Fault detection and isolation method and device for process industry

Abstract

The present invention provides a method and device for fault detection and isolation of process industries, wherein the method comprises the following steps: obtaining variable data of the process industry to be tested; preprocessing the variable data to obtain training data, wherein the training data has no abnormal values ​​and no missing values; constructing a window sample set of the training data; obtaining a window sample image and a window sample label according to the window sample set; feature processing the window sample image; and performing fault detection and isolation on the process industry to be tested according to the feature-processed window sample image and the window sample label. The present invention can output the fault detection and isolation results of the process industry to be tested in the form of an image frame, thereby realizing the coordination of fault detection and isolation. In addition, it can also retain complete data information, thereby realizing feature extraction within the window range, and avoiding the problems of loss of process information and waste of storage space under the existing data visualization mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field of process industry, and in particular to a fault detection and isolation method and a fault detection and isolation device for process industry. Background Art

[0002] Fault detection and isolation technology in process industry is an analytical method that uses the easily measurable data of the production process to mine hidden information in the data, establish process models and evaluation indicators, and realize online qualitative and quantitative rapid detection of the detected process. It has advantages that conventional detection methods cannot match, such as high real-time performance, low detection cost, strong portability, and low requirements for production process background and process mechanism. After years of technological development and improvement, this technology has been widely used in national pillar industries such as petroleum, chemical industry, steel, building materials, medicine, and food. With the rapid development of my country's market economy and the continuous improvement of the level of informatization and intelligence of production processes, the traditional data-driven process model based on statistical analysis methods can no longer meet the real-time needs of enterprises for updating production process information in the new era due to its two-step (detection-isolation) monitoring mechanism. With the help of machine vision and neural network technology, the process model that replaces the traditional data flow form with visual data can achieve the coordination of fault detection and isolation. However, the easily measurable data of complex production processes generally have high dimensions and certain correlations, and the existing process modeling methods based on visual data flow have the following disadvantages:

[0003] (1) The visualization mechanism of easy-to-measure process data has limitations. The existing easy-to-measure process variable visualization mechanism is basically based on the direct use of single-variable waveform graphs. For multi-dimensional variables, especially multi-dimensional process variable waveform graphs under standardized data, it is impossible to establish a corresponding detection model. Specifically, for overlapping intervals, non-top-level waveforms will be covered by top-level waveform images, resulting in loss of process information. In addition, the retention of areas without waveforms also wastes image storage space, indirectly increasing the detection cost.

[0004] (2) The coordination of fault detection and isolation is highly dependent on the form of training data. Although the modeling method using image data improves the traditional data-driven model's analysis method of first detecting and then isolating, and can achieve coordinated analysis of the two, the training samples need to be manually labeled with the fault range and source, which requires high requirements on the operator's relevant process background knowledge and the accuracy of the labeled range. At the same time, the labeling work of a large number of samples requires additional labor costs. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a fault detection and isolation method for process industry. By performing image processing on variable data, taking window sample images as input for feature extraction, and combining with corresponding fault information tags, the coordination of fault detection and isolation is achieved, thereby improving the efficiency of image sample fault labeling, reducing the cost of manual labeling, and improving the real-time and effectiveness of the monitoring process. In addition, complete data information can be retained, thereby achieving feature extraction within the window range, avoiding the loss of process information and waste of storage space under the existing data visualization mechanism.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A method for fault detection and isolation of a process industry comprises the following steps: obtaining variable data of the process industry to be tested; preprocessing the variable data to obtain training data, wherein the training data has no abnormal values ​​and no missing values; constructing a window sample set of the training data; obtaining a window sample image and a window sample label according to the window sample set; feature processing the window sample image; and performing fault detection and isolation on the process industry to be tested according to the feature-processed window sample image and the window sample label.

[0008] According to an embodiment of the present invention, the preprocessing of the variable data to obtain the training data specifically includes the following steps: preprocessing the variable data to obtain the training data using the piecewise 3σ principle.

[0009] According to an embodiment of the present invention, the step of constructing the window sample set of the training data specifically includes the following steps: obtaining data corresponding to the fault-free process in the training data; constructing a principal component analysis model for the fault-free process data; and constructing a T 2 Monitoring indicator control limit; according to the T 2 The monitoring indicator control limits construct a window sample set of the training data under a rolling window mechanism.

[0010] According to an embodiment of the present invention, the step of obtaining a window sample image and a window sample label according to the window sample set specifically comprises the following steps: 2The monitoring indicator control limit determines whether there is a window sample corresponding to the fault process data in the window sample set; if so, the fault contribution of the window sample corresponding to the fault process data is analyzed, and a fault isolation model is constructed according to the fault contribution; the fault contribution rate of each window sample is obtained according to the fault isolation model; it is determined whether the fault contribution rate of each window sample exceeds the fault contribution rate threshold; if so, the fault interval range of the corresponding window sample is recorded; the window sample label of the window sample set is constructed according to the fault interval range; an image conversion model is constructed; according to the image conversion model, the data points in each window sample are converted into corresponding pixel parameter values ​​to obtain the window sample image of the window sample set.

[0011] According to one embodiment of the present invention, the fault detection and isolation of the process industry to be tested is performed based on the window sample image after feature processing and the window sample label, which specifically includes the following steps: using a fully connected neural network to establish a mapping relationship between the window sample image after feature processing and the window sample label; and obtaining the fault detection and isolation result of the process industry to be tested based on the mapping relationship.

[0012] A fault detection and isolation device for a process industry, comprising: an acquisition module, the acquisition module is used to acquire variable data of the process industry to be tested; a preprocessing module, the preprocessing module is used to preprocess the variable data to obtain training data, wherein the training data has no abnormal values ​​and no missing values; a data construction module, the data construction module is used to construct a window sample set of the training data; a data conversion marking module, the data conversion marking module is used to obtain a window sample image and a window sample label according to the window sample set; a feature extraction module, the feature extraction module is used to feature process the window sample image; and a fault detection and isolation module, the fault detection and isolation module is used to perform fault detection and isolation on the process industry to be tested according to the window sample image and the window sample label after feature processing.

[0013] According to an embodiment of the present invention, the preprocessing module is specifically used to preprocess the variable data by adopting the piecewise 3σ principle to obtain the training data.

[0014] According to an embodiment of the present invention, the data construction module is specifically used to obtain data corresponding to the fault-free process in the training data, construct a principal component analysis model for the fault-free process data, and construct T according to the principal component analysis model. 2 Monitoring indicator control limits, according to the T 2 The monitoring indicator control limits construct a window sample set of the training data under a rolling window mechanism.

[0015] According to an embodiment of the present invention, the data conversion marking module is specifically used to: 2 The monitoring indicator control limit determines whether there is a window sample corresponding to the fault process data in the window sample set; if so, the fault contribution of the window sample corresponding to the fault process data is analyzed, and a fault isolation model is constructed according to the fault contribution; the fault contribution rate of each window sample is obtained according to the fault isolation model; it is determined whether the fault contribution rate of each window sample exceeds the fault contribution rate threshold; if so, the fault interval range of the corresponding window sample is recorded; the window sample label of the window sample set is constructed according to the fault interval range; an image conversion model is constructed; according to the image conversion model, the data points in each window sample are converted into corresponding pixel parameter values ​​to obtain the window sample image of the window sample set.

[0016] According to one embodiment of the present invention, the fault detection and isolation module is specifically used to use a fully connected neural network to establish a mapping relationship between the window sample image after feature processing and the window sample label, and obtain the fault detection and isolation result of the process industry to be tested according to the mapping relationship.

[0017] The beneficial effects of the present invention are as follows:

[0018] 1) The present invention converts the variable data of the process industry to be tested into image data, and can output the fault detection and isolation results of the process industry to be tested in the form of an image frame, thereby realizing the coordination of fault detection and isolation;

[0019] 2) The present invention can retain complete data information by converting variable data of the process industry to be tested into image pixel values, thereby realizing feature extraction within the window range and avoiding the problems of loss of process information and waste of storage space under the existing data visualization mechanism;

[0020] 3) The present invention constructs window sample labels through principal component analysis, thereby ensuring labeling accuracy and reducing the labor cost of labeling work. In addition, by using principal component analysis, when new working conditions appear in the production process or the working point drifts slowly, it is only necessary to use the covariance equivalence principle to add new data on the basis of the original fixed parameter dimension to achieve the update effect, which is convenient for subsequent maintenance and updating of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A flow chart of a method for fault detection and isolation in a process industry according to an embodiment of the present invention;

[0022] Figure 2 A schematic diagram of a fault detection and isolation process in a process industry according to a specific embodiment of the present invention;

[0023] Figure 3 It is a block diagram of a fault detection and isolation device for process industry according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] Figure 1 The figure is a flow chart of a fault detection and isolation method for process industry according to an embodiment of the present invention.

[0026] like Figure 1 As shown, the fault detection and isolation method for process industry in an embodiment of the present invention includes the following steps:

[0027] S1, obtain variable data of the process industry to be tested.

[0028] Specifically, the variable data, namely, the process data of the process industry to be tested, can be obtained through the distributed control system of the process industry to be tested. It should be noted that the variable data, namely, the process data of the process industry to be tested, can be online data.

[0029] S2, preprocessing variable data to obtain training data, wherein the training data has no outliers and no missing values.

[0030] Specifically, the segmented 3σ principle can be used to preprocess the variable data to obtain the training data. Of course, the preprocessing method is not unique. For example, in other embodiments of the present invention, the variable data can also be preprocessed by manual screening to obtain the training data.

[0031] S3, construct a window sample set of training data.

[0032] Specifically, the data corresponding to the fault-free process in the training data can be obtained first, and then the principal component analysis model of the fault-free process data can be constructed, and T can be constructed according to the principal component analysis model. 2 Monitoring indicator control limits can be used according to T 2 The monitoring indicator control limits construct a window sample set of training data under the rolling window mechanism.

[0033] More specifically, the expert system can select the data set X corresponding to the fault-free process from the training data NF (Expert system refers to integrating expert experience into information system, and computers replace manual work to complete the selection of fault-free data sets), and then the data set X can beNF Constructing the principal component analysis model:

[0034] X NF =TP T +E

[0035] Among them, T, P, and E represent the data set X NF The score matrix, loading matrix, and residual matrix in principal component analysis.

[0036] Furthermore, the principal component analysis module can be used to construct T 2 Monitoring indicator control limits:

[0037]

[0038] in, q represents the rank of the score matrix, and α represents the confidence.

[0039] Furthermore, in obtaining T 2 After monitoring the control limits of the indicators, we can construct a window sample set of training data under the rolling window mechanism:

[0040] X MW = {X1, X2, X3, ..., X N}

[0041] Where N represents the number of data in the window sample set.

[0042] S4, obtaining a window sample image and a window sample label according to the window sample set.

[0043] Specifically, according to the principal component analysis model and T 2 The monitoring indicator control limit determines whether there is a window sample corresponding to the fault process data in the window sample set. If so, the fault contribution of the window sample corresponding to the fault process data is analyzed, and a fault isolation model is constructed based on the fault contribution. Then, the fault contribution rate of each window sample can be obtained based on the fault isolation model, and it is determined whether the fault contribution rate of each window sample exceeds the fault contribution rate threshold. If so, the fault interval range of the corresponding window sample is recorded, and then the window sample label of the window sample set can be constructed according to the fault interval range, and an image conversion model is constructed. Then, the data points in each window sample can be converted into corresponding pixel parameter values ​​according to the image conversion model to obtain the window sample set and the window sample image.

[0044] More specifically, the process of obtaining a window sample label according to a window sample set may include the following steps:

[0045] First, the principal component analysis model and T 2 The monitoring index control limit performs fault analysis and judgment on each window sample in the window sample set:

[0046]

[0047] Where n represents the rolling window length, l = 1, 2, ..., N.

[0048] Furthermore, for the existence of more than T 2 Window sample data x in the window sample set of the monitoring index control limit j (j=1,2,3,…), a fault isolation model can be constructed based on contribution analysis:

[0049]

[0050] Among them, m represents the window sample dimension, Represents an m-dimensional row vector whose i-th element is 1 and the rest of the elements are 0.

[0051] Furthermore, the fault contribution rate of each window sample is obtained according to the fault isolation model, and the fault interval range is recorded for the corresponding window sample whose fault contribution rate exceeds the fault contribution threshold λ, and the window sample label of the window sample set is constructed:

[0052] y k =[(y k,1 ,y k,a1 ,y k,b1 )(y k,2 ,y k,a2 ,y k,b2 )…(y k,m ,y k,am ,y k,bm )](k=1…N)

[0053] Among them, y k represents the k-th window sample label, y k,1 ,y k,2 …y k,m is the reference variable of window sample 1, window sample 2…window sample m, (y k,a1 ,y k,b1 )、(y k,a2 ,y k,b2 )…(y k,am ,y k,bm ) is the corresponding variable fault interval. For the fault-free window, the window sample label is (0,0,0), thereby obtaining N window sample labels.

[0054] More specifically, the process of obtaining a window sample image according to a window sample set may include the following steps:

[0055] First, for each data point x in the window sample set, u,v(u=1…n,v=1…m), and build an image conversion model by matching the numerical interval, that is, the three primary colors (RGB) image conversion model:

[0056]

[0057] Among them, (a * b * ) is a certain interval in the range of variable v, For x u,v The red, green, and blue parameter values ​​of the corresponding pixel (u, v) when converted into an image.

[0058] Furthermore, according to an image conversion model, ie, a three-primary-color (RGB) image conversion model, a row of white pixels is inserted at both ends of the v-th row of window images to obtain N n×(1+2)m×3-dimensional window sample images.

[0059] S5, feature processing window sample image.

[0060] Specifically, a convolutional neural network can be used to extract image features from the N n×(1+2)m×3-dimensional window sample images obtained. For example, C The convolution-pooling operation is performed to obtain the feature vectors of N window sample images.

[0061] S6, performing fault detection and isolation on the process industry to be tested according to the feature-processed window sample images and window sample labels.

[0062] Specifically, a fully connected neural network can be used to establish a mapping relationship between a feature-processed window sample image and a window sample label, and the fault detection and isolation results of the process industry to be tested can be obtained according to the mapping relationship.

[0063] For example, the feature vectors of the N window sample images can be used as the input of the fully connected neural network. F The fully connected operation is performed at each layer to construct a mapping relationship between the feature vectors of N window sample images and the corresponding N window sample labels, so that the fault detection and isolation results of the process industry to be tested can be output in the form of an image frame.

[0064] The following will be combined Figure 2 , further illustrating the practical application process of the fault detection and isolation method for process industry in an embodiment of the present invention.

[0065] like Figure 2 As shown, variable data of the process industry to be tested can be obtained, and then a window sample set can be obtained based on the variable data. Principal component analysis, control limit, and fault contribution analysis can be performed on the window sample set to obtain a window sample label. In addition, the window sample set can be converted into an image to obtain a window sample image.

[0066] Furthermore, if Figure 2 As shown, the feature vector of the window sample image can be obtained by inputting the window sample image into the convolutional neural network for feature extraction, and the mapping relationship between the feature vector of the window sample image and the window sample label can be established by using a fully connected neural network, thereby realizing the coordinated fault detection and isolation of the process industry to be tested.

[0067] The beneficial effects of the present invention are as follows:

[0068] 1) The present invention converts the variable data of the process industry to be tested into image data, and can output the fault detection and isolation results of the process industry to be tested in the form of an image frame, thereby realizing the coordination of fault detection and isolation;

[0069] 2) The present invention can retain complete data information by converting variable data of the process industry to be tested into image pixel values, thereby realizing feature extraction within the window range and avoiding the problems of loss of process information and waste of storage space under the existing data visualization mechanism;

[0070] 3) The present invention constructs window sample labels through principal component analysis, thereby ensuring labeling accuracy and reducing the labor cost of labeling work. In addition, by using principal component analysis, when new working conditions appear in the production process or the working point drifts slowly, it is only necessary to use the covariance equivalence principle to add new data on the basis of the original fixed parameter dimension to achieve the update effect, which is convenient for subsequent maintenance and updating of the model.

[0071] Corresponding to the fault detection and isolation method for process industry proposed in the above embodiment, the present invention also proposes a fault detection and isolation device for process industry.

[0072] like Figure 3 As shown, the fault detection and isolation device for process industry of the embodiment of the present invention comprises an acquisition module 10, a preprocessing module 20, a data construction module 30, a data conversion marking module 40, a feature extraction module 50 and a fault detection and isolation module 60. Among them, the acquisition module 10 is used to acquire the variable data of the process industry to be tested; the preprocessing module 20 is used to preprocess the variable data to obtain training data, wherein the training data has no abnormal values ​​and no missing values; the data construction module 30 is used to construct a window sample set of training data; the data conversion marking module 40 is used to obtain a window sample image and a window sample label according to the window sample set; the feature extraction module 50 is used to feature process the window sample image; and the fault detection and isolation module 60 is used to perform fault detection and isolation on the process industry to be tested according to the window sample image and the window sample label after feature processing.

[0073] In one embodiment of the present invention, the acquisition module 10 can acquire variable data, namely, process data of the process industry to be tested, through the distributed control system of the process industry to be tested. It should be noted that the variable data, namely, process data of the process industry to be tested, can be online data.

[0074] In one embodiment of the present invention, the preprocessing module 20 may use the segmented 3σ principle to preprocess the variable data to obtain the training data. Of course, the preprocessing method is not unique. For example, in other embodiments of the present invention, the variable data may be preprocessed by manual screening to obtain the training data.

[0075] In one embodiment of the present invention, the data construction module 30 may first obtain data corresponding to the fault-free process in the training data, then construct a principal component analysis model for the fault-free process data, and construct T according to the principal component analysis model. 2 Monitoring indicator control limits can be used according to T 2 The monitoring indicator control limits construct a window sample set of training data under the rolling window mechanism.

[0076] More specifically, the expert system can select the data set X corresponding to the fault-free process from the training data NF (Expert system refers to integrating expert experience into information system, and computers replace manual work to complete the selection of fault-free data sets), and then the data set X can be NF Constructing the principal component analysis model:

[0077] X NF =TP T +E

[0078] Among them, T, P, and E represent the data set X NF The score matrix, loading matrix, and residual matrix in principal component analysis.

[0079] Furthermore, the principal component analysis module can be used to construct T 2 Monitoring indicator control limits:

[0080]

[0081] in, q represents the rank of the score matrix, and α represents the confidence.

[0082] Furthermore, in obtaining T 2 After monitoring the control limits of the indicators, we can construct a window sample set of training data under the rolling window mechanism:

[0083] X MW = {X1, X2, X3, ..., X N}

[0084] Where N represents the number of data in the window sample set.

[0085] In one embodiment of the present invention, the data conversion marking module 40 can be based on the principal component analysis model and T 2 The monitoring indicator control limit determines whether there is a window sample corresponding to the fault process data in the window sample set. If so, the fault contribution of the window sample corresponding to the fault process data is analyzed, and a fault isolation model is constructed based on the fault contribution. Then, the fault contribution rate of each window sample can be obtained based on the fault isolation model, and it is determined whether the fault contribution rate of each window sample exceeds the fault contribution rate threshold. If so, the fault interval range of the corresponding window sample is recorded, and then the window sample label of the window sample set can be constructed according to the fault interval range, and an image conversion model is constructed. Then, the data points in each window sample can be converted into corresponding pixel parameter values ​​according to the image conversion model to obtain the window sample set and the window sample image.

[0086] More specifically, the process of obtaining a window sample label according to a window sample set may include the following steps:

[0087] First, the principal component analysis model and T 2 The monitoring index control limit performs fault analysis and judgment on each window sample in the window sample set:

[0088]

[0089] Where n represents the rolling window length, l = 1, 2, ..., N.

[0090] Furthermore, for the existence of more than T 2 Window sample data x in the window sample set of the monitoring index control limit j (j=1,2,3,…), a fault isolation model can be constructed based on contribution analysis:

[0091]

[0092] Among them, m represents the window sample dimension, Represents an m-dimensional row vector whose i-th element is 1 and the rest of the elements are 0.

[0093] Furthermore, the fault contribution rate of each window sample is obtained according to the fault isolation model, and the fault interval range is recorded for the corresponding window sample whose fault contribution rate exceeds the fault contribution threshold λ, and the window sample label of the window sample set is constructed:

[0094] y k =[(y k,1 ,y k,a1 ,y k,b1 )(y k,2 ,yk,a2 ,y k,b2 )…(y k,m ,y k,am ,y k,bm )](k=1…N)

[0095] Among them, y k represents the kth window sample label, y k,1 ,y k,2 …y k,m is the reference variable of window sample 1, window sample 2…window sample m, (y k,a1 ,y k,b1 )、(y k,a2 ,y k,b2 )…(y k,am ,y k,bm ) is the corresponding variable fault interval. For the fault-free window, the window sample label is (0,0,0), thereby obtaining N window sample labels.

[0096] More specifically, the process of obtaining a window sample image according to a window sample set may include the following steps:

[0097] First, for each data point x in the window sample set, u,v (u=1…n,v=1…m), and build an image conversion model by matching the numerical interval, that is, the three primary colors (RGB) image conversion model:

[0098]

[0099] Among them, (a * b * ) is a certain interval in the range of variable v, For x u,v The red, green, and blue parameter values ​​of the corresponding pixel (u, v) when converted into an image.

[0100] Furthermore, according to an image conversion model, ie, a three-primary-color (RGB) image conversion model, a row of white pixels is inserted at both ends of the v-th row of window images to obtain N n×(1+2)m×3-dimensional window sample images.

[0101] In one embodiment of the present invention, the feature extraction module 50 may use a convolutional neural network to extract image features from the N n×(1+2)m×3-dimensional window sample images obtained. For example, the feature extraction module 50 may use a convolutional neural network to extract image features from the N n×(1+2)m×3-dimensional window sample images obtained. C The convolution-pooling operation is performed to obtain the feature vectors of N window sample images.

[0102] In one embodiment of the present invention, the fault detection and isolation module 60 may use a fully connected neural network to establish a mapping relationship between the feature-processed window sample image and the window sample label, and may obtain the fault detection and isolation results of the process industry to be tested based on the mapping relationship.

[0103] For example, the feature vectors of the N window sample images can be used as the input of the fully connected neural network. F The fully connected operation is performed at each layer to construct a mapping relationship between the feature vectors of N window sample images and the corresponding N window sample labels, so that the fault detection and isolation results of the process industry to be tested can be output in the form of an image frame.

[0104] The following will be combined Figure 2 , further illustrating the practical application process of the fault detection and isolation device for process industry according to the embodiment of the present invention.

[0105] like Figure 2 As shown, variable data of the process industry to be tested can be obtained, and then a window sample set can be obtained based on the variable data. Principal component analysis, control limit, and fault contribution analysis can be performed on the window sample set to obtain a window sample label. In addition, the window sample set can be converted into an image to obtain a window sample image.

[0106] Furthermore, if Figure 2 As shown, the feature vector of the window sample image can be obtained by inputting the window sample image into the convolutional neural network for feature extraction, and the mapping relationship between the feature vector of the window sample image and the window sample label can be established by using a fully connected neural network, thereby realizing the coordinated fault detection and isolation of the process industry to be tested.

[0107] The beneficial effects of the present invention are as follows:

[0108] 1) The present invention converts the variable data of the process industry to be tested into image data, and can output the fault detection and isolation results of the process industry to be tested in the form of an image frame, thereby realizing the coordination of fault detection and isolation;

[0109] 2) The present invention can retain complete data information by converting variable data of the process industry to be tested into image pixel values, thereby realizing feature extraction within the window range and avoiding the problems of loss of process information and waste of storage space under the existing data visualization mechanism;

[0110] 3) The present invention constructs window sample labels through principal component analysis, thereby ensuring labeling accuracy and reducing the labor cost of labeling work. In addition, by using principal component analysis, when new working conditions appear in the production process or the working point drifts slowly, it is only necessary to use the covariance equivalence principle to add new data on the basis of the original fixed parameter dimension to achieve the update effect, which is convenient for subsequent maintenance and updating of the model.

[0111] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. "Multiple" means two or more, unless otherwise clearly and specifically defined.

[0112] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0113] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being "above", "above" or "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below", "below" or "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0114] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

Claims

1. A fault detection and isolation method for process industry, characterized in that: The following steps are involved: Obtain variable data of the process industry to be tested; Preprocessing the variable data to obtain training data, wherein the training data has no abnormal values ​​and no missing values; Constructing a window sample set of the training data; Obtaining a window sample image and a window sample label according to the window sample set; feature processing the window sample image; According to the feature-processed window sample image and the window sample label, fault detection and isolation are performed on the process industry to be tested. The window sample set of the training data is constructed, specifically comprising the following steps: obtaining data corresponding to the fault-free process in the training data; constructing a principal component analysis model for the fault-free process data; and constructing a window sample set of the training data according to the principal component analysis model. Monitoring indicator control limits; according to the The monitoring indicator control limits construct a window sample set of the training data under the rolling window mechanism, Specifically, the expert system selects the data set corresponding to the fault-free process from the training data. , and then the dataset Constructing the principal component analysis model: ， in, Represents the data set The score matrix, loading matrix and residual matrix in principal component analysis, Based on principal component analysis model Monitoring indicator control limits: ， in, represents the rank of the score matrix, Indicates confidence, In getting After monitoring the indicator control limits, construct a window sample set of training data under the rolling window mechanism: ， in, N Indicates the number of data in the window sample set, The step of obtaining a window sample image and a window sample label according to the window sample set specifically comprises the following steps: The monitoring index control limit determines whether there is a window sample corresponding to the fault process data in the window sample set; if so, the fault contribution of the window sample corresponding to the fault process data is analyzed, and a fault isolation model is constructed according to the fault contribution; the fault contribution rate of each window sample is obtained according to the fault isolation model; it is determined whether the fault contribution rate of each window sample exceeds the fault contribution rate threshold; if so, the fault interval range of the corresponding window sample is recorded; the window sample label of the window sample set is constructed according to the fault interval range; an image conversion model is constructed; and the data points in each window sample are converted into corresponding pixel parameter values ​​according to the image conversion model to obtain a window sample image according to the window sample set. Specifically, first, through the principal component analysis model and The monitoring index control limit performs fault analysis and judgment on each window sample in the window sample set: ， in, Indicates the length of the rolling window, , Furthermore, for the existence of more than Window sample data in the window sample set of the monitoring indicator control limit , build a fault isolation model based on contribution analysis: ， in, represents the window sample dimension, Indicates elements are 1 and the rest are 0 dimensional row vector, Furthermore, the fault contribution rate of each window sample is obtained according to the fault isolation model, and the fault contribution rate exceeds the fault contribution threshold. The corresponding window sample records the fault interval range and constructs the window sample label of the window sample set: ， in, Indicates window sample labels, , Window sample 1, window sample 2, ... window sample m The referent variable of ( , ) is the corresponding variable fault interval. For the fault-free window, the window sample label is , thus obtaining window sample labels, The process of obtaining a window sample image according to a window sample set includes the following steps: First, for each data point in the window sample set , an image conversion model is constructed by matching numerical intervals, namely, the three-primary-color RGB image conversion model: ， in, For variables A certain interval in the range of variation, for The corresponding pixel when converted into an image The red, green, and blue parameter values ​​in Further, according to the image conversion model, that is, the three primary colors RGB image conversion model, Insert a row of white pixels at each end of the row window image to obtain indivual dimensional window sample image, The method of performing fault detection and isolation on the process industry to be tested based on the feature-processed window sample image and the window sample label specifically includes the following steps: using a fully connected neural network to establish a mapping relationship between the feature-processed window sample image and the window sample label; and obtaining the fault detection and isolation result of the process industry to be tested based on the mapping relationship.

2. The fault detection and isolation method for process industry according to claim 1, characterized in that: The preprocessing of the variable data to obtain training data specifically includes the following steps: Use segmentation The variable data is preprocessed in principle to obtain the training data.

3. A fault detection and isolation device for process industry, characterized in that: include: An acquisition module, the acquisition module is used to acquire variable data of the process industry to be tested; A preprocessing module, the preprocessing module is used to preprocess the variable data to obtain training data, wherein the training data has no abnormal values ​​and no missing values; A data construction module, wherein the data construction module is used to construct a window sample set of the training data; A data conversion marking module, the data conversion marking module is used to obtain a window sample image and a window sample label according to the window sample set; A feature extraction module, the feature extraction module is used for feature processing of the window sample image; A fault detection and isolation module, the fault detection and isolation module is used to perform fault detection and isolation on the process industry to be tested according to the window sample image and the window sample label after feature processing, The data construction module is specifically used to obtain data corresponding to the fault-free process in the training data, construct a principal component analysis model for the fault-free process data, and construct Monitoring indicator control limits, according to the The monitoring indicator control limits construct a window sample set of the training data under the rolling window mechanism, Specifically, the expert system selects the data set corresponding to the fault-free process from the training data. , and then the dataset Constructing the principal component analysis model: ， in, Represents the data set The score matrix, loading matrix and residual matrix in principal component analysis, Based on principal component analysis model Monitoring indicator control limits: ， in, represents the rank of the score matrix, Indicates confidence, In getting After monitoring the indicator control limits, construct a window sample set of training data under the rolling window mechanism: ， in, N Indicates the number of data in the window sample set, The data conversion marking module is specifically used to: The monitoring index control limit determines whether there is a window sample corresponding to the fault process data in the window sample set; if so, the fault contribution of the window sample corresponding to the fault process data is analyzed, and a fault isolation model is constructed according to the fault contribution; the fault contribution rate of each window sample is obtained according to the fault isolation model; it is determined whether the fault contribution rate of each window sample exceeds the fault contribution rate threshold; if so, the fault interval range of the corresponding window sample is recorded; the window sample label of the window sample set is constructed according to the fault interval range; an image conversion model is constructed; and the data points in each window sample are converted into corresponding pixel parameter values ​​according to the image conversion model to obtain a window sample image according to the window sample set. Specifically, first, through the principal component analysis model and The monitoring index control limit performs fault analysis and judgment on each window sample in the window sample set: ， in, Indicates the length of the rolling window, , Furthermore, for the existence of more than Window sample data in the window sample set of the monitoring indicator control limit , build a fault isolation model based on contribution analysis: ， in, represents the window sample dimension, Indicates elements are 1 and the rest are 0 dimensional row vector, Furthermore, the fault contribution rate of each window sample is obtained according to the fault isolation model, and the fault contribution rate exceeds the fault contribution threshold. The corresponding window sample records the fault interval range and constructs the window sample label of the window sample set: ， in, Indicates window sample labels, , Window sample 1, window sample 2, ... window sample m The referent variable of ( , ) is the corresponding variable fault interval. For the fault-free window, the window sample label is , thus obtaining window sample labels, The process of obtaining a window sample image according to a window sample set includes the following steps: First, for each data point in the window sample set , an image conversion model is constructed by matching numerical intervals, namely, the three-primary-color RGB image conversion model: ， in, For variables A certain interval in the range of variation, for The corresponding pixel when converted into an image The red, green, and blue parameter values ​​in Further, according to the image conversion model, that is, the three primary colors RGB image conversion model, Insert a row of white pixels at each end of the row window image to obtain indivual dimensional window sample image, The fault detection and isolation module is specifically used to establish a mapping relationship between the window sample image after feature processing and the window sample label using a fully connected neural network, and obtain the fault detection and isolation result of the process industry to be tested according to the mapping relationship.

4. The fault detection and isolation device for process industry according to claim 3, characterized in that: The preprocessing module is specifically used to adopt segmentation The variable data is preprocessed in principle to obtain the training data.

Images