Machine equipment fault diagnosis system based on vibration signals

Abstract

The invention discloses a machine equipment fault diagnosis system based on vibration signals, which comprises a vibration signal acquisition module, a vibration signal processing and feature extraction module, a multi-path classification module and a display module which are connected in sequence, and is suitable for the technical field of machine equipment fault diagnosis. The method comprises the following steps: acquiring machinery equipment under different working conditions through an acceleration sensor to obtain vibration signals of the machinery equipment during operation; decomposing the vibration signals through a successive variational mode decomposition method, extracting entropy features, and combining the entropy features and time-frequency domain features to obtain a feature vector matrix; the feature vector matrix is classified and recognized through a multi-path classifier, and a classification result is obtained; the fault condition of the machine equipment is judged according to the classification result, and a diagnosis result is displayed on the server terminal; through vibration signal processing and multi-path classification training, various influences of machinery equipment under multiple working conditions can be effectively solved, and the fault diagnosis and recognition accuracy is improved to a great extent.

Description

Technical Field

[0001] The invention belongs to the technical field of machine equipment fault diagnosis, and in particular is a machine equipment fault diagnosis system based on vibration signals. Background Art

[0002] At present, most of the traditional machine equipment fault diagnosis methods are based on manual experience. Generally, maintenance engineers go to the site to diagnose and repair the machine equipment. This method not only wastes time, but also easily causes production and work stoppages. With the continuous development of computer and artificial intelligence technology, intelligent fault diagnosis technology has emerged.

[0003] Since vibration signals contain rich information and can reflect various physical phenomena and problems during equipment operation, machine equipment fault diagnosis technology based on vibration signals has attracted the attention of many researchers.

[0004] The key technologies include two aspects: one is the processing and feature extraction of vibration signals, and the other is the classification of vibration signal features using a classifier. At present, the processing and feature extraction of vibration signals mainly use signal decomposition methods, and feature classification mostly uses a single-channel classifier. Although certain effects have been achieved, there are still the following problems:

[0005] (1) The vibration signals collected during the operation of machinery and equipment often contain noise. Traditional signal decomposition methods usually cannot well preserve the characteristics of the original signal when processing complex signals containing noise.

[0006] (2) Machinery and equipment often operate under different loads and working conditions, which causes their vibration signals to

[0007] There may be significant differences in the spectrum and amplitude, and a single-path classifier can only extract features through one path, making it difficult to capture the diversity and complexity of vibration signals at the same time. Therefore, when processing vibration signal features, it may not be possible to fully capture the changes in vibration characteristics under multiple working conditions, resulting in a low accuracy rate in fault diagnosis and identification of machine equipment.

[0008] Therefore, how to improve the accuracy of fault diagnosis of machinery and equipment under multiple working conditions becomes a problem to be solved.

[0009] The signal decomposition of the present invention adopts the successive variational modal decomposition method to effectively separate different modes of the signal, reduce the interference of noise, and can more clearly analyze the real abnormal vibration situation. The feature classification adopts a multi-way classifier, and the vibration signal features of multiple different working conditions are simultaneously input into the multi-way classifier, which can reduce the interference and confusion between the features of different working conditions, and can also improve the robustness and generalization ability of the model, and greatly improve the accuracy of classification and recognition. Summary of the invention

[0010] The purpose of the present invention is to overcome the defects of the prior art and provide a machine equipment fault diagnosis system based on vibration signals.

[0011] To achieve the above object, the present invention adopts the following technical solutions:

[0012] A machine equipment fault diagnosis system based on vibration signals, comprising a vibration signal acquisition module, a vibration signal processing and feature extraction module, a multi-channel classification module and a display module connected in sequence;

[0013] The vibration signal acquisition module is configured to: collect vibration signals of machines and equipment in different working conditions through an acceleration sensor to obtain vibration signals of the machines and equipment when they are running;

[0014] The vibration signal processing and feature extraction module is configured to: decompose the vibration signal and extract entropy features by a successive variational mode decomposition method, combine the entropy features with time-frequency domain features, and obtain a feature vector matrix;

[0015] The multi-channel classification module is configured to: classify and identify the feature vector matrix through a multi-channel classifier to obtain a classification result;

[0016] The display module is configured to determine the fault condition of the machine equipment according to the classification result and display the diagnosis result on the server terminal.

[0017] Preferably, the vibration signal processing and feature extraction module decomposes the vibration signal and extracts entropy features by a successive variational mode decomposition method, combines the entropy features with the time-frequency domain features, and obtains a feature vector matrix, specifically comprising the following steps:

[0018] Step S21: input vibration signal and initialize successive variational modal decomposition parameters;

[0019] Specifically, the parameters of the successive variational modal decomposition include equilibrium parameters, time steps, tolerance of convergence criteria and stopping criteria;

[0020] Step S22: decomposing the signal input in step S21, and calculating the correlation coefficient between each intrinsic mode function and the input signal;

[0021] Step S23: selecting the eigenmode function with the highest correlation coefficient, and extracting the entropy feature of the eigenmode function;

[0022] Specifically, the entropy features include energy entropy, permutation entropy, approximate entropy, fuzzy entropy, information entropy, envelope entropy and sample entropy;

[0023] Step S24: combining the entropy feature obtained in step S23 with the time-frequency domain feature to obtain a feature vector matrix;

[0024] Specifically, the time-frequency domain features are extracted from the vibration signal by using a statistical analysis algorithm, which include: maximum value, minimum value, mean value, kurtosis, variance, peak value, rectified mean value, standard deviation, root mean square, peak factor, pulse factor, waveform factor, and margin factor, and the frequency domain features are extracted from the vibration signal by using a fast Fourier transform algorithm, which include: average frequency, center frequency, and root mean square frequency.

[0025] Preferably, the multi-way classifier is an MTCN-LSTM-Attention model, and the multi-way classifier is used to classify and identify the feature vector matrix to obtain a classification result, which specifically includes the following steps:

[0026] Step S31: Divide the feature vector matrix into a training set and a test set according to a preset ratio;

[0027] Step S32: training the pre-constructed MTCN-LSTM-Attention model using the training set to obtain a trained MTCN-LSTM-Attention model;

[0028] Step S33: classify the test set using the trained MTCN-LSTM-Attention model to obtain the classification result.

[0029] The beneficial effects of the present invention are as follows: by decomposing the vibration signals of the machine equipment during operation, the real abnormal vibration situation can be analyzed more clearly; by simultaneously inputting the vibration signal features of a variety of different working conditions into a multi-channel classifier, the interference and confusion between the features of different working conditions can be reduced, and the robustness and generalization ability of the model can be improved, thereby greatly improving the accuracy of classification and recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A block diagram of a machine equipment fault diagnosis system based on vibration signals provided by an embodiment of the present invention;

[0031] Figure 2 A schematic diagram of the structure of the MTCN-LSTM-Attention model provided in an embodiment of the present invention;

[0032] Figure 3 A schematic diagram of the structure of a TCN module provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the technical solution of the present invention and implement it, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the scope of protection of the present invention.

[0034] Before introducing the solutions of the present invention, the terms involved in the embodiments are explained first.

[0035] Temporal Convolutional Network (TCN): is a deep learning model for processing time series data. It is based on the idea of ​​convolutional neural network, extracts and learns features in time series data through convolution operations, and achieves good results in a series of time series prediction and classification tasks.

[0036] Multipath Temporal Convolutional Network (MTCN): A TCN with multi-channel input that can process multiple different data sources or feature sets. Each channel can extract and process specific features separately, making better use of information from different data sources.

[0037] Long Short Term Memory (LSTM): is a special type of recurrent neural network that can learn long-term dependencies in data and performs well in processing and predicting sequence data.

[0038] Attention mechanism: It can dynamically assign weights to different parts of the input data, thereby improving the model's focusing ability, focusing on important information in the input features, and enhancing the model's ability to recognize key features.

[0039] Example:

[0040] A machine equipment fault diagnosis system based on vibration signals, such as Figure 1 As shown, it includes a vibration signal acquisition module 1, a vibration signal processing and feature extraction module 2, a multi-channel classification module 3 and a display module 4 which are connected in sequence;

[0041] The vibration signal acquisition module 1 is configured to: collect vibration signals of machines and equipment in different working conditions through an acceleration sensor to obtain vibration signals of the machines and equipment when they are running;

[0042] The vibration signal processing and feature extraction module 2 is configured as follows: decomposing the vibration signal and extracting entropy features by a successive variational mode decomposition method, combining the entropy features with the time-frequency domain features to obtain a feature vector matrix;

[0043] In a possible implementation, the vibration signal processing and feature extraction module 2 decomposes the vibration signal and extracts entropy features by a successive variational mode decomposition method, combines the entropy features with the time-frequency domain features, and obtains a feature vector matrix, specifically including the following steps:

[0044] Step S21: input vibration signal and initialize successive variational modal decomposition parameters;

[0045] Specifically, the parameters of the successive variational modal decomposition include equilibrium parameters, time steps, tolerance of convergence criteria and stopping criteria;

[0046] Step S22: decomposing the signal input in step S21, and calculating the correlation coefficient between each intrinsic mode function and the input signal;

[0047] Step S23: selecting the eigenmode function with the highest correlation coefficient, and extracting the entropy feature of the eigenmode function;

[0048] Specifically, the entropy features include energy entropy, permutation entropy, approximate entropy, fuzzy entropy, information entropy, envelope entropy and sample entropy;

[0049] Step S24: Combine the entropy feature obtained in step S23 with the time-frequency domain feature to obtain a feature vector matrix.

[0050] In one possible implementation, the time-frequency domain features are extracted from the vibration signal by using a statistical analysis algorithm, which includes: maximum value, minimum value, mean value, kurtosis, variance, peak value, rectified mean value, standard deviation, root mean square, peak factor, pulse factor, waveform factor, and margin factor, and the frequency domain features are extracted from the vibration signal by using a fast Fourier transform algorithm, which includes: average frequency, center frequency, and root mean square frequency.

[0051] The multi-way classification module 3 is configured to: classify and identify the feature vector matrix through a multi-way classifier to obtain a classification result;

[0052] In a possible implementation, the multi-way classifier is an MTCN-LSTM-Attention model, and the feature vector matrix is ​​classified and identified by the multi-way classifier to obtain a classification result, which specifically includes the following steps:

[0053] Step S31: Divide the feature vector matrix into a training set and a test set according to a preset ratio;

[0054] Step S32: training the pre-constructed MTCN-LSTM-Attention model using the training set to obtain a trained MTCN-LSTM-Attention model;

[0055] Step S33: classify the test set using the trained MTCN-LSTM-Attention model to obtain the classification result.

[0056] In a possible implementation, Figure 2 As shown in the figure, the MTCN-LSTM-Attention model structure includes an input layer, multiple parallel TCN modules, an LSTM layer, an attention layer, a concatenation layer, a fully connected layer, a Softmax layer, and a classification layer. The classification process includes the following steps:

[0057] Step A1: Send the training set in step S31 to the input layer, and normalize the feature data received by the input layer. The calculation expression is:

[0058]

[0059] in, represents the normalized data, Represents feature data, and Indicates the maximum and minimum values ​​of the characteristic data, and Set the upper and lower limits of the normalized value range respectively;

[0060] Step A2: The data normalized by step A1 is simultaneously passed to each parallel TCN module to capture feature relationships at different scales;

[0061] Step A3: Use the concatenation layer to concatenate the outputs of multiple TCN modules to form a comprehensive feature representation;

[0062] Specifically, the "plurality" in the plurality of TCN modules corresponds to the number of machine equipment operating conditions.

[0063] Step A4: The LSTM layer receives the output of the concatenation layer and further processes the complex relationship between features;

[0064] Step A5: Use the attention mechanism to focus on important information in the input features and enhance the model's ability to identify key features;

[0065] Step A6: The fully connected layer maps the output features of the LSTM layer and the attention layer to a lower-dimensional space to further process the complex relationship between features;

[0066] Step A7: The Softmax layer converts the output into a probability distribution, and the classification layer defines the loss function and calculates the classification error.

[0067] In a possible implementation, Figure 3As shown in the figure, each TCN module structure consists of 4 layers, namely convolution, layer normalization, ReLU and Dropout layers.

[0068] The display module 4 is configured to determine the fault condition of the machine equipment according to the classification result and display the diagnosis result on the server terminal.

[0069] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A machine equipment fault diagnosis system based on vibration signals, Features: It comprises a vibration signal acquisition module (1), a vibration signal processing and feature extraction module (2), a multi-channel classification module (3) and a display module (4) which are connected in sequence; The vibration signal acquisition module (1) is configured to: acquire vibration signals of machines and equipment in different working conditions through an acceleration sensor to obtain vibration signals of the machines and equipment when they are in operation; The vibration signal processing and feature extraction module (2) is configured to: decompose the vibration signal and extract entropy features by a successive variational mode decomposition method, and combine the entropy features with time-frequency domain features to obtain a feature vector matrix; The multi-channel classification module (3) is configured to: classify and identify the feature vector matrix through a multi-channel classifier to obtain a classification result; The display module (4) is configured to: determine the fault condition of the machine equipment according to the classification result, and display the diagnosis result on the server terminal.

2. A machine equipment fault diagnosis system based on vibration signals as claimed in claim 1, Features: The vibration signal processing and feature extraction module decomposes the vibration signal and extracts entropy features by a successive variational mode decomposition method, combines the entropy features with the time-frequency domain features, and obtains a feature vector matrix, specifically comprising the following steps: Step S21: input vibration signal and initialize successive variational modal decomposition parameters; Specifically, the parameters of the successive variational modal decomposition include equilibrium parameters, time steps, tolerance of convergence criteria and stopping criteria; Step S22: decomposing the signal input in step S21, and calculating the correlation coefficient between each intrinsic mode function and the input signal; Step S23: selecting the eigenmode function with the highest correlation coefficient, and extracting the entropy feature of the eigenmode function; Specifically, the entropy features include energy entropy, permutation entropy, approximate entropy, fuzzy entropy, information entropy, envelope entropy and sample entropy; Step S24: combining the entropy feature obtained in step S23 with the time-frequency domain feature to obtain a feature vector matrix; Specifically, the time-frequency domain features are extracted from the vibration signal by using a statistical analysis algorithm, which include: maximum value, minimum value, mean value, kurtosis, variance, peak value, rectified mean value, standard deviation, root mean square, peak factor, pulse factor, waveform factor, and margin factor, and the frequency domain features are extracted from the vibration signal by using a fast Fourier transform algorithm, which include: average frequency, center frequency, and root mean square frequency.

3. A machine equipment fault diagnosis system based on vibration signals as claimed in claim 1, Features: The multi-channel classifier is an MTCN-LSTM-Attention model. The multi-channel classifier is used to classify and identify the feature vector matrix to obtain a classification result, which specifically includes the following steps: Step S31: Divide the feature vector matrix into a training set and a test set according to a preset ratio; Step S32: training the pre-constructed MTCN-LSTM-Attention model using the training set to obtain a trained MTCN-LSTM-Attention model; Step S33: classify the test set using the trained MTCN-LSTM-Attention model to obtain the classification result.

Images