MOS gas sensor array fault diagnosis method for electronic nose

By combining the sliding time window algorithm, Gated Transformer Network, and improved deep residual network, the problems of high false alarm rate and inaccurate isolation under multiple fault modes in MOS gas sensor array fault detection are solved, achieving efficient fault identification and type identification.

CN118937407AActive Publication Date: 2024-11-12HARBIN UNIV OF SCI & TECH
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202410982719.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-12
Estimated Expiration
2044-07-22

AI Technical Summary

Technical Problem

Existing fault detection methods for MOS gas sensor arrays suffer from high false alarm rates, inaccurate isolation under multiple fault modes, and the inability of simulated signals to fully reproduce the physical fault characteristics.

Method used

We employ a sliding time window (STW) algorithm combined with a gated transformer network (GTN) to distinguish between normal and fault signals by constructing a sliding time window dataset and a fault detection model. Then, we use a method combining Gram angle difference field and Markov transfer field to extract fault signal features and convert them into thermal images. Finally, we use an improved deep residual network IResNet for fault identification, and enhance feature representation and attention through a multi-scale channel attention mechanism and a local importance pooling module.

Benefits of technology

It effectively reduces the false alarm rate of fault detection, improves the accuracy and generalization ability of fault identification, and can accurately identify the fault type of sensor array.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118937407A_ABST
    Figure CN118937407A_ABST
Patent Text Reader

Abstract

The invention discloses an MOS gas sensor array fault diagnosis method for an electronic nose. The method comprises the following steps: 1, constructing an STW fault detection data set; 2, inputting the multi-dimensional response signal subjected to STW segmentation processing into a GTN fault detection model to obtain a fault detection result, and if the detection result is that a fault is detected, executing a step 3; 3, constructing a GADF-MTF fault identification data set; 4, constructing a fault identification network based on IResnet; and 5, inputting the GADF-MTF fault identification data set into the fault identification network based on the IResnet, identifying the response signal type of each sensor by using the fault identification network based on the IResnet according to the characteristics of the heat map, determining the position and the fault type of the fault sensor, and obtaining a fault identification result. The method can effectively reduce the false alarm rate of fault detection, has good fault identification performance, and is suitable for long-term popularization and application.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a gas sensor array fault diagnosis method, and in particular to a MOS gas sensor array fault diagnosis method based on a sliding time window algorithm (STW)-Gated Transformer Network (GTN) and GADF-MTF-IResnet (Gram angle difference field-Markov transfer field-improved deep residual network). Background Art

[0002] The gas sensor array is a key information acquisition component of the electronic nose system. Sudden failure of the sensor will cause the output signal of the sensor array to change, thus affecting the analysis accuracy of the existing intelligent analysis model. As the application scenarios of the electronic nose system become more and more complex, gas sensors will inevitably have sudden failures such as open welding of contact points, heater degradation, power supply fluctuations, etc. during long-term use, resulting in unstable sensor array response output. Therefore, it is necessary to monitor the abnormal state of the gas sensor array in the electronic nose system. Timely fault detection, isolation and diagnosis of sensors are of great significance to improving the reliability and maintainability of the electronic nose system.

[0003] Traditional fault detection, isolation and diagnosis methods based on PCA and similar methods use the normal signal of the MOS gas sensor array without the target gas to model. First, SPE or T 2 Statistics are used to detect faults, and the contribution graph method is used to isolate the faulty sensor. Finally, the classification algorithm is used to solve the fault diagnosis problem. However, the response of the MOS sensor array to the target gas during operation will also cause SPE or T 2 The mutation of statistics will cause a certain degree of false alarm of faults. Most of the existing MOS gas sensor fault detection methods do not consider the fault detection when the sensor responds normally to the gas. In addition, the fault isolation method based on the contribution graph has a high isolation accuracy for a single faulty sensor, but it cannot accurately isolate multiple faulty sensors in a multi-fault mode, and there are problems of false alarm or omission of fault location in the MOS gas sensor array. Finally, the existing research uses simulation software to superimpose simulation signals of different fault modes on the normal response signal to obtain experimental samples, which cannot completely reproduce the characteristics of the fault signal at the physical level. Summary of the invention

[0004] In view of the above problems existing in the traditional MOS gas sensor array fault detection method and the fault isolation method based on contribution graph, the purpose of the present invention is to provide a MOS gas sensor array fault diagnosis method for electronic nose. The method can effectively reduce the false alarm rate of fault detection, has good fault identification performance, and is suitable for long-term promotion and application.

[0005] The objective of the present invention is achieved through the following technical solutions:

[0006] A MOS gas sensor array fault diagnosis method for an electronic nose comprises the following steps:

[0007] Step 1: Construct a STW fault detection dataset and convert the sensor array multidimensional response signal dataset into a sliding time window dataset;

[0008] Step 2: Input the multidimensional response signal after STW segmentation processing into the GTN fault detection model, use the GTN fault detection model to implement window-by-window fault detection, accurately distinguish the multidimensional time series segments of normal signals and fault signals, and obtain the fault detection result. If the detection result is that a fault is detected, execute step 3;

[0009] Step 3: Construct a GADF-MTF fault recognition dataset, and convert the sensor array multi-dimensional response signal dataset into a GADF-MTF thermal image dataset;

[0010] Step 4: Build a fault identification network based on IResnet;

[0011] Step 5: Input the GADF-MTF fault identification data set into the fault identification network based on IResnet. According to the heat map features, the fault identification network based on IResnet is used to identify the response signal type of each sensor, determine the location and fault type of the faulty sensor, and obtain the fault identification result.

[0012] Compared with the prior art, the present invention has the following advantages:

[0013] 1. The present invention utilizes the fault detection model based on STW and GTN to fully extract the time and space correlation characteristics of the time window samples, thereby effectively distinguishing the normal response signal of the sensor array from the fault signal, and solving the problem of high false alarm rate of the sensor array's normal response to the target gas during the fault detection process.

[0014] 2. The GADF-MTF signal encoding method combines the time-dependence characteristics of GADF for signals and the first-order Markov chain transition probability characteristics of MTF for signals, enhancing the feature expression capability of sensor array response signals and effectively highlighting the inter-class differences of signals.

[0015] 3. The present invention utilizes PSA to improve the deep residual network's ability to express multi-scale information of feature maps, and at the same time enhances the network's effective attention to important feature information through LIP, effectively improving the fault diagnosis model's recognition accuracy and generalization ability for fault signal types. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of the MOS gas sensor array fault diagnosis method for electronic nose proposed by the present invention.

[0017] Figure 2 This is a diagram of the overall network architecture proposed by the present invention.

[0018] Figure 3 This is the IResnet network structure diagram proposed by the present invention.

[0019] Figure 4 It is a flow chart of the overall implementation of the present invention.

[0020] Figure 5 This is a structural diagram of the GTN fault detection model used in the present invention.

[0021] Figure 6 This is the implementation process of the LIP module.

[0022] Figure 7 It is a PSA module;

[0023] Figure 8 The fault detection results of six types of signals are: a) heating power change, b) load resistance change, c) pin short circuit, d) power failure, e) voltage fluctuation, and f) normal response. DETAILED DESCRIPTION

[0024] The technical solution of the present invention is further described below in conjunction with the accompanying drawings, but is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention should be included in the protection scope of the present invention.

[0025] The present invention provides a MOS gas sensor array fault diagnosis method for an electronic nose. First, STW is combined with GTN for fault detection. The multidimensional response signal of the sensor array is segmented into a group of time window samples with high time correlation characteristics by STW, and then GTN is used to fully extract the time and space correlation characteristics of the multivariate time series, accurately distinguish the multidimensional time series fragments of normal signals and fault signals, and effectively solve the problem of high fault false alarm rate of normal gas response signals. Then, the Gram angle difference field is combined with the Markov transfer field to extract fault signal features, and the original multidimensional fault signal is converted into a GADF-MTF thermal image channel by channel. The image contains both the time dependency characteristics of the gas sensor response signal and the transition probability characteristics of the first-order Markov chain, which effectively increases the inter-class difference of the fault signal. Finally, the improved deep residual neural network model is called channel by channel to realize the identification of each sensor fault type. The model uses ResNet50 as the backbone network, introduces a multi-scale channel attention mechanism (PSA) to improve the network's ability to express features of different scales in thermal images, and then uses the local importance pooling module (LIP) to enhance the network's effective attention to important fault feature information, which can achieve effective fault identification. Figure 1 As shown, the specific steps of the method are as follows:

[0026] A1. Construct a STW fault detection dataset, convert the sensor array multi-dimensional response signal dataset into a sliding time window dataset, and divide it into a training set and a test set according to a certain ratio. The specific steps are as follows:

[0027] B1. Obtain multi-dimensional response fault signals and normal signals of the MOS gas sensor array.

[0028] In this step, the multi-dimensional response fault signals used are all obtained through the electronic nose experimental system combined with the fault generation device.

[0029] B2. Use STW to segment the multidimensional response fault signal into continuous time window samples.

[0030] In this step, the original multidimensional response signal data set is divided into a continuous sliding time window data set. The advantage of STW is that it can not only realize real-time fault detection window by window, but also retain the high spatiotemporal correlation characteristics of the original multidimensional response signal.

[0031] B3. Assign labels to the segmented time window samples and randomly divide the multi-dimensional response fault signals into training sets and test sets.

[0032] In this step, the segmented time window samples are labeled with faulty or normal class labels in turn, and the data set is randomly divided into training set and test set using a script.

[0033] A2. Input the multi-dimensional response signal after STW segmentation processing into the GTN fault detection model to obtain a fault detection result. If the detection result is that a fault is detected, execute step A3 to enter the fault identification and isolation link.

[0034] Figure 5 This is a structural diagram of the GTN fault detection model. The model retains the encoder structure in the original Transformer, removes the decoder part, and expands the model into a dual-tower structure by adding a set of channel encoders in parallel. At the same time, the Gate mechanism is introduced to generate a two-dimensional gating signal by learning the output features of the two towers to control the flow of feature information in the subsequent network to achieve the optimal feature fusion effect.

[0035] During the gating process, the dual-tower outputs Encoding1 and Encoding2 are first concatenated together through Concat. Subsequently, the two-dimensional gating signal is obtained through linear mapping of the Gate layer, and converted into gating probabilities g1 and g2 using the Softmax function, which correspond to the first and second columns of probability information of the two-dimensional gating signal, respectively. Finally, Encoding1 and Encoding2 are weightedly combined according to the gating probabilities g1 and g2 to obtain the fully fused dual-tower feature output result Encoding. Encoding is processed through the Linear layer to obtain the final output result of the GTN fault detection model, that is, the probability that the input sample is predicted to be the corresponding category.

[0036] The entire gating process is expressed as follows using formula (1):

[0037]

[0038] Among them, C and S represent the output feature vectors of the dual towers of Encoding1 and Encoding2 respectively, W represents the weight configuration in the linear mapping of the Gate layer, b represents the bias vector in the linear mapping, h is the two-dimensional gating signal obtained by linear mapping, and g1 and g2 are the gating probabilities calculated by the Softmax function. The output feature vectors of the dual towers are concatenated according to the gating probabilities to obtain the final output vector y that fully integrates the feature information of the dual towers, and then linear processing is performed on it to obtain the final classification result.

[0039] The dual-tower structure of GTN enables the model to simultaneously capture the temporal and spatial related information of multi-dimensional time series data. The Gate mechanism can fully integrate the dual-tower feature information to improve the final classification performance of the model.

[0040] A3. Construct a GADF-MTF fault recognition dataset, convert the sensor array multi-dimensional response signal dataset into a GADF-MTF thermal image dataset, and divide it into a training set and a test set according to a certain ratio. The specific steps are as follows:

[0041] B4. Separate the multi-dimensional response fault signal into a single-channel one-dimensional signal, and process the single-channel one-dimensional signal using GADF and MTF signal coding methods respectively.

[0042] In this step, GADF can retain the time-dependent characteristics of the data, and at the same time has greater sparsity, eliminating redundant information between multiple modalities. MTF fully obtains the state change information of the time series in different periods by calculating the transition probability between the current state and the remaining states.

[0043] B5. Perform weighted fusion of GADF and MTF feature matrices to generate a GADF-MTF thermal image.

[0044] In this step, the thermal image generated by weighted fusion of feature matrices contains both the time-dependent characteristics of the sensor response signal and the transition probability characteristics of the first-order Markov chain. The feature information is richer and the inter-class differences of fault signals are effectively amplified.

[0045] B6. Assign fault category labels to the GADF-MTF thermal image samples and randomly divide them into training and test sets.

[0046] In this step, the generated thermal image samples are labeled with fault categories, and the script is used to randomly divide the dataset into training and test sets.

[0047] A4. Construction Figure 3 The fault identification network based on IResnet shown in the figure uses Resnet50 as the basic model, optimizes its convolutional layer network structure, and introduces LIP and PSA modules. The specific steps are as follows:

[0048] B7. Change the network structure of the traditional Resnet50 and replace the first convolutional layer of the traditional Resnet50 with two convolutional layers with a convolution kernel size of 3×3 to enhance the expressive power of the network in the initial stage and reduce model parameters.

[0049] B8. Introduce the LIP module to optimize the traditional pooling structure and residual structure.

[0050] like Figure 6 As shown in Figure 1, the LIP module automatically enhances the recognition features by adaptively learning the importance weights of the input tensor during the downsampling process of the neural network. This module can determine which features are more important during the downsampling process, thereby automatically enhancing and retaining key features.

[0051] The LIP module actively learns feature discrimination criteria through the network G to avoid manual control of the importance function. G is named the logit module, which is implemented by a small fully convolutional network (FCN). The logit module can learn the features of each pixel in the input image and generate an importance feature map. In order to make the importance weight non-negative and easy to optimize, the exp(·) operation is added after the network G to obtain:

[0052] W=F(I)=exp(G(I)) (2)

[0053] Figure 6 The implementation process of LIP is shown. First, the tensor I is input into the logit module to obtain the corresponding importance weight W. Then, the product of the importance weight and the input tensor I×W and W are average pooled respectively, and the result of the I×W average pooling is divided by the result of the W average pooling to obtain the final output of the LIP module:

[0054]

[0055] B9. Insert PSA modules before and after each residual module.

[0056] like Figure 7 As shown in the figure, the PSA module can fully extract the multi-scale spatial information of the input feature map and enhance the expression ability of channel features. The implementation steps of this module are as follows:

[0057] (1) Use the SPC module to transform the input feature map in the channel dimension X Divide into S parts, represented as [X0,X1,…,X S-1 ]. The initial number of channels is C, and the number of channels in each part after segmentation is Use a convolution kernel size of K i The multi-scale grouped convolution (i=0,1,…,S-1) extracts the spatial information of different scales of each part of the channel feature map, and the number of groups corresponding to each part is G i (i=0,1,…,S-1), as shown in formula (4). The output feature maps of each convolutional layer are concatenated in the channel dimension to obtain a new feature map containing multi-scale channel features, as shown in formula (5).

[0058] F i =Conv(K i ×K i ,G i )(X i ),i=0,1,2,…,S-1,F i ∈R C'×H×W (4)

[0059] F=Cat(F0,F1,…,FS-1 ) (5)

[0060] (2) Use the SEWeight module to calculate the channel attention weights corresponding to the output feature maps of each convolutional layer, as shown in formula (6), and concatenate these weights in the channel dimension, as shown in formula (7).

[0061] Z i =SEWeight(F i ),i=0,1,2,…,S-1,Z i ∈R C×1×1 (6)

[0062] Z=Z0+Z1+…+Z S-1 (7)

[0063] (3) Use Softmax to normalize the multi-scale channel attention vector to obtain a new attention vector, as shown in formula (8).

[0064]

[0065] (4) The feature map F i And the attention vector att obtained after normalization i Multiply them in the channel dimension to obtain the feature map after multi-scale channel attention weighting, as shown in formula (9). The above feature maps are concatenated in the channel dimension to output a feature map with richer multi-scale information, as shown in formula (10).

[0066] Y i =F i ⊙att i ,i=0,1,2,…,S-1 (9)

[0067] Out=Cat(Y0,Y1,…,Y S-1 ) (10)

[0068] A5. Input the GADF-MTF fault identification data set into the fault identification network based on IResnet to obtain the fault identification results. Figure 4 As shown, it can be divided into the following steps:

[0069] First, input the STW fault detection dataset after STW segmentation;

[0070] Then, the fault detection results are obtained using the GTN fault detection model;

[0071] Secondly, the response signal is preprocessed by GADF-MTF method;

[0072] Then, the GADF-MTF fault identification dataset is input;

[0073] Finally, the fault type identification results are obtained using the fault identification network based on IResnet.

[0074] Example:

[0075] Dataset:

[0076] This embodiment uses the electronic nose experimental system with a fault generation device to construct a fault detection data set and a fault identification data set. Among them, the original fault signal data set mainly includes five types: power failure fault, voltage fluctuation fault, load resistance change fault, heating power change fault and pin short circuit fault.

[0077] In order to discuss the influence of sliding time window parameters on the GTN fault detection model, this embodiment selects sliding time window widths of 100, 200, and 300, and window moving steps of 10 and 20 to perform sliding time window segmentation on the MOS gas sensor array response signal, and uses the time window samples segmented from the training samples to train the GTN fault detection model. Finally, the performance of the GTN fault detection model is tested using the test samples.

[0078] The GADF-MTF signal encoding method is used to convert the response signal of a single sensor into a heat map, and a fault identification dataset is constructed based on it. The dataset contains a total of 300 samples, including 210 samples in the training set and 90 samples in the test set.

[0079] Evaluation indicators:

[0080] (1) Fault detection evaluation indicators

[0081] Because the main goal of fault detection is to identify samples with faults, fault samples are usually regarded as positive events and normal samples are regarded as negative events. The present invention selects fault detection rate (FDR), false positive rate (FPR) and false negative rate (FNR) as fault detection evaluation indicators, as shown in formulas (11)(12)(13).

[0082]

[0083] Among them, the meanings of TP, TN, FP, and FN are as follows:

[0084] (1) True Positive (TP): A true positive sample is a sample that is actually a fault sample and is correctly predicted as a fault sample by the model.

[0085] (2) True Negative (TN): True negative examples. They are actually normal samples and are correctly predicted as normal samples by the model.

[0086] (3) False Positive (FP): False positive examples. They refer to samples that are actually normal but are mistakenly predicted by the model as faulty samples. This situation is usually called "false positive" or "first type error".

[0087] (4) False Negative (FN): False negative examples. They are actually faulty samples, but are mistakenly predicted as normal samples by the model. This situation is usually called "false negative" or "second type error".

[0088] (2) Fault identification evaluation indicators

[0089] Fault type identification uses Precision and Recall as model performance evaluation indicators.

[0090]

[0091] Among them, the meanings of TP, FP, and FN are as follows:

[0092] (1) TP: indicates that samples in a specific fault category are correctly predicted by the model as belonging to that category.

[0093] (2) FP: Indicates the situation where samples in other fault categories are incorrectly predicted by the model as a specific fault category.

[0094] (3) FN: Indicates the situation where samples in a specific fault category are incorrectly predicted by the model as other categories.

[0095] Experimental results and analysis:

[0096] (1) Fault detection experiment

[0097] In order to verify the performance of the fault detection model, six sliding time windows of different sizes are used to segment the samples, and the average fault detection rate, average fault false alarm rate and average fault omission rate of various response signals are obtained as shown in Tables 1, 2 and 3.

[0098] Table 1 Average fault detection rate of fault detection model under different sliding time window sizes

[0099]

[0100] As shown in Table 1, the GTN fault detection model trained with six different sliding window sizes can achieve a high fault detection rate for various response signals. As the window width and moving step size increase, the overall average fault detection rate increases. Among them, when the signal is segmented using a sliding time window with a width of 100, the fault detection rates of the two types of response signals, voltage fluctuation fault and load resistance change fault, are significantly lower than those of other types of signals. The above two types of signal samples both contain minor fault samples. In the case of unstable power supply and environmental noise interference, the feature information contained in the time window samples with a width of 100 is difficult to distinguish minor fault samples from normal samples, resulting in a low fault detection rate for the two types of response signals.

[0101] Table 2 Average fault false alarm rate of fault detection model under different sliding time window sizes

[0102]

[0103] The size of the average fault false alarm rate shown in Table 2 reflects the recognition effect of the GTN fault detection model on the normal response time window samples under different sliding time window sizes. As the time window width increases, the average fault false alarm rate shows an overall downward trend. Consistent with the average fault detection rate, the false alarm rate of the GTN fault detection model for voltage fluctuation faults and load resistance change faults is significantly higher than that of other types of signals.

[0104] Table 3 Average fault omission rate of fault detection model under different sliding time window sizes

[0105]

[0106] Table 3 reflects the fault omissions of the GTN fault detection model under six different sliding time window sizes. As the window width increases and the step size increases, the overall average fault omission rate shows a downward trend.

[0107] When using a sliding time window of size (w=300, s=10), the fault detection effects of the six types of samples are shown as follows: Figure 8 As shown in the figure, 0 indicates a normal state and 1 indicates a fault state. Figure 8It can be seen that the fully trained GatedTransformer model can perform real-time fault detection on the signals collected from the sensor array, and the six types of sensor signals can achieve good detection results. In addition, the model is sensitive to the time point of the fault burst, and can more accurately judge the critical state of the fault and detect when the fault occurs. This model solves the problem that traditional fault detection models such as PCA and KPCA are difficult to solve, that is, the problem of effectively distinguishing normal gas response signals from fault signals. The state of the normal gas response signal in each time window is displayed as 0, that is, normal and fault-free, which effectively solves the problem of high misjudgment rate of normal gas response faults.

[0108] (2) Fault identification experiment

[0109] In order to verify the superiority of the MOS gas sensor fault identification model proposed in the present invention, it is compared with a variety of other models.

[0110] Table 4 Comparison of fault identification results of IResnet and other models

[0111]

[0112] It can be seen from the results in Table 4 that the precision and recall rates of the other nine fault recognition models are lower than the IResnet fault recognition model proposed in the present invention. When processing thermal images, the convolutional neural network model retains the characteristics of translation invariance and correlation between local pixels, while the Transformer model lacks these features, resulting in poor recognition performance on smaller data sets. Although the deeper the network depth, the stronger the feature extraction ability, it is also more prone to overfitting, especially when the amount of training data is relatively small. Therefore, models with moderate network depth may be more suitable for solving recognition problems with limited training data. The fault recognition model proposed in the present invention changes the initial layer convolution structure on the basis of the traditional Resnet50, and enhances the network's ability to express image features in the initial stage. After introducing the attention mechanism, the feature space is enriched by effectively acquiring and utilizing feature information of different scales, and the network's attention to important feature information is enhanced by utilizing local importance pooling, which effectively improves the fault recognition performance of the model.

[0113] The IResnet fault diagnosis model proposed in this invention adds LIP and PSA modules to the original Resnet50 network, and replaces the 7×7 convolution of the first layer of the network with two 3×3 convolutions. Table 5 verifies the effectiveness of this improved method through ablation experiments, where No means not to perform this operation and Yes means to perform this operation.

[0114] Table 5 Ablation experiment

[0115]

[0116] As can be seen from Table 5, when the three improvements in Model2 to Model4 are applied to the original model separately, they can achieve certain improvement effects. When the three operations are combined in pairs, the classification effect obtained is improved over that of a single operation. When the three operations are applied to the original model together, the diagnostic accuracy and recall rate reach 100%. The above ablation experiment proves the effectiveness of the improved Resnet50 fault diagnosis model proposed in the present invention.

Claims

1. A MOS gas sensor array fault diagnosis method for electronic nose, characterized in that The method comprises the following steps: Step 1: Construct a STW fault detection dataset and convert the sensor array multidimensional response signal dataset into a sliding time window dataset; Step 2: Input the multidimensional response signal after STW segmentation processing into the GTN fault detection model, use the GTN fault detection model to implement window-by-window fault detection, accurately distinguish the multidimensional time series segments of normal signals and fault signals, and obtain the fault detection result. If the detection result is that a fault is detected, execute step 3; Step 3: Construct a GADF-MTF fault recognition dataset, and convert the sensor array multi-dimensional response signal dataset into a GADF-MTF thermal image dataset; Step 4: Build a fault identification network based on IResnet; Step 5: Input the GADF-MTF fault identification data set into the fault identification network based on IResnet. According to the heat map features, the fault identification network based on IResnet is used to identify the response signal type of each sensor, determine the location and fault type of the faulty sensor, and obtain the fault identification result.

2. The MOS gas sensor array fault diagnosis method for electronic nose according to claim 1 is characterized in that The specific steps of step one are as follows: Step 11, obtaining a multi-dimensional response fault signal and a normal signal of the MOS gas sensor array; Step 1 and 2: Use STW to segment the multidimensional response fault signal into continuous time window samples; Step 1: Assign labels to the segmented time window samples and randomly divide the multi-dimensional response fault signals into training sets and test sets.

3. The MOS gas sensor array fault diagnosis method for electronic nose according to claim 2 is characterized in that In the steps 1 and 3, the segmented time window samples are labeled with fault class or normal class in turn, and the data set is randomly divided into a training set and a test set using a script.

4. The MOS gas sensor array fault diagnosis method for electronic nose according to claim 1 is characterized in that The specific steps of step three are as follows: Step 31: Separate the multi-dimensional response fault signal into a single-channel one-dimensional signal, and process the single-channel one-dimensional signal using GADF and MTF signal coding methods respectively; Step 32: Perform weighted fusion of GADF and MTF feature matrices to generate a GADF-MTF thermal image; Step 3: Assign fault category labels to the GADF-MTF thermal image samples and randomly divide them into training sets and test sets.

5. The MOS gas sensor array fault diagnosis method for electronic nose according to claim 1 is characterized in that In step 33, the generated thermal image samples are labeled with fault categories, and a script is used to randomly divide the data set into a training set and a test set.

6. The MOS gas sensor array fault diagnosis method for electronic nose according to claim 1, characterized in that The specific steps of step 4 are as follows: Step 41: Change the network structure of the traditional Resnet50 and replace the first convolutional layer of the traditional Resnet50 with two convolutional layers with a convolution kernel size of 3×3 to enhance the expression ability of the network in the initial stage and reduce the model parameters; Step 42: Introduce the LIP module to optimize the traditional pooling structure and residual structure; Step 43: Insert PSA modules before and after each residual module.

7. The electronic nose-oriented MOS gas sensor array fault diagnosis method according to claim 6, characterized in that The LIP module actively learns the feature discrimination criterion through the network G, which is named the logit module. The exp(·) operation is added after the network G to obtain: W = F (I) = exp (G (I)) (2); First, the tensor I is input into the logit module to obtain the corresponding importance weight W. Then, the product of the importance weight and the input tensor I×W and W are average pooled, and the result of the I×W average pooling is divided by the result of the W average pooling to obtain the final output of the LIP module:

8. The method for fault diagnosis of MOS gas sensor array for electronic nose according to claim 6, characterized in that The implementation steps of the PSA module are as follows: (1) Use the SPC module to transform the input feature map in the channel dimension X Divide into S parts, represented as [X0,X1,…,X S-1 ], the initial number of channels is C, and the number of channels in each part after segmentation is Use a convolution kernel size of K i The multi-scale grouped convolution (i=0,1,…,S-1) extracts the spatial information of different scales of each part of the channel feature map, and the number of groups corresponding to each part is G i (i=0,1,…,S-1), as shown in formula (4), the output feature maps of each convolutional layer are concatenated in the channel dimension to obtain a new feature map containing multi-scale channel features, as shown in formula (5): F i =Conv(K i ×K i ,G i )(X i ),i=0,1,2,…,S-1,F i ∈R C’×H×W (4); F=Cat(F0,F1,…,F S-1 ) (5); (2) Use the SEWeight module to calculate the channel attention weights corresponding to the output feature maps of each convolutional layer, as shown in formula (6), and concatenate these weights in the channel dimension, as shown in formula (7): Z i =SEWeight(F i ),i=0,1,2,…,S-1,Z i ∈R C′×1×1 (6); Z=Z0+Z1+…+Z S-1 (7); (3) Use Softmax to normalize the multi-scale channel attention vector to obtain a new attention vector, as shown in formula (8): (4) The feature map F i And the attention vector att obtained after normalization i Multiply them in the channel dimension to obtain the feature map after multi-scale channel attention weighting, as shown in formula (9). Concatenate the above feature maps in the channel dimension to output a feature map with richer multi-scale information, as shown in formula (10): Y i =F i ⊙att i ,i=0,1,2,…,S-1 (9); Out=Cat(Y0,Y1,…,Y S-1 ) (10)。

Citation Information

Patent Citations

  • Fault detection and diagnosis method for chemical sensor array

    CN112098600A

  • Multi-source multi-domain information entropy fusion and model self-optimization method for bearing fault diagnosis

    CN114048769A

  • Rolling bearing fault diagnosis method and system and computer storage medium

    CN115855505A

  • Intelligent gearbox fault diagnosis method based on multi-sensor fusion

    CN117890099A

  • Distribution transformer operation fault telemetering method and system

    CN118130940A