Industrial process fault classification method based on trend and amplitude consistency fusion

Abstract

The invention provides an industrial process fault classification method and device based on trend and amplitude consistency fusion, and relates to the technical field of industrial process fault diagnosis. The method comprises the following steps: preprocessing historical production data based on a sliding window method and a Z-score standardization algorithm to obtain training data; according to the training data, performing data enhancement on the data enhancement model to obtain an optimized data enhancement model and second enhancement data; performing redundant data elimination on the second enhanced data based on a minimum redundancy and maximum correlation analysis method to obtain optimized data; training the fault classification model by using the optimized data to obtain an optimized fault classification model; and according to actual production data, fault classification prediction is carried out through the optimized data enhancement model and the optimized fault classification model, and an actual fault classification result is obtained. The method is an efficient and accurate industrial process fault classification method for nonlinear industrial process tiny faults.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial process fault diagnosis, and in particular to an industrial process fault classification method and device based on trend and amplitude consistency fusion. Background Art

[0002] Fault detection and quantitative evaluation technology for quality anomalies is an effective means to ensure safe production and obtain reliable product quality. It is one of the current research hotspots in the field of process control. The nonlinear and high-dimensional characteristics of complex industrial processes have brought great challenges to its application. Therefore, it is necessary to establish an accurate and reliable monitoring model for quality anomaly fault diagnosis to provide support for on-site maintenance.

[0003] Process monitoring is a marginal discipline developed based on Fault Detection and Diagnosis (FDD) technology. Its main motivation is to monitor the operating status of the system, continuously provide quantitative and qualitative analysis, and help process operators and managers to understand the operating status of the process in a timely manner, so as to eliminate abnormal behavior of the process, prevent catastrophic accidents and reduce fluctuations in product quality. The research content of process monitoring generally includes fault detection, fault diagnosis and identification, fault separation, fault countermeasures and repair, etc.

[0004] The entire process of complex industry is organically connected by multiple production equipment or processes. The process is long, the working conditions are complex and changeable, the processes are coupled with each other, and the integrated automation system is multi-level and cooperatively related. As enterprises have lower and lower tolerance for performance degradation, productivity decline and safety hazards, the requirements for the reliability and safety of control systems are getting higher and higher. With the emergence of computer control, communication networks and information technology, a large amount of operating data related to process status has been collected, which not only provides the possibility for the emergence of new FDD methods, but also brings huge challenges. However, traditional monitoring methods have poor recognition performance for small faults with small amplitudes, slow changes, unclear fault characteristics, and easy to be disturbed or masked by noise. Many industrial process faults have a slow change process. If they can be detected and diagnosed in time at the early stage of the fault, it will help avoid the occurrence of serious faults.

[0005] Therefore, the prior art lacks an efficient and accurate industrial process fault classification method for minor faults in nonlinear industrial processes. Summary of the invention

[0006] In order to solve the technical problem that the prior art has poor recognition performance for small faults with small amplitude, slow change, unclear fault characteristics, and easy to be disturbed or masked by noise, the embodiment of the present invention provides an industrial process fault classification method and device based on trend and amplitude consistency fusion. The technical solution is as follows:

[0007] On the one hand, a method for industrial process fault classification based on trend and amplitude consistency fusion is provided, the method is implemented by an industrial process fault classification device, and the method includes:

[0008] Acquire historical production data of the industrial process; preprocess the historical production data based on a sliding window method and a Z-score normalization algorithm to obtain training data;

[0009] According to the training data, data enhancement is performed on the data enhancement model based on the fusion trend consistency error and the amplitude consistency error to obtain an optimized data enhancement model and second enhanced data;

[0010] Eliminate redundant data from the second enhanced data based on a minimum redundancy maximum correlation analysis method to obtain optimized data;

[0011] Using the optimized data, training a fault classification model to obtain an optimized fault classification model;

[0012] In the actual production process, data is collected from the industrial control server to obtain actual production data of the industrial process; based on the actual production data, fault classification prediction is performed through the optimized data enhancement model and the optimized fault classification model to obtain actual fault classification results.

[0013] On the other hand, an industrial process fault classification device based on trend and amplitude consistency fusion is provided, and the device is applied to the industrial process fault classification method based on trend and amplitude consistency fusion, and the device includes:

[0014] A data processing module is used to obtain historical production data of industrial processes; based on a sliding window method and a Z-score normalization algorithm, the historical production data is preprocessed to obtain training data;

[0015] A data enhancement module, used to perform data enhancement on a data enhancement model based on a fusion trend consistency error and an amplitude consistency error according to the training data, to obtain an optimized data enhancement model and second enhanced data;

[0016] A data optimization module, configured to eliminate redundant data from the second enhanced data based on a minimum redundancy maximum correlation analysis method to obtain optimized data;

[0017] A model optimization module, used to train a fault classification model using the optimization data to obtain an optimized fault classification model;

[0018] The fault classification module is used to collect data from the industrial control server during the actual production process to obtain the actual production data of the industrial process; based on the actual production data, fault classification prediction is performed through the optimized data enhancement model and the optimized fault classification model to obtain the actual fault classification result.

[0019] On the other hand, an industrial process fault classification device is provided, which includes: a processor; a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, any one of the above-mentioned industrial process fault classification methods based on trend and amplitude consistency fusion is implemented.

[0020] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement any one of the above-mentioned industrial process fault classification methods based on trend and amplitude consistency fusion.

[0021] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0022] The present invention proposes an industrial process fault classification method based on the fusion of trend and amplitude consistency, combining autoencoders and adversarial generative networks to enhance the data of industrial processes, solving the problem of few data samples and much interference in the manufacturing process; considering the consistency of autocorrelation and cross-correlation, a new pre-training loss function is designed to characterize the features involved in the trend fault signal, making the extracted features more accurate; when only a small number of fault samples are contaminated, due to the presence of strong noise, the present invention can provide a more advanced learning mechanism for the deep neural network to accurately extract the potential weak features of minor faults. The present invention is an efficient and accurate industrial process fault classification method for minor faults in nonlinear industrial processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 It is a flow chart of an industrial process fault classification method based on trend and amplitude consistency fusion provided by an embodiment of the present invention;

[0025] Figure 2 It is a block diagram of an industrial process fault classification device based on trend and amplitude consistency fusion provided by an embodiment of the present invention;

[0026] Figure 3 It is a structural schematic diagram of an industrial process fault classification device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0028] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0029] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0030] In the embodiments of the present invention, sometimes the subscripts such as W 1 It may be written in non-subscript form such as W1. When the difference is not emphasized, the meaning is the same.

[0031] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0032] The embodiment of the present invention provides an industrial process fault classification method based on trend and amplitude consistency fusion, which can be implemented by an industrial process fault classification device, which can be a terminal or a server. Figure 1 The flowchart of the industrial process fault classification method based on trend and amplitude consistency fusion is shown. The processing flow of the method may include the following steps:

[0033] S1. Obtain historical production data of industrial processes; preprocess the historical production data based on the sliding window method and Z-score normalization algorithm to obtain training data.

[0034] Optionally, based on a sliding window method and a Z-score normalization algorithm, historical production data is preprocessed to obtain training data, including:

[0035] Based on the random partitioning method, the historical production data is randomly partitioned to obtain the partitioned historical data;

[0036] Based on the preset sliding window, the divided historical data is cleaned to obtain the cleaned historical data;

[0037] Based on the cleaned historical data, the Z-score standardization algorithm is used to standardize the data to obtain training data.

[0038] In a feasible implementation, the industrial process data includes data of several process variables and manipulated variables, and the data set is divided using a random partitioning method, and the data is cleaned using a sliding window algorithm, and the data set is gradually traversed through a preset window. , determine the fixed length The window starts from the beginning of the data set and slides one step each time. , at each window position, calculate the mean of the data in the window, for k windows , calculate the required mean statistic As shown in the following formula (1):

[0039] (1);

[0040] Among them, the starting position of the kth window is , the end position is , the data of the kth window is , based on which a new dataset is constructed .

[0041] The data were normalized using the Z-score algorithm to transform the data into a standard normal distribution with zero mean and unit variance. As shown in the following formula (2):

[0042] (2);

[0043] in, is the mean of the data set, is the variance of the data set.

[0044] S2. According to the training data, data enhancement is performed on the data enhancement model based on the fusion trend consistency error and the amplitude consistency error to obtain an optimized data enhancement model and second enhanced data.

[0045] Among them, the model structure of the data enhancement model is constructed based on the autoencoder structure and the generative adversarial network structure;

[0046] The data augmentation model consists of an encoder part, a generator part, and a discriminator part;

[0047] The encoder part is used to map the input model data into a low-dimensional latent space;

[0048] The generator part is used to generate enhanced data based on the data in the latent space;

[0049] The discriminator part is used to guide the generator to generate augmented data.

[0050] In a feasible implementation manner, the present invention proposes a data enhancement model based on the fusion of trend consistency error and amplitude consistency error to achieve data enhancement.

[0051] The autoencoder has a symmetrical 3-layer unsupervised neural network structure, which is divided into an input layer, a hidden layer, and an output layer; the hidden layer maps the input data to a low-dimensional latent space, and the output layer reconstructs the input data from the representation of the latent space. The training goal of the network is to minimize the difference between the input and the reconstructed output. The network parameters are usually adjusted by minimizing the reconstruction error so that the hidden layer learns the input features and obtains the best data expression.

[0052] When the input data is , the hidden layer maps the input data x to a representation z in a latent space. The mathematical expression of the mapped data is shown in the following formula (3):

[0053] (3);

[0054] in, is the encoder weight matrix, is the bias vector of the hidden layer, is the Relu activation function.

[0055] The task of the output layer is to restore the representation z of the latent space to an output similar to the input data. , the mathematical expression of this process is shown in the following formula (4):

[0056] (4);

[0057] in, is the weight matrix of the decoder, is the bias vector of the output layer, is the Relu activation function. The error loss between the input data and the reconstructed data is .

[0058] The data enhancement model of the autoencoder and the generative adversarial network uses the advantages of both to generate higher quality data. By combining the generative ability of the autoencoder with the discriminative ability of the generative adversarial network (GAN), more realistic data can be generated. It consists of three parts: the encoder part, the generator part, and the discriminator part.

[0059] Optionally, according to the training data, data enhancement is performed on the data enhancement model based on the fusion trend consistency error and the amplitude consistency error to obtain an optimized data enhancement model and second enhanced data, including:

[0060] Use the training data to train the data enhancement model to obtain an optimized data enhancement model;

[0061] The training data is input into the optimized data enhancement model for data enhancement to obtain the second enhanced data.

[0062] In a feasible implementation, before performing data enhancement on the training data, it is necessary to use the training data to train the data enhancement model to ensure that the second enhanced data generated based on the training data can extract small fault features with small amplitude, slow changes, and disturbed or masked by noise.

[0063] Optionally, the data enhancement model is trained using the training data to obtain an optimized data enhancement model, including:

[0064] Input the training data into the encoder part for latent space mapping to obtain mapping data;

[0065] Based on the discriminator part, according to the mapping data, the generator part generates data to obtain first enhanced data;

[0066] Calculate the amplitude consistency error based on the training data and the first enhanced data;

[0067] Calculate the trend consistency error according to the training data and the first enhanced data; the trend consistency error includes an autocorrelation error and a cross-correlation error;

[0068] The loss function is calculated based on the amplitude consistency error and the trend consistency error to obtain the data enhancement loss;

[0069] According to the data enhancement loss, the data enhancement model is reversely optimized to obtain the optimized data enhancement model.

[0070] In a feasible implementation, the amplitude consistency error and trend consistency error are fused to improve the traditional data enhancement technology instead of using the amplitude consistency error. The amplitude consistency error is the error loss between the input data and the reconstructed data. , the trend consistency error includes the autocorrelation error and cross-correlation error of the data, and the mathematical expression of the amplitude consistency error is shown in the following formula (5):

[0071] (5);

[0072] in, is the total number of samples, represents the i-th sample in the original input data, Represents the i-th sample obtained after reconstruction or processing.

[0073] The mathematical expression of the cross-correlation error is as follows (6):

[0074] (6);

[0075] in, is the mean of the original data, is the mean of the reconstructed data.

[0076] The mathematical expression of the autocorrelation error is as follows:

[0077] (7);

[0078] The generator and discriminator of GAN are used to replace the output layer Decoder in the autoencoder. The data is processed by the encoder and input into the generator. The discriminator guides the generator to generate data. The goal of the generator is to minimize the judgment error of the discriminator on the generated data. The loss function of GAN is as follows (8):

[0079] (8);

[0080] in, is the output of the discriminator for the real data, indicating the probability that the data is the real data. is the output of the discriminator for the generated data G(z), indicating the probability that the generated data is real data. represents the expectation of the probability distribution p of the real data, represents the expectation of the probability distribution p of the generated data.

[0081] The data enhancement loss of the data enhancement model combining the autoencoder and the GAN network is calculated as follows (9):

[0082] (9);

[0083] S3. Eliminate redundant data from the second enhanced data based on a minimum redundancy maximum correlation analysis method to obtain optimized data.

[0084] In a feasible implementation, the enhanced data is subjected to minimum redundancy maximum correlation analysis to eliminate process variables irrelevant to the fault variable and retain the features with the maximum correlation to the fault variable, which maximizes the optimization objective function. As shown in formula (10):

[0085] (10);

[0086] in, Representing features and target variables The mutual information between them is maximized to select the features most relevant to the target; Represents the mutual information between features and , minimizes this term, removes redundant features, Represents the total number of samples.

[0087] S4. Use the optimized data to train the fault classification model to obtain an optimized fault classification model.

[0088] Among them, the model structure of the fault classification model is constructed based on the one-dimensional convolutional neural network structure and the self-attention model structure;

[0089] The fault classification model consists of 1 convolutional layer, 2 residual modules, 2 self-attention layers and 1 fully connected layer.

[0090] In a feasible implementation, the convolution layer of a one-dimensional convolutional neural network uses a convolution kernel to slide on the input data to extract local features. The convolution operation can capture local patterns in space or time. The pooling layer reduces the size of the feature map through maximum pooling or average pooling, reduces computational complexity, and retains important features. The fully connected layer uses the extracted features for classification and introduces an activation function to achieve nonlinear propagation. The mathematical expression of the input sequence of the one-dimensional convolutional neural network is as follows (11):

[0091] (11);

[0092] The mathematical expression of the convolution kernel is as follows (12):

[0093] (12);

[0094] After the convolution operation, the output sequence is No. The mathematical expression of each element is as follows (13):

[0095] (13);

[0096] For a pooling window size of , the mathematical expression of the pooling operation is as follows (14):

[0097] (14);

[0098] The output of the pooling layer Flatten to vector , and passes through the fully connected layer. The mathematical expression of this process is as follows (15):

[0099] (15);

[0100] in, is the weight matrix, is the bias vector, is the activation function.

[0101] The self-attention mechanism generates attention weights by calculating the correlation between each element in the input sequence and other elements, and uses these weights to weighted sum the input to get the output. Each input element generates a query vector Q, a key vector K, and a value vector through linear transformation. , use the dot product to calculate the similarity between the query and the key, and normalize it through the softmax function. The self-attention weight calculation process is as follows (16):

[0102] (16);

[0103] Where T is the matrix transpose operation, are the query vector Q, key vector K and value vector Dimension.

[0104] The self-attention model is combined with the convolutional network, and after multiple convolutions and pooling, it is input into the self-attention model to extract the correlation of variables. The self-attention model is combined with the convolutional network, and after multiple convolutions and pooling, it is input into the self-attention model to extract the correlation of variables, and finally input into the fully connected layer to achieve the classification effect.

[0105] Optionally, the fault classification model is trained using the optimized data to obtain an optimized fault classification model, including:

[0106] Inputting the optimized data into the fault classification model to perform fault classification and obtain the fault classification result;

[0107] Based on the cross-validation method, the loss function is calculated according to the fault classification results and the optimized data to obtain the classification loss;

[0108] Based on the gradient descent method, the fault classification model is reversely optimized according to the classification loss to obtain the optimized fault classification model.

[0109] In a feasible implementation manner, the k-fold cross-validation of the present invention divides the optimization data into k subsets of equal size, selects one of the subsets as a validation set, and the remaining k-1 subsets as training sets, trains the model on the training set, and evaluates the model performance on the validation set, and repeats this step until all k subsets are used as validation sets.

[0110] The accuracy of k-fold cross validation can be expressed as follows (17):

[0111] (17);

[0112] in, represents the accuracy of the i-th validation set, and k is the number of subsets.

[0113] The Adam algorithm calculates the first-order and second-order moment estimates of the gradient to dynamically adjust the learning rate of each parameter. In the process of calculating the gradient The mathematical expression of is as follows (18):

[0114] (18);

[0115] in, is the parameter for time step t Derivative operation.

[0116] The mathematical expression for updating the first-order moment estimate is as follows (19):

[0117] (19);

[0118] in, The mathematical expression for updating the second-order moment estimate by the first-order moment estimate at time step t is as follows:

[0119] (20);

[0120] in, represents the second-order moment estimate at time step t, and is the decay rate of the first-order and second-order moments, which is a custom hyperparameter and usually takes a value between (0, 1).

[0121] The deviation correction of the first-order and second-order moments is calculated as follows (21):

[0122] (twenty one);

[0123] in, and After bias correction and .

[0124] The mathematical expression of the parameter update process is as follows (22):

[0125] (twenty two);

[0126] in, is a custom learning rate, is a small constant to prevent the denominator from being zero.

[0127] Adam automatically adjusts the learning rate during training and is robust to sparse gradients and noise.

[0128] S5. In the actual production process, data is collected from the industrial control server to obtain the actual production data of the industrial process; based on the actual production data, fault classification prediction is performed by optimizing the data enhancement model and optimizing the fault classification model to obtain the actual fault classification result.

[0129] In a feasible implementation, actual production data is collected and the optimized model is used to classify faults, and a confusion matrix and a receiver operating characteristic (ROC) curve are used to evaluate the fault classification effect.

[0130] The confusion matrix is ​​a table used to describe the performance of a classification model, showing the comparison between the model's predictions and the actual results; including the number of samples correctly predicted by the model as positive, the number of samples correctly predicted by the model as negative, the number of samples incorrectly predicted by the model as positive, and the number of samples incorrectly predicted by the model as negative.

[0131] A confusion matrix is ​​generated based on simulation data; the simulation data includes normal data and 7 types of faults (fault 1-fault 7); the generated confusion matrix is ​​an 8x8 table, the horizontal and vertical axes of the table are all normal, fault 1-fault 7, the first one in the upper left corner of the diagonal is the number of samples predicted as normal, and the other 7 are the number of samples predicted as faults. Taking the diagonal as the benchmark, the lower left of the diagonal is the number of samples that the model mistakenly predicts as normal as faults, and the upper right of the diagonal is the number of samples that the model mistakenly predicts as faults. By analyzing the diagonal and non-diagonal elements, the classification performance of the model can be comprehensively and clearly evaluated, the advantages and disadvantages of the model can be identified, and corresponding improvements can be made.

[0132] Generate confusion matrix diagram based on simulation data; through full-category cross-validation (accurately displaying the number of correct classifications for each category and the distribution of specific types of misclassifications), error type directional analysis (distinguishing between "over-sensitivity" errors that misclassify normal samples as faults and "failure tolerance" errors that miss faulty samples), and multi-level indicator derivation capabilities (capable of calculating category-specific recall rate / precision rate / F1 score), the present invention breaks through the limitations of a single global indicator in fine-grained diagnosis, error attribution and tracing, and is particularly suitable for industrial detection scenarios that require accurate positioning models to identify blind spots for specific fault types.

[0133] The ROC curve is a graphical tool for evaluating the performance of a binary classification model, showing the performance of the model at different thresholds. The ROC curve evaluates the classification ability of the model by plotting the relationship between the true positive rate and the false positive rate.

[0134] Generate ROC curve based on simulation data; Since the fault classification of industrial control process is a multi-classification task, the fault classification task is divided into 8 binary classification tasks, and the ROC curve of each class is calculated, and a total of 8 curves are obtained; the curve is an arc with 8 arcs bending to the upper left corner. The horizontal axis of the curve is the false positive rate, which indicates the proportion of negative samples misclassified as positive samples; the vertical axis is the true positive rate, which indicates the proportion of positive samples correctly classified as positive samples, and the value range is (0, 1). The ROC curve should be as close to the upper left corner as possible, indicating that the model maintains a low false positive rate while having a high recall rate, that is, the model is very effective in identifying positive samples and also performs well in avoiding misclassification of negative samples. This performance combination makes the model more reliable and effective in practical applications and can meet the needs. The area under the curve (AUC) is the area under the ROC curve, and the value range is between 0 and 1. The closer the AUC is to 1, the better the classification effect.

[0135] Based on the ROC curve; through threshold-independent comprehensive evaluation, performance measurement resistant to sample distribution influence, visualization of multi-dimensional trade-offs, AUC quantitative comparison and cost-sensitive decision support and other characteristics, the present invention overcomes the limitations of traditional indicators in threshold dependence, category imbalance sensitivity, single-dimensional evaluation, etc., and provides a more comprehensive, robust and operational analysis framework for fault classification tasks.

[0136] The present invention proposes an industrial process fault classification method based on the fusion of trend and amplitude consistency, combining autoencoders and adversarial generative networks to enhance the data of industrial processes, solving the problem of few data samples and much interference in the manufacturing process; considering the consistency of autocorrelation and cross-correlation, a new pre-training loss function is designed to characterize the features involved in the trend fault signal, making the extracted features more accurate; when only a small number of fault samples are contaminated, due to the presence of strong noise, the present invention can provide a more advanced learning mechanism for the deep neural network to accurately extract the potential weak features of minor faults. The present invention is an efficient and accurate industrial process fault classification method for minor faults in nonlinear industrial processes.

[0137] Figure 2 The present invention is a block diagram of an industrial process fault classification device based on trend and amplitude consistency fusion according to an exemplary embodiment. The device is used in an industrial process fault classification method based on trend and amplitude consistency fusion. Figure 2 The device includes a data processing module 210, a data enhancement module 220, a data optimization module 230, a model optimization module 240 and a fault classification module 250. Among them:

[0138] The data processing module 210 is used to obtain the historical production data of the industrial process; based on the sliding window method and the Z-score normalization algorithm, the historical production data is preprocessed to obtain training data;

[0139] A data enhancement module 220 is used to perform data enhancement on the data enhancement model based on the fusion trend consistency error and the amplitude consistency error according to the training data to obtain an optimized data enhancement model and second enhanced data;

[0140] A data optimization module 230, configured to eliminate redundant data from the second enhanced data based on a minimum redundancy maximum correlation analysis method to obtain optimized data;

[0141] A model optimization module 240 is used to train the fault classification model using the optimization data to obtain an optimized fault classification model;

[0142] The fault classification module 250 is used to collect data from the industrial control server during the actual production process to obtain the actual production data of the industrial process; based on the actual production data, the fault classification prediction is performed by optimizing the data enhancement model and the fault classification model to obtain the actual fault classification result.

[0143] Optionally, the data processing module 210 is further configured to:

[0144] Based on the random partitioning method, the historical production data is randomly partitioned to obtain the partitioned historical data;

[0145] Based on the preset sliding window, the divided historical data is cleaned to obtain the cleaned historical data;

[0146] Based on the cleaned historical data, the Z-score standardization algorithm is used to standardize the data to obtain training data.

[0147] Among them, the model structure of the data enhancement model is constructed based on the autoencoder structure and the generative adversarial network structure;

[0148] The data augmentation model consists of an encoder part, a generator part, and a discriminator part;

[0149] The encoder part is used to map the input model data into a low-dimensional latent space;

[0150] The generator part is used to generate enhanced data based on the data in the latent space;

[0151] The discriminator part is used to guide the generator to generate augmented data.

[0152] Optionally, the data enhancement module 220 is further configured to:

[0153] Use the training data to train the data enhancement model to obtain an optimized data enhancement model;

[0154] The training data is input into the optimized data enhancement model for data enhancement to obtain the second enhanced data.

[0155] Optionally, the data enhancement module 220 is further configured to:

[0156] Input the training data into the encoder part for latent space mapping to obtain mapping data;

[0157] Based on the discriminator part, according to the mapping data, the generator part generates data to obtain first enhanced data;

[0158] Calculate the amplitude consistency error based on the training data and the first enhanced data;

[0159] Calculate the trend consistency error according to the training data and the first enhanced data; the trend consistency error includes an autocorrelation error and a cross-correlation error;

[0160] The loss function is calculated based on the amplitude consistency error and the trend consistency error to obtain the data enhancement loss;

[0161] According to the data enhancement loss, the data enhancement model is reversely optimized to obtain the optimized data enhancement model.

[0162] Among them, the model structure of the fault classification model is constructed based on the one-dimensional convolutional neural network structure and the self-attention model structure;

[0163] The fault classification model consists of 1 convolutional layer, 2 residual modules, 2 self-attention layers and 1 fully connected layer.

[0164] Optionally, the model optimization module 240 is further configured to:

[0165] Inputting the optimized data into the fault classification model to perform fault classification and obtain the fault classification result;

[0166] Based on the cross-validation method, the loss function is calculated according to the fault classification results and the optimized data to obtain the classification loss;

[0167] Based on the gradient descent method, the fault classification model is reversely optimized according to the classification loss to obtain the optimized fault classification model.

[0168] The present invention proposes an industrial process fault classification method based on the fusion of trend and amplitude consistency, combining autoencoders and adversarial generative networks to enhance the data of industrial processes, solving the problem of few data samples and much interference in the manufacturing process; considering the consistency of autocorrelation and cross-correlation, a new pre-training loss function is designed to characterize the features involved in the trend fault signal, making the extracted features more accurate; when only a small number of fault samples are contaminated, due to the presence of strong noise, the present invention can provide a more advanced learning mechanism for the deep neural network to accurately extract the potential weak features of minor faults. The present invention is an efficient and accurate industrial process fault classification method for minor faults in nonlinear industrial processes.

[0169] Figure 3 is a schematic diagram of the structure of an industrial process fault classification device provided by an embodiment of the present invention, such as Figure 3 As shown, the industrial process fault classification device may include the above Figure 2 The industrial process fault classification device based on trend and amplitude consistency fusion is shown. Optionally, the industrial process fault classification device 310 may include a first processor 2001 .

[0170] Optionally, the industrial process fault classification device 310 may further include a memory 2002 and a transceiver 2003 .

[0171] The first processor 2001, the memory 2002 and the transceiver 2003 may be connected via a communication bus.

[0172] Combine the following Figure 3 The components of the industrial process fault classification device 310 are specifically introduced:

[0173] The first processor 2001 is the control center of the industrial process fault classification device 310, and may be a processor or a general term for multiple processing elements. For example, the first processor 2001 is one or more central processing units (CPUs), or may be application specific integrated circuits (ASICs), or may be one or more integrated circuits configured to implement the embodiments of the present invention, such as one or more microprocessors (digital signal processors, DSPs), or one or more field programmable gate arrays (field programmable gate arrays, FPGAs).

[0174] Optionally, the first processor 2001 may perform various functions of the industrial process fault classification device 310 by running or executing a software program stored in the memory 2002 and calling data stored in the memory 2002 .

[0175] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Figure 3 CPU0 and CPU1 are shown in FIG.

[0176] In a specific implementation, as an embodiment, the industrial process fault classification device 310 may also include multiple processors, such as Figure 3 The first processor 2001 and the second processor 2004 are shown in FIG. Each of these processors may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor here may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0177] The memory 2002 is used to store the software program for executing the solution of the present invention, and is controlled to be executed by the first processor 2001. The specific implementation method can refer to the above method embodiment, which will not be repeated here.

[0178] Optionally, the memory 2002 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 2002 may be integrated with the first processor 2001, or may exist independently and access the first processor 2001 through the interface circuit ( Figure 3 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0179] The transceiver 2003 is used to communicate with a network device or a terminal device.

[0180] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Figure 3 The receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0181] Optionally, the transceiver 2003 may be integrated with the first processor 2001, or may exist independently and communicate with the first processor 2001 through the interface circuit ( Figure 3 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0182] It should be noted that Figure 3 The structure of the industrial process fault classification device 310 shown in the figure does not constitute a limitation on the router, and the actual knowledge structure identification device may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0183] In addition, the technical effects of the industrial process fault classification device 310 can refer to the technical effects of the industrial process fault classification method based on trend and amplitude consistency fusion described in the above method embodiment, which will not be repeated here.

[0184] It should be understood that the first processor 2001 in the embodiment of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0185] It should also be understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0186] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0187] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0188] In the present invention, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0189] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0190] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0191] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0192] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0193] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0194] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0195] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0196] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An industrial process fault classification method based on trend and amplitude consistency fusion, characterized in that: The method comprises: Acquire historical production data of the industrial process; preprocess the historical production data based on a sliding window method and a Z-score normalization algorithm to obtain training data; According to the training data, data enhancement is performed on the data enhancement model based on the fusion trend consistency error and the amplitude consistency error to obtain an optimized data enhancement model and second enhanced data; Eliminate redundant data from the second enhanced data based on a minimum redundancy maximum correlation analysis method to obtain optimized data; Using the optimized data, training a fault classification model to obtain an optimized fault classification model; In the actual production process, data is collected from the industrial control server to obtain actual production data of the industrial process; based on the actual production data, fault classification prediction is performed through the optimized data enhancement model and the optimized fault classification model to obtain actual fault classification results.

2. The industrial process fault classification method based on trend and amplitude consistency fusion according to claim 1 is characterized in that: The historical production data is preprocessed based on the sliding window method and the Z-score normalization algorithm to obtain training data, including: Based on a random partitioning method, the historical production data is randomly partitioned to obtain the partitioned historical data; Based on a preset sliding window, data cleaning is performed on the divided historical data to obtain cleaned historical data; According to the cleaned historical data, the Z-score standardization algorithm is used to perform data standardization to obtain training data.

3. The industrial process fault classification method based on trend and amplitude consistency fusion according to claim 1 is characterized in that: The model structure of the data enhancement model is constructed based on the autoencoder structure and the generative adversarial network structure; The data augmentation model includes an encoder part, a generator part and a discriminator part; The encoder part is used to map the input model data to a low-dimensional latent space; The generator part is used to generate enhanced data according to the data in the latent space; The discriminator part is used to guide the generator to generate enhanced data.

4. The industrial process fault classification method based on trend and amplitude consistency fusion according to claim 1 is characterized in that: According to the training data, data enhancement is performed on the data enhancement model based on the fusion trend consistency error and the amplitude consistency error to obtain an optimized data enhancement model and second enhanced data, including: Using the training data, training the data enhancement model to obtain an optimized data enhancement model; The training data is input into the optimized data enhancement model for data enhancement to obtain second enhanced data.

5. The industrial process fault classification method based on trend and amplitude consistency fusion according to claim 4 is characterized in that: The step of using the training data to train the data enhancement model to obtain an optimized data enhancement model includes: Inputting the training data into the encoder part for latent space mapping to obtain mapping data; Based on the discriminator part, according to the mapping data, the generator part generates data to obtain first enhanced data; Calculating an amplitude consistency error according to the training data and the first enhanced data; Calculate a trend consistency error according to the training data and the first enhanced data; the trend consistency error includes an autocorrelation error and a cross-correlation error; A loss function is calculated according to the amplitude consistency error and the trend consistency error to obtain a data enhancement loss; According to the data enhancement loss, reverse parameter optimization is performed on the data enhancement model to obtain an optimized data enhancement model.

6. The industrial process fault classification method based on trend and amplitude consistency fusion according to claim 1 is characterized in that: The model structure of the fault classification model is constructed based on a one-dimensional convolutional neural network structure and a self-attention model structure; The fault classification model includes 1 convolutional layer, 2 residual modules, 2 self-attention layers and 1 fully connected layer.

7. The industrial process fault classification method based on trend and amplitude consistency fusion according to claim 1 is characterized in that: The method of using the optimized data to train the fault classification model to obtain the optimized fault classification model includes: Inputting the optimized data into a fault classification model to perform fault classification and obtain a fault classification result; Based on the cross-validation method, a loss function is calculated according to the fault classification result and the optimization data to obtain the classification loss; Based on the gradient descent method, reverse parameter optimization is performed on the fault classification model according to the classification loss to obtain an optimized fault classification model.

8. An industrial process fault classification device based on trend and amplitude consistency fusion, the industrial process fault classification device based on trend and amplitude consistency fusion is used to implement the industrial process fault classification method based on trend and amplitude consistency fusion as claimed in any one of claims 1 to 7, characterized in that: The device comprises: A data processing module is used to obtain historical production data of industrial processes; based on a sliding window method and a Z-score normalization algorithm, the historical production data is preprocessed to obtain training data; A data enhancement module, used to perform data enhancement on a data enhancement model based on a fusion trend consistency error and an amplitude consistency error according to the training data, to obtain an optimized data enhancement model and second enhanced data; A data optimization module, configured to eliminate redundant data from the second enhanced data based on a minimum redundancy maximum correlation analysis method to obtain optimized data; A model optimization module, used to train a fault classification model using the optimization data to obtain an optimized fault classification model; The fault classification module is used to collect data from the industrial control server during the actual production process to obtain the actual production data of the industrial process; based on the actual production data, fault classification prediction is performed through the optimized data enhancement model and the optimized fault classification model to obtain the actual fault classification result.

9. An industrial process fault classification device, characterized in that: The industrial process fault classification device comprises: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program codes, which can be called by a processor to execute the method according to any one of claims 1 to 7.

Images