CN112001421B - Industrial control loop state identification method - Google Patents
Industrial control loop state identification method Download PDFInfo
- Publication number
- CN112001421B CN112001421B CN202010729517.XA CN202010729517A CN112001421B CN 112001421 B CN112001421 B CN 112001421B CN 202010729517 A CN202010729517 A CN 202010729517A CN 112001421 B CN112001421 B CN 112001421B
- Authority
- CN
- China
- Prior art keywords
- control loop
- data
- image
- stage
- industrial control
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses a state identification method of an industrial control loop, which comprises the following steps: s1: acquiring annotated industrial control loop image data; s2: constructing two-stage image feature learners based on convolution kernels with different sizes; s3: and identifying the state of the control loop according to the two-stage image feature learning device and the classifier based on a multi-time scale integrated decision-making method. The method replaces the traditional mode of carrying out complicated manual feature extraction and processing on industrial process data, has no requirement on the length of the data, can quickly, effectively and automatically extract potential features in control loop data, and can be used for control loop abnormity identification and performance evaluation.
Description
Technical Field
The invention relates to the technical field of industrial control, in particular to an industrial control loop state identification method.
Background
With the continuous development of scientific technology, an analog feedback control loop based on a PID (proportional-Integral-Derivative) controller, a PI (proportional-Integral-Derivative) controller, a PD (proportional-Derivative) controller, a state observer, a phase compensator, various filters, a feedforward controller, a cascade loop controller, and a 2 × 2 multi-loop controller is generally applied to the fields of industrial production, automation control, and the like to form a corresponding feedback control system. For a feedback control system, how to identify the state of a control loop in the feedback control system is an extremely important technical problem.
The in-depth analysis of the control loop data is beneficial to monitoring and optimizing the control loop. In particular, there are a number of control loops for temperature, pressure, flow, etc. in the process industry. The effective monitoring and analysis of the loop data is beneficial to improving the performance of the whole loop and reducing the energy consumption and saving the cost. When a control loop has just been put into use, its performance is often better. However, the degradation of the performance of the control loop may occur due to aging of the actuator in the control loop or characteristic change of the control object. In actual production, stopping the control loop operation and detecting it is not a feasible solution, and therefore it is very valuable to mine features in the loop from production data to evaluate the operating state of the control loop.
The existing control loop identification only utilizes observation characteristics on images, actually processed data or industrial process time sequence data, the characteristics of the control loop in the traditional method are often manually extracted according to the experience of an operator or the production mechanism of the control loop, and different characteristics are often required to reflect different states, so that the working difficulty of field operators is greatly increased.
Disclosure of Invention
The invention provides an industrial control loop state identification method based on image feature automatic learning, aiming at solving the problem that the existing control loop needs manual identification, and the method has important significance for analyzing the operation state of the control loop.
In order to achieve the purpose, the invention adopts the following technical scheme:
the method for identifying the state of the industrial control loop comprises the following steps:
s1: acquiring annotated industrial control loop image data;
s2: constructing two-stage image feature learners based on convolution kernels with different sizes;
s3: and identifying the state of the control loop according to the two-stage image feature learning device and the classifier based on a multi-time scale integrated decision-making method.
In the method of the present invention, the step S1 specifically includes the steps of:
s101: acquiring historical image data of an industrial production process, calibrating the working condition corresponding to the image according to the working condition, and expressing an image data set as T { (img)1,c1),(img2,c2),...,(imgN,cN) Where N denotes the total number of acquired images and the annotated class is denoted ciWhere i ∈ 1iE, 1, C, C represents the total number of categories;
s102: unifying pixel sizes of all images in the image dataset.
In the method of the present invention, the step S2 specifically includes the steps of:
s201: constructing a first-stage feature learner based on convolution;
s202: and constructing a second-stage feature learner based on the convolution.
In the method of the present invention, the pixel size is 32 × 32.
In the method of the present invention, the first stage feature learner is composed of 8 convolution kernels of 5 × 5 and maximum pooling level of 2 × 2, and the second stage feature learner is composed of 20 convolution kernels of 3 × 3 and maximum pooling level of 2 × 2.
In the method of the present invention, the step S3 specifically includes the steps of:
s301: given new control loop data, set toAndt represents the length of the data, and the set of time scales is set as L epsilon { L1,l2,...,lMIn which liRepresenting the ith time scale, obtaining data of OP and PV data at the ith time scale asAndwill be provided withAndthe formed image is set as the ith image, and M images can be finally obtained by each control loop on the assumption that M different time scales exist;
s302: m recognition results are obtained through a two-stage image feature learning device and are expressed asM is an odd number;
s303: and obtaining a category identification result by adopting a strategy of integrating voting.
In the method of the present invention, the category identification result takes the category with a large number of occurrences as the category identification result.
Compared with the prior art, the method replaces the traditional mode of carrying out complicated manual feature extraction and processing on industrial process data, obtains a control loop mode recognizer with higher practicability and accuracy by training by inputting historical image data into a designed convolutional neural network, can judge the running state of a control loop more reliably based on a multi-time scale recognition strategy, can enhance the robustness and reliability of a final result, has no requirement on the length of the data, and can quickly, effectively and automatically extract potential features in the control loop data, thereby further realizing effective recognition of different running modes of the industrial control loop.
Drawings
FIG. 1 is a flow chart of a method for identifying the status of an industrial control loop according to the present invention;
FIG. 2 is an exemplary graph of image data collected by an industrial process of the present invention;
FIG. 3 is an exemplary diagram of the features automatically extracted by the first level learner of the present invention;
FIG. 4 is an exemplary diagram of the features automatically extracted by the second stage learner of the present invention;
FIG. 5 is a diagram of an exemplary multi-time scale image data under a single loop.
Detailed Description
The following embodiments of the present invention are provided by way of specific examples, and other advantages and effects of the present invention will be readily apparent to those skilled in the art from the disclosure herein. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. It is to be noted that the features in the following embodiments and examples may be combined with each other without conflict.
It should be noted that the drawings provided in the following embodiments are only for illustrating the basic idea of the present invention, and the components related to the present invention are only shown in the drawings rather than drawn according to the number, shape and size of the components in actual implementation, and the type, quantity and proportion of the components in actual implementation may be changed freely, and the layout of the components may be more complicated.
The invention provides a state identification method of an industrial control loop, which comprises the following steps as shown in figure 1:
s1: acquiring annotated industrial control loop image data;
s2: constructing two-stage image feature learners based on convolution kernels with different sizes;
s3: and identifying the state of the control loop according to the two-stage image feature learning device and the classifier based on a multi-time scale integrated decision-making method.
Further, in an embodiment of the present invention, the step S1 specifically includes the steps of:
s101: acquiring historical image data of an industrial production process, calibrating the working condition corresponding to the image according to the working condition, and expressing an image data set as T { (img)1,c1),(img2,c2),...,(imgN,cN)},Where N represents the total number of acquired images and the annotated class is denoted ciWhere i ∈ 1iE, 1, C, C represents the total number of categories;
s102: unifying pixel sizes of all images in the image data set.
Specifically, taking the detection of control loop viscosity as an example, the control loop viscosity is mainly caused by insensitivity of a valve, and severe viscosity can cause oscillation of a loop to cause performance degradation, and image data with category labels is obtained from industrial production historical data, as shown in fig. 2 (the abscissa represents OP, the ordinate represents PV, the loop 1 is labeled as a viscous loop, and the loop 2 is labeled as a non-viscous loop). The image pixel size is set to 32 x 32. Fig. 2 shows sample examples of a viscous loop image and a non-viscous loop image.
Further, in an embodiment of the present invention, the step S2 specifically includes the steps of:
s201: constructing a first-stage feature learner based on convolution;
s202: and constructing a second-stage feature learner based on the convolution.
Specifically, the first-stage feature learner is composed of 8 convolution kernels of 5 × 5 and a maximum pooling layer of 2 × 2, and the first-stage feature learner uses a larger convolution kernel size to learn a larger range or more obvious feature in the image, as shown in fig. 3, taking a loop 1 as an example, the upper row of fig. 3 represents the feature extracted by the convolution kernel, and the lower row represents the feature passing through the maximum pooling layer.
The second level of feature learner, which consists of 20 convolution kernels of 3 x 3 and 2 x 2 of the largest pooling layer, uses a smaller convolution kernel size to learn smaller or finer features in the image, with the extracted features shown in fig. 4.
By adopting two stages of convolution kernels with different sizes, the characteristics of large scale and small scale in the image can be captured simultaneously. But also different numbers of convolution kernels can ensure that as many features as possible are learned. Let the input image be characterized as input (1,32,32), and the features extracted by the convolution kernel areWhereinRepresenting the weighting factor on the output jth channel. The parameters of the two-stage feature learner are shown in the following table.
The features extracted by the two-stage learner are represented as feats (N)2,H2,W2) Where N2 denotes the number of convolution kernels of the second stage feature learner, H2 and W2 respectively characterize the final extracted image height and width. And inputting the group of characteristics into a designed full-connection neural network classifier, wherein the output of the full-connection neural network is the corresponding control loop type. Before the network is fully connected, a dropout layer is added, and the dropout layer has the function of randomly screening the features to avoid generating redundant features. Where the final output layer parameter C represents the total number of categories of the acquired image.
It can be seen that through two-stage feature extraction, image features under different scales can be obtained at the same time, and multiple features can also be obtained under the same scale. And inputting the extracted features into a designed fully-connected neural network classifier, and finally outputting whether a control loop is sticky.
Further, in an embodiment of the present invention, the step S3 specifically includes the steps of:
s301: given new control loop data, set toAndt represents the length of the data, and the set of time scales is set as L epsilon { L1,l2,...,lMH, wherein liRepresenting the ith time scale, obtaining data of OP and PV data at the ith time scale asAndwill be provided withAndthe formed image is set as the ith image, and M images can be finally obtained by each control loop on the assumption that M different time scales exist;
s302: m recognition results are obtained through a two-stage image feature learning device and are expressed asM is an odd number;
s303: and obtaining a category identification result by adopting a strategy of integrating voting.
And finally judging the identification result, and adopting an integrated voting strategy. The final decision result can be expressed as:where N and S represent different categories, respectively, | N | and | S | represent the number of occurrences of each category in the final result R, respectively. The final result takes the category with more occurrence times as the recognition result.
When a new control loop is obtained, the control loop OP/PV data is first plotted into OP-PV images at multiple time scales. An example of a multi-time scale image is shown in fig. 5. This example selects 5 different time scales. The 5 images are identified simultaneously by the model in step S2, and the identification result is shown as loop 1 in table 1, while the identification of the remaining 4 control loops is summarized in the present embodiment as shown in the following table.
Loop circuit | Dimension 1 | |
Dimension 3 | |
|
Final decision | |
1 | Viscous glue | Viscous glue | Viscous glue | Viscous glue | Viscous | Viscous glue | |
2 | Viscous glue | Viscous glue | Non-stick | Viscous glue | Viscous glue | Viscous glue | |
3 | Viscous glue | Non-stick | Viscous glue | Non-stick | Non-stick | Non-stick | |
4 | Viscous glue | Viscous glue | Viscous glue | Viscous glue | Viscous | Viscous glue | |
5 | Viscous glue | Viscous glue | Viscous glue | Non-stick | Non-stick | Viscous glue |
The method replaces the traditional mode of carrying out complicated manual feature extraction and processing on industrial process data, obtains a control loop mode recognizer with higher practicability and accuracy by inputting historical image data into a designed convolutional neural network, can judge the running state of a control loop more reliably based on a multi-time scale recognition strategy, can enhance the robustness and reliability of a final result, has no requirement on the length of the data, and can quickly, effectively and automatically extract potential features in the control loop data, thereby further realizing the effective recognition of different running modes of the industrial control loop.
It is to be noted that the foregoing is only illustrative of the preferred embodiments of the present invention and the technical principles employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
Claims (6)
1. An industrial control loop state identification method is characterized by comprising the following steps:
s1: acquiring annotated industrial control loop image data;
s2: constructing two-stage image feature learners based on convolution kernels with different sizes;
s3: identifying the state of a control loop according to the two-stage image feature learning device and the classifier based on a multi-time scale integrated decision-making method;
the step S3 specifically includes the steps of:
s301: given new control loop data, set toAndt represents the length of the data, and the set of time scales is set as L epsilon { L1,l2,…,lMIn which liRepresenting the ith time scale, obtaining data of OP and PV data at the ith time scale asAndwill be provided withAndthe formed image is set as the ith image, and M images can be finally obtained by each control loop on the assumption that M different time scales exist;
s302: m recognition results are obtained through a two-stage image feature learning device and are expressed asM is an odd number;
s303: and obtaining a category identification result by adopting a strategy of integrating voting.
2. The industrial control loop state identification method according to claim 1,
the step S1 specifically includes the steps of:
s101: acquiring historical image data of an industrial production process, calibrating the working condition corresponding to the image according to the working condition, and expressing an image data set as T { (img)1,c1),(img2,c2),…,(imgN,cN) Where N denotes the total number of acquired images and the annotated class is denoted ci'Where i' is e.1i'E, 1, C, C represents the total number of categories;
s102: unifying pixel sizes of all images in the image dataset.
3. The industrial control loop state identification method according to claim 2,
the step S2 specifically includes the steps of:
s201: constructing a first-stage feature learner based on convolution;
s202: and constructing a second-stage feature learner based on the convolution.
4. The industrial control loop state identification method according to claim 2,
the pixel size is 32 x 32.
5. The industrial control loop state identification method according to claim 3,
the first stage of feature learner is composed of 8 convolution kernels of 5 × 5 and maximum pooling level of 2 × 2, and the second stage of feature learner is composed of 20 convolution kernels of 3 × 3 and maximum pooling level of 2 × 2.
6. The industrial control loop state identification method according to claim 5,
and the category identification result takes the category with more occurrence times as the category identification result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010729517.XA CN112001421B (en) | 2020-07-27 | 2020-07-27 | Industrial control loop state identification method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010729517.XA CN112001421B (en) | 2020-07-27 | 2020-07-27 | Industrial control loop state identification method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112001421A CN112001421A (en) | 2020-11-27 |
CN112001421B true CN112001421B (en) | 2022-06-10 |
Family
ID=73467225
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010729517.XA Active CN112001421B (en) | 2020-07-27 | 2020-07-27 | Industrial control loop state identification method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112001421B (en) |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3824328B2 (en) * | 1997-12-05 | 2006-09-20 | シチズン時計株式会社 | Ferroelectric liquid crystal device with burn-in prevention and recovery processing means |
US10755334B2 (en) * | 2016-06-30 | 2020-08-25 | Varmour Networks, Inc. | Systems and methods for continually scoring and segmenting open opportunities using client data and product predictors |
CN109060001B (en) * | 2018-05-29 | 2021-02-26 | 浙江工业大学 | Multi-working-condition process soft measurement modeling method based on feature transfer learning |
US10747207B2 (en) * | 2018-06-15 | 2020-08-18 | Honeywell International Inc. | System and method for accurate automatic determination of “alarm-operator action” linkage for operator assessment and alarm guidance using custom graphics and control charts |
US11294367B2 (en) * | 2018-12-21 | 2022-04-05 | Canvass Analytics Inc. | Predictive analysis of industrial processes |
CN110555479B (en) * | 2019-09-05 | 2023-05-16 | 北京工业大学 | Fault feature learning and classifying method based on 1DCNN and GRU fusion |
-
2020
- 2020-07-27 CN CN202010729517.XA patent/CN112001421B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN112001421A (en) | 2020-11-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2020244134A1 (en) | Multi-task feature sharing neural network-based intelligent fault diagnosis method | |
CN112101220A (en) | Rolling bearing service life prediction method based on unsupervised model parameter migration | |
Li et al. | Attention-based deep meta-transfer learning for few-shot fine-grained fault diagnosis | |
CN113052218A (en) | Multi-scale residual convolution and LSTM fusion performance evaluation method for industrial process | |
CN112541552B (en) | Air handling unit fault detection and diagnosis method combining DCCNN and LGBM | |
CN113837000A (en) | Small sample fault diagnosis method based on task sequencing meta-learning | |
CN113297885A (en) | Belt conveyor surface state detection method and device based on convolutional neural network | |
WO2022142426A1 (en) | Deep learning-based test anomaly analysis system for integrated circuits, and method therefor | |
CN110757510A (en) | Method and system for predicting remaining life of robot | |
Akpudo et al. | An automated sensor fusion approach for the rul prediction of electromagnetic pumps | |
Variz et al. | Machine learning applied to an intelligent and adaptive robotic inspection station | |
CN115169709A (en) | Power station auxiliary machine fault diagnosis method and system based on data driving | |
CN112001421B (en) | Industrial control loop state identification method | |
Liu et al. | Explainable fault diagnosis of oil-gas treatment station based on transfer learning | |
KR20230003819A (en) | Facility management system that enables preventive maintenance using deep learning | |
CN116929815A (en) | Equipment working state monitoring system and method based on Internet of things | |
CN115587299A (en) | Transferable multi-scale rotating machine fault diagnosis method and system | |
CN115578325A (en) | Image anomaly detection method based on channel attention registration network | |
CN115358137A (en) | Bearing metal temperature prediction method based on CNN-GRU algorithm | |
CN112016240B (en) | Prediction method for residual stable service life of incomplete degradation equipment with similar evidence | |
Li et al. | Comparison of data-driven prognostics models: A process perspective | |
CN109523195B (en) | Sample multi-modal classification-based fault detection method for hierarchical non-Gaussian algorithm | |
Sun et al. | Bearing feature extraction and fault diagnosis algorithm based on convolutional neural networks | |
MacGregor | Data-based latent variable methods for process analysis, monitoring and control | |
CN110554667A (en) | convolutional Neural Network (CNN) based intermittent industrial process fault diagnosis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |