Multi-modal marine bearing fault detection method based on acoustic characteristics
Through the multimodal marine bearing fault detection method based on deep learning, the time domain and frequency domain characteristics are combined with LSTM, residual network and attention mechanism, the problem of insufficient generalization of model in marine bearing fault diagnosis is solved, and the accuracy of cross-domain diagnosis and fault detection is improved.
Patent Information
- Application Number
- CN202311448911.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-02
- Publication Date
- 2025-05-06
AI Technical Summary
In marine bearing fault diagnosis, the training set and test set usually come from different distributions, resulting in insufficient generalization of the model and degradation of diagnostic accuracy.
A multimodal marine bearing fault detection method based on deep learning is adopted to build a multimodal detection model based on acoustic characteristics by collecting and preprocessing vibration signal data. This model uses time domain and frequency domain features, combined with LSTM, residual network and attention mechanism, to perform feature extraction and fusion to achieve cross-domain diagnosis.
It effectively improves the ability to extract fault features, narrows the difference in feature distribution caused by environmental factors in the working conditions, realizes cross-domain diagnosis between different working conditions, and improves the accuracy and reliability of fault detection.
Smart Images
Figure CN119935549A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of mechanical fault detection, and in particular relates to a multi-mode marine bearing fault detection method based on acoustic characteristics. Background Art
[0002] Industrial systems are gradually developing towards intelligent manufacturing, and various mechanical equipment are tending towards automation and complexity. Rolling bearings, as key components of the transmission devices of large industrial equipment, are closely related to the normal operation of the equipment. Intelligent research on bearing fault diagnosis is of great significance to ensuring production and avoiding accidents.
[0003] At present, most of the research on mechanical fault diagnosis focuses on signal feature selection and diagnostic classification. Excessive vibration is usually the main factor leading to bearing failure. Modeling and analyzing the vibration signal of the equipment collected by sensors has become one of the main bases for operating status diagnosis. In the diagnosis method based on signal analysis and processing, the time domain and frequency domain are two angles for observing the implicit laws within the original vibration data, and have different sensitivities to fault modes. The time domain characteristics reflect the change of signal amplitude over time, and the frequency domain characteristics study the distribution law of signal energy in each frequency band. For fault diagnosis tasks, multimodal information of the same data can mine more signal features.
[0004] The deep learning model can achieve good diagnostic results based on the premise that the training set and the test set come from the same distribution. However, machinery such as marine bearings has certain particularities. In actual industrial scenarios, the diversity of working conditions and equipment models leads to the fact that the training set and the test set usually come from different distributions. It is also difficult to collect labeled data that is sufficient to support the training of deep learning models under all conditions. At this time, the lack of model generalization leads to a decrease in diagnostic accuracy. Therefore, improving the ability to extract fault features and reducing the differences in feature distribution caused by factors such as working environment, and implementing effective cross-domain diagnosis between different working conditions are two key issues facing intelligent fault diagnosis. Summary of the invention
[0005] In view of the above problems, the object of the present invention is to provide a multi-modal marine bearing fault detection method based on acoustic characteristics.
[0006] The specific technical solution for achieving the purpose of the present invention is:
[0007] A multi-modal marine bearing fault detection method based on acoustic characteristics comprises the following steps:
[0008] Step 1: Collect and preprocess marine bearing data to form a data set;
[0009] Step 2: Build a multimodal marine bearing fault detection model based on acoustic characteristics based on deep learning;
[0010] Step 3: training and testing the multimodal marine bearing fault detection model based on the data set in step 1;
[0011] Step 4: Perform bearing fault detection based on the trained multimodal marine bearing fault detection model.
[0012] Compared with the prior art, the present invention has the following beneficial effects:
[0013] (1) The method of the present invention makes full use of the multi-angle characterization information of the vibration signal, takes the time domain and frequency domain as multi-modal information input, focuses on the timing information in the vibration signal through the LSTM mechanism, uses the residual network to extract acoustic features, and further uses the attention mechanism to fuse features of multiple modes to fully mine the effective fault information contained in the signal;
[0014] (2) The scheme of the present invention uses multimodal comprehensive analysis to model the bearing operation process based on the signal energy intensity, time-frequency domain variation characteristics and amplitude domain statistical characteristics of the acoustic emission signal. The acoustic feature analysis part focuses on the energy intensity of the acoustic emission signal, extracts highly correlated and robust acoustic features from a multi-scale and multi-level perspective, and provides the most effective pattern recognition information for multimodal analysis; based on the time series analysis part, it focuses on the temporal dynamic changes of the time-frequency domain signal to accurately describe the dynamic change trend of the frequency domain signal, and pays attention to and provides the relevant time series intervals of multimodal analysis to avoid the limitation that a single mode cannot accurately locate the ultrasonic noise characteristics, the abnormal working condition occurrence interval and the working condition state change characteristics. The multimodal comprehensive analysis module fully considers the complementary and redundant information between the acoustic features and the time series features, highlights the acoustic effective part in the time series modeling, and extracts and fuses the more essential features of the bearing grinding signal from different modes, such as the time domain and frequency domain of the signal, so as to achieve cross-domain diagnosis between different working conditions.
[0015] The present invention is further described below in conjunction with specific implementation modes. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 It is a schematic diagram of the architecture of the multi-modal marine bearing fault detection model of the present invention.
[0017] Figure 2 Schematic diagram of the acoustic feature extraction unit in the fault detection model of the present invention.
[0018] Figure 3 It is a schematic diagram of extracting time-frequency domain sequence features in the fault detection model of the present invention.
[0019] Figure 4 It is a schematic diagram of a multi-modal comprehensive analysis unit in the fault detection model of the present invention.
[0020] Figure 5 Schematic diagram of the fully connected layer in the fault detection model of an embodiment of the present invention. DETAILED DESCRIPTION
[0021] A multi-modal marine bearing fault detection method based on acoustic characteristics comprises the following steps:
[0022] Step 1, collecting marine bearing data and preprocessing it to form a data set;
[0023] Among them, the preprocessing of marine bearing data includes:
[0024] Collect the original vibration signal data of marine bearings under different working conditions;
[0025] Perform edge processing on the collected data and network the data to form a marine bearing data set;
[0026] The data set is divided into training set and test set according to a certain ratio.
[0027] Step 2: Build a multimodal marine bearing fault detection model based on acoustic characteristics based on deep learning:
[0028] The multimodal marine bearing fault detection is based on a deep learning algorithm to construct a black box model between input parameters and output parameters;
[0029] The multimodal bearing fault detection model based on acoustic characteristics includes an acoustic feature extraction unit based on acoustic emission signals, a time series analysis unit, and a multimodal comprehensive analysis unit. By comprehensively analyzing the multi-dimensional features of the acoustic signal, the multimodal information of the time domain and the frequency domain of the data is integrated at the feature level to construct a multimodal bearing fault detection model and output the fault detection result.
[0030] The acoustic feature extraction unit of the acoustic emission signal successively performs spectrogram processing, acoustic feature analysis and deep acoustic feature extraction on the acoustic emission signal;
[0031] Among them, the acoustic feature extraction part of the acoustic emission signal performs spectrogram processing on the acoustic signal;
[0032] The acoustic feature analysis part targets the spectrogram of the acoustic emission signal and reduces irrelevant acoustic features based on principal component analysis and linear discriminant analysis;
[0033] Deep acoustic feature extraction uses a deep multi-scale convolutional neural network to extract acoustic features from acoustic emission signals after acoustic feature analysis.
[0034] The time series analysis part successively performs time-frequency domain signal processing, time-frequency domain feature analysis and time-frequency domain sequence feature extraction on the acoustic emission signal;
[0035] Wherein, the time-frequency domain signal processing performs time-frequency domain processing on the acoustic emission signal based on wavelet transform;
[0036] The time-frequency domain feature analysis performs feature analysis on the time-frequency spectrum after wavelet transformation to extract the frequency transfer dynamics of the acoustic emission signal in a short-term, stable time series;
[0037] The time-frequency domain sequence feature extraction is based on the MSCNN integrated module, which analyzes the frequency domain features of the acoustic emission signal at different scales in parallel, reduces the information resolution layer by layer, and improves the field to obtain deep acoustic signal features. In the long- and short-term time series feature aggregation stage, the long- and short-term memory network uses cell units to save the effective part of the hidden state generated at the previous moment, and effectively learns the time series structure characteristics of the abnormal state of the acoustic emission signal.
[0038] The multimodal comprehensive analysis unit fuses the acoustic features and time series features obtained by the acoustic feature extraction unit and the time series analysis unit based on a multi-head attention mechanism.
[0039] Step 3: Train and test the multimodal marine bearing fault detection model based on the data set in step 1:
[0040] Specifically, based on the training set in step 1, the parameters of each part are optimized through iterative training using the gradient descent algorithm to obtain the trained fault detection model, and then the model is tested through the test set until the test meets the requirements.
[0041] Step 4: Perform bearing fault detection based on the trained multimodal marine bearing fault detection model.
[0042] A multi-modal marine bearing fault detection system based on acoustic characteristics, including the following modules:
[0043] Data processing module: used to collect marine bearing data and perform preprocessing to form a data set;
[0044] Model building module: used to build a multimodal marine bearing fault detection model based on acoustic characteristics based on deep learning;
[0045] Model training module: used to train and test the multimodal marine bearing fault detection model based on the data set in step 1;
[0046] Fault detection module: used to perform bearing fault detection based on the trained multimodal marine bearing fault detection model.
[0047] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the following steps are implemented:
[0048] Step 1: Collect and preprocess marine bearing data to form a data set;
[0049] Step 2: Build a multimodal marine bearing fault detection model based on acoustic characteristics based on deep learning;
[0050] Step 3: training and testing the multimodal marine bearing fault detection model based on the data set in step 1;
[0051] Step 4: Perform bearing fault detection based on the trained multimodal marine bearing fault detection model.
[0052] A computer storable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:
[0053] Step 1: Collect and preprocess marine bearing data to form a data set;
[0054] Step 2: Build a multimodal marine bearing fault detection model based on acoustic characteristics based on deep learning;
[0055] Step 3: training and testing the multimodal marine bearing fault detection model based on the data set in step 1;
[0056] Step 4: Perform bearing fault detection based on the trained multimodal marine bearing fault detection model.
[0057] Example
[0058] Combination Figure 1 , a multi-modal marine bearing fault detection method based on acoustic characteristics, comprising the following steps:
[0059] Step 1: Collect and preprocess marine bearing data to form a data set;
[0060] Among them, the preprocessing of marine bearing data includes:
[0061] Collect the original vibration signal data of marine bearings under different working conditions;
[0062] Perform edge processing on the collected data and network the data to form a marine bearing data set;
[0063] The data set is divided into training set and test set according to a certain ratio.
[0064] In this embodiment, a bearing fault data set is used to draw two-dimensional waveforms in the time domain and frequency domain, and then the data samples are divided into a training set and a test set in a ratio of 8:2. Both working conditions contain 36 normal states and 52 fault states. The data details are shown in Table 1 below:
[0065]
[0066] Step 2: Build a multimodal marine bearing fault detection model based on acoustic characteristics based on deep learning:
[0067] The multimodal marine bearing fault detection is based on a deep learning algorithm to construct a black box model between input parameters and output parameters;
[0068] The multimodal bearing fault detection model based on acoustic characteristics includes an acoustic feature extraction unit based on acoustic emission signals, a time series analysis unit, and a multimodal comprehensive analysis unit. By comprehensively analyzing the multi-dimensional features of the acoustic signal, the multimodal information of the time domain and the frequency domain of the data is integrated at the feature level to construct a multimodal bearing fault detection model and output the fault detection result.
[0069] Combination Figure 2 , the acoustic feature extraction unit of the acoustic emission signal successively performs spectrogram processing, acoustic feature analysis and deep acoustic feature extraction on the acoustic emission signal;
[0070] Among them, the acoustic feature extraction part of the acoustic emission signal performs spectrogram processing on the acoustic signal, using a fast Fourier transform sequence length of 1024 and a distance of 512 between adjacent sliding window frames, and converts it into a spectrogram, which displays a large amount of information related to acoustic characteristics. It combines the characteristics of the spectrum graph and the time domain waveform to show the change of the spectrum over time;
[0071] Acoustic feature analysis is performed on the spectrogram of the acoustic emission signal to reduce irrelevant acoustic features based on principal component analysis and linear discriminant analysis;
[0072] Deep acoustic feature extraction uses a deep multi-scale convolutional neural network to extract highly correlated and robust acoustic features from the acoustic emission signal after acoustic feature analysis. At the same time, the ResNet network is used, and a combination of convolutional layers and normalization layers is adopted. A residual structure is added to increase the depth of model building and reduce feature loss. A one-dimensional convolution of size 3 is used, and a ReLU activation function is used.
[0073] Combination Figure 3 The time series analysis part successively performs time-frequency domain signal processing, time-frequency domain feature analysis and time-frequency domain sequence feature extraction on the acoustic emission signal;
[0074] Wherein, the time-frequency domain signal processing performs time-frequency domain processing on the acoustic emission signal based on wavelet transform;
[0075] The time-frequency domain feature analysis performs feature analysis on the time-frequency spectrum after wavelet transformation to extract the frequency transfer dynamics of the acoustic emission signal in a short-term, stable time series;
[0076] The time-frequency domain sequence feature extraction is based on the MSCNN integrated module, which analyzes the frequency domain features of the acoustic emission signal at different scales in parallel, and obtains deep acoustic signal features while reducing the information resolution layer by layer. In the long- and short-term temporal feature aggregation stage, the long- and short-term memory network uses cell units to save the effective part of the hidden state generated at the previous moment, and effectively learns the temporal structure characteristics of the abnormal state of the acoustic emission signal.
[0077] In this embodiment, the time series analysis unit first uses 8 layers of convolution to reduce the dimension of the data, using a 3×3 convolution kernel with a step size of 2, and then extracts the time series features through a 2-layer LSTM, with a GeLU activation function in the middle.
[0078] Combination Figure 4 The multimodal comprehensive analysis unit fuses the acoustic features and time series features obtained by the acoustic feature extraction unit and the time series analysis unit based on a multi-head attention mechanism.
[0079] The multimodal comprehensive analysis unit will fully consider the complementary and redundant information between acoustic features and time series features, highlight the acoustic effective part of time series modeling, and fuse the features extracted by the acoustic feature extraction module and the time series analysis module through the attention mechanism;
[0080] The fused features are mapped to the bearing failure probability through the FC layer and the softmax layer, as shown in Figure 5 As shown, the bearing fault detection is completed;
[0081] Step 3: Train and test the multimodal marine bearing fault detection model based on the data set in step 1:
[0082] Specifically, based on the training set in step 1, the parameters of each part are optimized through iterative training using the gradient descent algorithm to obtain the trained fault detection model, and then the model is tested through the test set until the test meets the requirements.
[0083] Step 4: Perform bearing fault detection based on the trained multimodal marine bearing fault detection model.
[0084] The above embodiments show and describe the basic principles and main features of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.
Claims
1. A multi-modal marine bearing fault detection method based on acoustic characteristics, characterized in that: The following steps are involved: Step 1, collecting marine bearing data and preprocessing it to form a data set; Step 2: Build a multimodal marine bearing fault detection model based on acoustic characteristics based on deep learning; Step 3: training and testing the multimodal marine bearing fault detection model based on the data set in step 1; Step 4: Perform bearing fault detection based on the trained multimodal marine bearing fault detection model.
2. The multi-modal marine bearing fault detection method based on acoustic characteristics according to claim 1 is characterized in that: The preprocessing of the marine bearing data in step 1 includes: Collect the original vibration signal data of marine bearings under different working conditions; Perform edge processing on the collected data and network the data to form a marine bearing data set; The data set is divided into training set and test set according to a certain ratio.
3. The multi-modal marine bearing fault detection method based on acoustic characteristics according to claim 1 is characterized in that: The construction of a multi-modal marine bearing fault detection model in step 2 is specifically as follows: The multimodal marine bearing fault detection is based on a deep learning algorithm to construct a black box model between input parameters and output parameters; The multimodal bearing fault detection model based on acoustic characteristics includes an acoustic feature extraction unit based on acoustic emission signals, a time series analysis unit and a multimodal comprehensive analysis unit. By comprehensively analyzing the multi-dimensional characteristics of the acoustic signal, the multimodal information of the time domain and frequency domain of the data is integrated at the feature level to construct a multimodal bearing fault detection model and output the fault detection result.
4. The multi-modal marine bearing fault detection method based on acoustic characteristics according to claim 3 is characterized in that: The acoustic feature extraction unit of the acoustic emission signal successively performs spectrogram processing, acoustic feature analysis and deep acoustic feature extraction on the acoustic emission signal; Among them, the acoustic feature extraction part of the acoustic emission signal performs spectrogram processing on the acoustic signal; The acoustic feature analysis part targets the spectrogram of the acoustic emission signal and reduces irrelevant acoustic features based on principal component analysis and linear discriminant analysis; Deep acoustic feature extraction uses a deep multi-scale convolutional neural network to extract acoustic features from acoustic emission signals after acoustic feature analysis.
5. The multi-modal marine bearing fault detection method based on acoustic characteristics according to claim 3 is characterized in that: The time series analysis part successively performs time-frequency domain signal processing, time-frequency domain feature analysis and time-frequency domain sequence feature extraction on the acoustic emission signal; Wherein, the time-frequency domain signal processing performs time-frequency domain processing on the acoustic emission signal based on wavelet transform; The time-frequency domain feature analysis performs feature analysis on the time-frequency spectrum after wavelet transformation to extract the frequency transfer dynamics of the acoustic emission signal in a short-term, stable time series; The time-frequency domain sequence feature extraction is based on the MSCNN integrated module, which analyzes the frequency domain features of the acoustic emission signal at different scales in parallel, reduces the information resolution layer by layer while improving and obtaining the deep acoustic signal features. In the long- and short-term time series feature aggregation stage, the long- and short-term memory network uses cell units to save the effective part of the hidden state generated at the previous moment, and effectively learns the time series structure characteristics of the abnormal state of the acoustic emission signal.
6. The multi-modal marine bearing fault detection method based on acoustic characteristics according to claim 3 is characterized in that: The multimodal comprehensive analysis unit fuses the acoustic features and time series features obtained by the acoustic feature extraction unit and the time series analysis unit based on a multi-head attention mechanism; The obtained fused features are mapped to the bearing fault profile to complete the bearing fault detection.
7. The multi-modal marine bearing fault detection method based on acoustic characteristics according to claim 1 is characterized in that: When training the multimodal marine bearing fault detection model in step 3, based on the training set in step 1, the gradient descent algorithm is used to iteratively train and optimize the parameters of each part to obtain the trained fault detection model, and then the model is tested through the test set until the test meets the requirements.
8. A multi-modal marine bearing fault detection system based on acoustic characteristics, characterized in that: Includes the following modules: Data processing module: used to collect marine bearing data and perform preprocessing to form a data set; Model building module: used to build a multimodal marine bearing fault detection model based on acoustic characteristics based on deep learning; Model training module: used to train and test the multimodal marine bearing fault detection model based on the data set in step 1; Fault detection module: used to perform bearing fault detection based on the trained multimodal marine bearing fault detection model.
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, the steps of the method according to any one of claims 1 to 7 are implemented.
10. A computer storable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.