmic energy is subjected to discrete cosine transform to extract the MFCC features.
6. The new energy vehicle motor abnormal noise detection method based on decision-level fusion according to claim 3 is characterized in that: The method of using Mel spectrum to extract audio features of preprocessed audio signals is: Divide the continuous audio signal into multiple short-time frames and perform windowing on each frame signal; Perform fast Fourier transform on each frame of windowed signal to convert the time domain signal into a frequency domain signal to obtain the spectrum representation of the frame; Square the spectrum obtained by fast Fourier transform and calculate the power spectrum of each frequency component; Process the power spectrum through a set of Mel filters; Perform logarithmic operation on the energy output by the Mel filter bank to obtain the logarithmic energy spectrum.
7. The new energy vehicle motor abnormal noise detection method based on decision-level fusion according to claim 1 is characterized in that: The method of inputting the extracted multiple audio features into multiple independent support vector machine classifiers for training to obtain multiple trained support vector machine classifiers is: The extracted audio features are input into multiple independent support vector machine classifiers for training. During the training process, the performance of each support vector machine classifier is evaluated using five-fold cross validation to optimize its parameters, and finally multiple trained support vector machine classifiers are obtained.
8. The new energy vehicle motor abnormal noise detection method based on decision-level fusion according to claim 1 is characterized in that: The method for fusing the basic probability distribution evidence bodies from different support vector machine classifiers is: Using DS evidence theory, the basic probability distribution evidence bodies from different support vector machine classifiers are fused; For two support vector machine classifiers and BPA, where BPA represents the basic probability assignment, the fusion The definition is as follows: ; In the formula, Indicates fusion ; It's the complete set Any subset of ; and Support vector machine classifier and BPA in the denominator is the conflict coefficient, which indicates the degree to which two SVM classifiers support completely different categories. This term is used for normalization to ensure that the sum of BPA is 1; For the fusion of multiple support vector machine classifiers, the recursive characteristics of the DS evidence theory rules are used; for three support vector machine classifiers , first merge get Then with Fusion; and so on, to achieve the fusion of any number of support vector machine classifiers.
9. The new energy vehicle motor abnormal noise detection method based on decision-level fusion according to claim 8 is characterized in that: The basic probability distribution is used to represent the support degree of a certain support vector machine classifier for different categories, and the support quantity of each category is quantified into a probability value; it reflects the trust of the support vector machine classifier in the category, and is mapped from the power set of the entire set to a numerical value between [0,1]; probability distribution is performed for each possible category and its combination, and the undistributed part is used to represent uncertainty, and the sum of the basic probability distribution is 1.
10. A new energy vehicle motor abnormal noise detection system based on decision-level fusion, characterized in that: The system is used to implement the new energy vehicle motor abnormal noise detection method based on decision-level fusion according to any one of claims 1 to 9, specifically comprising: The audio signal acquisition and data set construction module is used to collect audio signal data of new energy vehicle motors in various working states, annotate the audio signal data to distinguish normal sounds from abnormal sounds, and construct a data set based on the annotated data; The data preprocessing and feature extraction module is used to preprocess the audio signals in the data set and extract multiple audio features from the preprocessed audio signals; A model training module is used to input the extracted multiple audio features into multiple independent support vector machine classifiers for training, thereby obtaining multiple trained support vector machine classifiers; The multi-classifier independent identification and BPA calculation module is used to independently identify and classify new audio signals based on different audio features using multiple trained support vector machine classifiers, generate classification results, and convert the classification results of each support vector machine classifier into a basic probability distribution evidence body; The BPA fusion module is used to fuse the basic probability allocation evidence bodies from different support vector machine classifiers to obtain the fusion result; The detection result judgment module is used to judge whether the sound signal of the new energy vehicle motor is normal based on the fusion result.
Citation Information
Patent Citations
Equipment abnormal sound detection method and detecting device
CN106323452A
Identification model determination method, identification method and system for identifying abnormal sound of treadmill
CN110992985A
Substation equipment sound anomaly identification method and system
CN112289341A
Belt conveyor fault diagnosis method based on sound signals
CN113405825A
Industrial robot speed reducer health state assessment method based on multi-source information fusion
CN115290317A
