Small sample fault diagnosis method based on SDWT-GAN and adaptive multiple channels

Abstract

The invention discloses a small sample fault diagnosis method based on SDWT-GAN and adaptive multiple channels, and belongs to the technical field of artificial intelligence. The method is characterized by comprising the following steps of: 1, acquiring a feature data set of a plurality of components by utilizing a plurality of sensors deployed in a spacecraft mass spectrometer; step 2, constructing a network model based on SDWT-GAN enhanced feature data, and enhancing features of small sample fault data; 3, correlation analysis is carried out on the signals of the multiple sensors, and a fault data set is generated according to fault characteristics and characteristics of different parts after the different parts generate data; and 4, building a multi-channel fault classification model, and respectively introducing each type of fault samples into the multi-channel fault classification model to obtain a fault diagnosis result. According to the small sample fault diagnosis method based on the SDWT-GAN and the self-adaptive multi-channel, the SDWT-GAN, correlation analysis and the self-adaptive multi-channel classifier are combined, so that the fault diagnosis algorithm has the functions of fault early warning and fault mode discrimination, and intelligent monitoring of the fault diagnosis function is realized.

Description

Technical Field

[0001] The invention discloses a small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel, belonging to the field of artificial intelligence technology. Background Art

[0002] The mass spectrometer is mainly composed of five parts: mass spectrometer injection assembly, ion source, analyzer, multiplier tube and vacuum assembly. The mass spectrometer injection assembly can be divided into direct injection and chromatographic injection. Single-component and high-boiling point liquid samples can be directly injected. The function of the ion source is to ionize the sample molecules to be analyzed into charged ions, and make these ions converge into ion beams with a certain geometric shape and energy under the action of the ion optical system, and then enter the mass analyzer to be separated. The analyzer uses a quadrupole mass analyzer, which is composed of four straight metal or metal-plated poles arranged parallel to the axis at equal distances. The ideal surface of the rod is a hyperbolic surface. The multiplier tube bombards the cathode surface of the electron multiplier tube with ions coming out of the mass analyzer to emit secondary electrons, and then bombards a series of electrodes with secondary electrons in turn to continuously multiply the secondary electrons. Finally, the anode receives the electron flow, so that the ion beam is amplified and detected. The vacuum assembly is mainly composed of an ion pump and an aspirator pump, which can keep the entire device in a high vacuum state to ensure the safe operation of each part.

[0003] As a high-end analytical scientific instrument, mass spectrometer has the characteristics of covering a wide range of technologies, high performance requirements such as stability and accuracy, and complex fault location and maintenance. Through the analysis of mass spectrometer mechanism and data mining, timely location of existing faults and early warning of potential faults are of great significance for the efficient use of mass spectrometers. Considering the diverse types, trends, and judgment criteria of mass spectrometer parameters, and the correlation and coupling of some parameters, a single algorithm cannot meet the needs of mass spectrometer fault diagnosis.

[0004] In recent years, with the continuous development of machine learning, data-driven fault handling methods have been improved in terms of data processing efficiency and accuracy. Wavelet denoising as a data preprocessing part can effectively reduce the noise of data. Adversarial neural network is an artificial neural network composed of a generator and a discriminator. The theory of zero-sum game can more effectively enhance the characteristics of various types of data. Model fusion can make full use of the characteristics of the data set itself to select the classifier model channel appropriately. On the other hand, model channel parallelism can process data more quickly, effectively improving the efficiency and accuracy of fault diagnosis. At present, it is mainly used in the fields of pattern recognition, artificial intelligence, etc., and has shown higher accuracy than traditional methods in specific applications such as environmental evaluation and prediction. Summary of the invention

[0005] The technical problem to be solved by the present invention is: to overcome the shortcomings of the prior art and provide a small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel, which combines SDWT-GAN, correlation analysis and adaptive multi-channel classifier to enable the fault diagnosis algorithm to have fault warning and fault mode discrimination functions, so as to realize intelligent monitoring of the fault diagnosis function.

[0006] The technical solution adopted by the present invention to solve the technical problem is: the small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel is characterized by comprising the following steps: Step 1, using multiple sensors deployed in a spacecraft mass spectrometer to obtain characteristic data sets of multiple components; Step 2: Build a network model based on SDWT-GAN to enhance the feature data and enhance the features of small sample fault data; Step 3: perform correlation analysis on multi-sensor signals and generate a fault data set based on the fault characteristics and the characteristics of different components after generating data; Step 4: Build a multi-channel fault classification model, and pass each type of fault sample into the multi-channel fault classification model to obtain the fault diagnosis result.

[0007] Preferably, in step 1, the plurality of components include a mass spectrometer injection assembly, an ion source, an analyzer, a multiplier tube, and a vacuum assembly.

[0008] Preferably, step 2 comprises the following steps: Step 2-1, using the principle of S transform and discrete wavelet denoising to perform feature extraction and denoising on the data set collected by multiple sensors, reconstructing the mass spectrometry signal to obtain { W 1, W 2,…, W i}Noise reduction dataset; Step 2-2, construct and train the SDWT-GAN neural network; Step 2-3, use the generator in GAN to generate { G 1, G 2,…, G i}and{ W 1, W 2,…, W i The dataset is passed into the discriminator for judgment to obtain the feature enhanced dataset { P 1, P 2,…, P i}; Step 2-4, the obtained data set { P 1,P 2,…, P i}Input into the feature verification part constructed by PCA and T-SNE dimensionality reduction methods.

[0009] Preferably, step 2-1 comprises the following steps: Step 2-1-1, S transform performs preliminary feature extraction on the original signal; Step 2-1-2, wavelet denoising.

[0010] Preferably, step 2-1-2 includes the following steps: Step 2-1-2-1, wavelet decomposition of the signal; Select a wavelet and determine the level N of wavelet decomposition, then perform N-level wavelet decomposition calculations on the signal. Different types of wavelets are selected for mass spectrometry signals with different characteristics through training; Step 2-1-2-2, threshold quantization of high-frequency coefficients of wavelet decomposition; For each high-frequency coefficient from the 1st layer to the Nth layer, a threshold is selected for threshold quantization processing; Step 2-1-2-3, wavelet reconstruction of the signal; based on the low-frequency coefficients of the Nth layer of wavelet decomposition and the high-frequency coefficients of the 1st to Nth layers after quantization, wavelet reconstruction of the mass spectrum signal is performed.

[0011] Preferably, steps 2-3 include the following steps: Step 2-3-1, pass the noise signal into two generators composed of a temporal convolutional neural network and a bidirectional long short-term memory network, and obtain two types of generated samples required by the discriminator: real samples and generated samples; Step 2-3-2, input the real samples and generated samples into the discriminator for judgment, and select the appropriate generator network as the network model of different feature samples according to the different types of real samples and the error size of the data output results; Step 2-3-2, perform PCA and T-SNE dimensionality reduction processing on the output results.

[0012] Preferably, step 3 comprises the following steps: Step 3-1: By establishing a visual correlation matrix, we can observe the correlation strength of the samples, classify the samples by the correlation strength, and obtain the fault sample data set. The data set of n types with different correlations is recorded as { X 1, X 2,…, X n}, { Y 1, Y 2,…, Y n}, …, {Z 1, Z 2,…, Z n}; Step 3-2, define the fault severity in each type of data set, label the data set through the mapping relationship, and obtain the data sample set of the fault classification model: The fault type labels of different data sets are X 1, X 2,…, X n}, { Y 1, Y 2,…, Y n}, …, { Z 1, Z 2,…, Z n} and for the failure data set of key components of mass spectrometers { X 1, X 2,…, X n}, { Y 1, Y 2,…, Y n}, …, { Z 1, Z 2,…, Z n}Establish a mapping relationship and define it as a training sample, where x , y , z represents the sample label of each type of fault dataset, X , Y , Z A signal sample representing different samples.

[0013] Preferably, in step 4, the multi-channel fault classification model is constructed by using logistic regression, decision tree or random forest, K-nearest neighbor and neural network.

[0014] Compared with the prior art, the present invention has the following beneficial effects: In this small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel, by combining SDWT-GAN, correlation analysis and adaptive multi-channel classifier, the fault diagnosis algorithm has fault warning and fault mode discrimination functions, so as to realize intelligent monitoring of fault diagnosis function. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Flowchart of the small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel.

[0016] Figure 2 Schematic diagram of the structure of SDWT-GAN and adaptive multi-channel classifier constructed based on the small sample fault diagnosis method of SDWT-GAN and adaptive multi-channel.

[0017] Figure 3 This is the confusion matrix diagram of the classification results of the small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel. DETAILED DESCRIPTION

[0018] Figures 1 to 3 The best embodiment of the present invention is shown below in conjunction with the attached Figures 1 to 3 The present invention is further described.

[0019] like Figure 1 As shown in FIG. 1 , a small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel (hereinafter referred to as the fault diagnosis method) includes the following steps: Step 1, using multiple sensors deployed in a spacecraft mass spectrometer to obtain a component feature data set; Multiple sensors deployed in the spacecraft mass spectrometer are used to obtain characteristic data sets of various key components, including the mass spectrometer injection assembly, ion source, analyzer, multiplier tube, and vacuum assembly. The sensors include temperature sensors for measuring particle source temperature, voltage sensors for analyzer voltage, current sensors for multiplier current, and pressure sensors for vacuum assembly pressure.

[0020] Step 2, construct a network model based on SDWT-GAN enhanced feature data; Construct a network model based on SDWT-GAN to enhance the feature data and enhance the features of small sample fault data. The specific steps include: Step 2-1, using the principle of S transform and discrete wavelet denoising to perform feature extraction and denoising on the data set collected by multiple sensors, reconstructing the mass spectrometry signal to obtain { W 1, W 2,…, W i}Noise reduction dataset; Combination Figure 2 , step 2-1 further includes the following steps: Step 2-1-1, S transform performs preliminary feature extraction on the original signal; Step 2-1-2, wavelet denoising: Wavelet denoising is a combination of feature extraction and low-pass filtering. A noisy mass spectrometry signal model can be expressed as follows: S ( k )= f ( k )+ ε*e ( k ) k= 0, 1, ..., n -1 in, f ( k ) is the useful signal in the mass spectrometry signal, S ( k ) is a noisy signal, e ( k ) is the noise, ε is the standard deviation of the noise system.

[0021] The noise reduction of small sample signals of mass spectrometer is divided into the following steps: Step 2-1-2-1, wavelet decomposition of the signal.

[0022] A wavelet is selected and the level N of wavelet decomposition is determined. Then, the signal is subjected to N-layer wavelet decomposition calculations. Mass spectrometry signals with different characteristics select different types of wavelets through training.

[0023] Step 2-1-2-2, threshold quantization of high-frequency coefficients of wavelet decomposition.

[0024] For each high-frequency coefficient from the 1st layer to the Nth layer, a threshold is selected for threshold quantization. Python provides many adaptive methods for selecting the threshold of each layer. The quantization methods mainly include hard threshold quantization and soft threshold quantization. The two different methods are used to achieve the following effect: the hard threshold method can well preserve local features such as signal edges, while the soft threshold processing is relatively smooth, but it will cause distortion such as edge blur; Step 2-1-2-3, wavelet reconstruction of the signal. The mass spectrum signal is reconstructed by wavelet based on the low-frequency coefficients of the Nth layer of wavelet decomposition and the high-frequency coefficients of the 1st to Nth layers after quantization.

[0025] Step 2-2, construct and train the SDWT-GAN neural network; Based on the existing GAN, SDWT is used as the main part of data preprocessing, and a generator is added to the GAN part. TCNs , BiLSTM ] and loss function [ l 2, wgangp ] selection and feature verification part, where: TCNs Use temporal convolutional networks to process data; BiLSTM The data is processed using a bidirectional long short-term memory network; l 2. Solve using the mean square error loss function; wgangp Gradient penalty is used to replace the weight clipping in the original GAN ​​for solution.

[0026] Step 2-3, use the generator in GAN to generate { G 1, G 2,…, G i}and{ W 1, W 2,…, W i The dataset is passed into the discriminator for judgment to obtain the feature enhanced dataset { P 1, P 2,…, P i}; In order to solve the data imbalance problem in small sample signals, in this fault diagnosis method, an adversarial neural network is built to enhance data features. The signal after wavelet denoising is saved as the real sample of the generator. The specific steps of data enhancement are: Step 2-3-1, pass the noise signal into two generators composed of a temporal convolutional neural network and a bidirectional long short-term memory network respectively. The two generators are parallel and can obtain two types of generated samples required by the discriminator for selection in step 2-3-2; In step 2-3-2, the real samples and generated samples are input into the discriminator for judgment, and the appropriate generator network is selected as the network model for different feature samples according to the different types of real samples and the error size of the data output results.

[0027] Step 2-3-2, perform PCA and T-SNE dimensionality reduction processing on the output results.

[0028] This process can remove irrelevant features in the enhanced data set by further mining the data, while reducing redundant parameters in subsequent processing and improving computational efficiency.

[0029] Step 2-4, the obtained data set { P 1, P 2,…, P i}Input into the feature verification part constructed by PCA and T-SNE dimensionality reduction methods.

[0030] Step 3, generating a fault data set based on the fault characteristics and the characteristics of different components after generating data; Perform correlation analysis on multi-sensor signals and generate fault data sets based on fault characteristics and the characteristics of different components after generating data; Step 3 includes the following steps: Step 3-1, the original signal is obtained by using multiple sensors to collect data from each key component of the mass spectrometer. There is a certain correlation between the type and change trend of the signal. By establishing a visual correlation matrix, the strength of the sample correlation can be clearly observed. The samples are classified according to the strength of the correlation to obtain the fault sample data set. The data set of n types with different correlations is recorded as { X 1, X 2,…, X n}, { Y 1, Y 2,…, Y n}, …, { Z 1, Z 2,…, Z n}; Step 3-2, define the fault severity in each type of data set, label the data set through mapping relationships, and obtain the data sample set of the fault classification model.

[0031] The fault type labels of different data sets are X 1, X 2,…, X n}, { Y 1, Y 2,…, Y n}, …, { Z 1, Z 2,…, Z n} and for the failure data set of key components of mass spectrometers { X 1, X 2,…, X n}, { Y 1, Y 2,…, Y n}, …, { Z 1, Z 2,…, Z n}Establish a mapping relationship and define it as a training sample, where x , y , z represents the sample label of each type of fault dataset, X , Y , Z A signal sample representing different samples.

[0032] Step 4: construct an adaptive multi-channel fault classifier model and perform fault diagnosis; A multi-channel fault classification model is built using logistic regression, decision tree or random forest, K-nearest neighbor and neural network. Each type of fault sample is passed into the multi-channel fault classification model to obtain the fault diagnosis result. The specific steps include: Step 4-1, model pre-training.

[0033] After initializing the model parameters, input the training samples to obtain the pre-trained weights of the model. This weight is used as the initialization weight of step 4-2, which can enable the model to find the optimal channel more quickly and avoid overfitting.

[0034] Step 4-2, secondary training.

[0035] When the test samples are input into the classification model, the model will find the optimal channel as quickly as possible and compare the results with other channels. In this way, each type of fault data set may have a diagnostic channel suitable for its own fault characteristics, verifying the effectiveness of the model.

[0036] like Figure 3 As shown in the figure, through this fault diagnosis algorithm, by enhancing the samples of different fault types collected by the sensor, a data set containing 5000 samples of each data type is obtained, and the data set is divided into a training set and a test set in a 7:3 manner. From the figure, it can be seen that the classification of the model has a high accuracy rate. The classification accuracy rate for the data types of normal, missing and bias is 100%, while the classification accuracy rates for spike and drift fault types are 99% and 98% respectively.

[0037] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel, characterized by: The steps include: Step 1, using multiple sensors deployed in a spacecraft mass spectrometer to obtain characteristic data sets of multiple components; Step 2: Build a network model based on SDWT-GAN to enhance the feature data and enhance the features of small sample fault data; Step 3: perform correlation analysis on multi-sensor signals and generate a fault data set based on the fault characteristics and the characteristics of different components after generating data; Step 4: Build a multi-channel fault classification model, and pass each type of fault sample into the multi-channel fault classification model to obtain the fault diagnosis result.

2. The small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel according to claim 1 is characterized in that: In step 1, multiple components include a mass spectrometer injection assembly, an ion source, an analyzer, a multiplier tube, and a vacuum assembly.

3. The small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel according to claim 1 is characterized in that: Step 2 includes the following steps: Step 2-1, using the principle of S transform and discrete wavelet denoising to perform feature extraction and denoising on the data set collected by multiple sensors, reconstructing the mass spectrometry signal to obtain { W 1, W 2,…, W i }Noise reduction dataset; Step 2-2, construct and train the SDWT-GAN neural network; Step 2-3, use the generator in GAN to generate { G 1, G 2,…, G i }and{ W 1, W 2,…, W i The dataset is passed into the discriminator for judgment to obtain the feature enhanced dataset { P 1, P 2,…, P i }; Step 2-4, the obtained data set { P 1, P 2,…, P i }Input into the feature verification part constructed by PCA and T-SNE dimensionality reduction methods.

4. The small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel according to claim 3 is characterized in that: Step 2-1 includes the following steps: Step 2-1-1, S transform performs preliminary feature extraction on the original signal; Step 2-1-2, wavelet denoising.

5. The small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel according to claim 3 is characterized in that: Step 2-1-2 includes the following steps: Step 2-1-2-1, wavelet decomposition of the signal; Select a wavelet and determine the level N of wavelet decomposition, then perform N-level wavelet decomposition calculations on the signal. Different types of wavelets are selected for mass spectrometry signals with different characteristics through training; Step 2-1-2-2, threshold quantization of high-frequency coefficients of wavelet decomposition; For each high-frequency coefficient from the 1st layer to the Nth layer, a threshold is selected for threshold quantization processing; Step 2-1-2-3, wavelet reconstruction of the signal; based on the low-frequency coefficients of the Nth layer of wavelet decomposition and the high-frequency coefficients of the 1st to Nth layers after quantization, wavelet reconstruction of the mass spectrum signal is performed.

6. The small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel according to claim 3 is characterized in that: Step 2-3 includes the following steps: Step 2-3-1, pass the noise signal into two generators composed of a temporal convolutional neural network and a bidirectional long short-term memory network, and obtain two types of generated samples required by the discriminator: real samples and generated samples; Step 2-3-2, input the real samples and generated samples into the discriminator for judgment, and select the appropriate generator network as the network model of different feature samples according to the different types of real samples and the error size of the data output results; Step 2-3-2, perform PCA and T-SNE dimensionality reduction processing on the output results.

7. The small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel according to claim 1 is characterized in that: Step 3 includes the following steps: Step 3-1: By establishing a visual correlation matrix, we can observe the correlation strength of the samples, classify the samples by the correlation strength, and obtain the fault sample data set. The data set of n types with different correlations is recorded as { X 1, X 2,…, X n }, { Y 1, Y 2,…, Y n }, …, { Z 1, Z 2,…, Z n }; Step 3-2, define the fault severity in each type of data set, label the data set through the mapping relationship, and obtain the data sample set of the fault classification model: The fault type labels of different data sets are X 1, X 2,…, X n }, { Y 1, Y 2,…, Y n }, …, { Z 1, Z 2,…, Z n } and for the failure data set of key components of mass spectrometers { X 1, X 2,…, X n }, { Y 1, Y 2,…, Y n }, …, { Z 1, Z 2,…, Z n }Establish a mapping relationship and define it as a training sample, where x , y , z Represents the sample label of each type of fault dataset, X , Y , Z A signal sample representing different samples.

8. The small sample fault diagnosis method based on SDWT-GAN and adaptive multi-channel according to claim 1 is characterized in that: In step 4, the multi-channel fault classification model is built by using logistic regression, decision tree or random forest, K-nearest neighbor and neural network.

Images