CN114841250A - Industrial system production abnormity detection and diagnosis method based on multi-dimensional sensing data - Google Patents

Industrial system production abnormity detection and diagnosis method based on multi-dimensional sensing data Download PDF

Info

Publication number
CN114841250A
CN114841250A CN202210375778.5A CN202210375778A CN114841250A CN 114841250 A CN114841250 A CN 114841250A CN 202210375778 A CN202210375778 A CN 202210375778A CN 114841250 A CN114841250 A CN 114841250A
Authority
CN
China
Prior art keywords
feature
subsample
layer
sensing data
training
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.)
Pending
Application number
CN202210375778.5A
Other languages
Chinese (zh)
Inventor
吕明琪
周丹
朱添田
陈铁明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN202210375778.5A priority Critical patent/CN114841250A/en
Publication of CN114841250A publication Critical patent/CN114841250A/en
Priority to PCT/CN2022/135714 priority patent/WO2023197617A1/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield 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
    • G06N3/04—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
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention discloses a method for detecting and diagnosing production abnormity of an industrial system based on multidimensional sensing data, which comprises the following steps: preprocessing a multi-dimensional sensing data sample, and dividing the preprocessed multi-dimensional sensing data sample into a plurality of sub-samples by adopting a sliding window; an automatic coding machine is adopted, and an abnormal detection model is obtained based on normal sub-sample training in an unsupervised training mode; training a classification model according to the anomaly detection model; and carrying out real-time detection and diagnosis on the production abnormity of the industrial system based on the abnormity detection model and the classification model. The invention solves the problem that the abnormality diagnosis is difficult to carry out under the condition of the abnormality detection of the multi-dimensional sensing data by using a black box model at present.

Description

Industrial system production abnormity detection and diagnosis method based on multi-dimensional sensing data
Technical Field
The invention belongs to the technical field of data mining, and particularly relates to a method for detecting and diagnosing production abnormity of an industrial system based on multi-dimensional sensing data.
Background
The industrial internet aims to achieve automation control and resource allocation of more acute and efficient industrial manufacturing systems while improving the production efficiency of intelligent factories. However, as the industrial internet breaks the boundaries of the network world and the physical world, industrial manufacturing systems are more vulnerable to external malicious activities. In addition, there are inevitable production problems in industrial manufacturing systems, such as equipment failure, performance degradation, quality defects, and the like. If the abnormal conditions such as intrusion, failure and the like in the industrial production cannot be detected in time, the serious loss can be brought to the whole manufacturing system. Therefore, the anomaly detection and diagnosis are basic requirements of the industrial Internet, and have very important significance for intelligent manufacturing enterprises.
With the rapid development of the industrial internet, the modern industrial manufacturing system realizes the sensing and recording of the production running state and process through a sensor, a large amount of industrial production data is accumulated, and a data driving method becomes a mainstream means for anomaly detection. In recent years, deep learning has become a mainstream technique of a data-driven method. However, since the deep learning model is too complex, contains a large number of non-linear transformations, and is a black box as a whole, the prediction result is not interpretable. In the industrial system abnormality detection, the interpretation of the detection result is very important, and is a basis for realizing the diagnosis of the abnormality detection result. For example, the anomaly detection result diagnosis can help locate which device and which time period an anomaly occurred.
The existing methods for explaining the deep learning model all focus on supervised learning models, such as SHAP, LIME and other deep learning interpretable frameworks. However, the industrial production data is very complex, and the cost of manual labeling is too high, so that the obtained industrial production data is basically label-free, and therefore the anomaly detection model needs to be trained in an unsupervised mode. In particular, the existing deep learning interpretable framework can hardly learn the association between the abnormal sample and the semantic feature, so that the deep unsupervised learning model can not be interpreted.
Disclosure of Invention
The invention aims to provide a method for detecting and diagnosing production abnormity of an industrial system based on multi-dimensional sensing data, and improve the accuracy of abnormity diagnosis.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
the industrial system production abnormity detection and diagnosis method based on the multi-dimensional sensing data comprises the following steps:
s1, preprocessing the multi-dimensional sensing data sample, and dividing the preprocessed multi-dimensional sensing data sample into a plurality of sub-samples by adopting a sliding window, wherein the sub-samples comprise normal sub-samples and abnormal sub-samples; given multi-dimensional sensing data sample s ∈ R N×T S is a two-dimensional matrix, wherein N is the characteristic dimension of s, namely the number of devices contained in the industrial system, and T is the data duration of s, namely the number of sampling points of the sensor;
s2, obtaining an abnormal detection model based on normal sub-sample training in an unsupervised training mode by adopting an automatic coding machine;
s3, training a classification model according to the abnormality detection model, including:
step 31, detecting the subsamples containing the normal subsamples and the abnormal subsamples by using an abnormal detection model, and adding marks to the subsamples according to a detection result to obtain a marked subsample set;
Step 32, assuming that F is a set of N features, and obtaining 2 by taking N features in the set F each time according to the combination of the features N -1 feature subset S, N-1, 2, …, N, generating a training subset from the set of labeled sub-samples according to each feature subset, the training subset comprising only features of the feature subset, training a classification model on each training subset in a supervised manner using XGBoost classifier, and obtaining 2 in total N -1 classification model;
s4, real-time detection and diagnosis are carried out on the production abnormity of the industrial system based on the abnormity detection model and the classification model, and the method comprises the following steps:
acquiring a real-time subsample to be detected, and finishing if the detection result of the abnormal detection model on the real-time subsample is a normal subsample; otherwise, the classification model is used for sequentially calculating the feature confidence coefficient corresponding to each feature dimension in the N feature dimensions according to the real-time subsamples, and abnormal features are diagnosed according to the feature confidence coefficients, namely, abnormal equipment in the industrial system is positioned.
Several alternatives are provided below, but not as an additional limitation to the above general solution, but merely as a further addition or preference, each alternative being combinable individually for the above general solution or among several alternatives without technical or logical contradictions.
Preferably, the preprocessing the multi-dimensional sensing data sample includes:
filling missing values in a multi-dimensional sensing data sample s by adopting an average value of front and back data;
the multidimensional sensing data sample s is normalized to have data in the range of [0,1 ].
Preferably, the dividing the preprocessed multidimensional sensing data sample into a plurality of sub-samples by using a sliding window includes:
dividing the multi-dimensional sensing data sample s by using a sliding window with the window size W to obtain continuous M sub-samples sS E for R N×W 。
Preferably, the network structure of the automatic coding machine comprises an input layer, a coding layer, a semantic layer, a decoding layer and an output layer, wherein:
the input layer: the input is subsamples Ss ∈ R N×W ；
The coding layer: two layers of LSTM are used as an encoder, and N-dimensional feature vectors x of W time instants in a subsample ss 1 ,x 2 ,…,x W Sequentially inputting each unit of the first layer of LSTM, sequentially inputting the obtained W hidden vectors into each unit of the second layer of LSTM, and obtaining W hidden vectors h 1 ,h 2 ,…,h W ；
The semantic layer: taking a hidden vector h W As a low-dimensional semantic vector after encoding;
the decoding layer: two layers of LSTM are adopted as a decoder, and an implicit vector h is generated W Repeating W times and inputting each unit of the first layer of LSTM in sequence to obtain W hidden vectorsInputting each unit of the second layer LSTM in sequence to obtain W hidden vectors g 1 ,g 2 ,…,g w ；
The output layer: adopting a full connection layer to hide W vectors g 1 ,g 2 ,…,g W Conversion into a vector y conforming to the dimension ss of the subsamples 1 ,y 2 ,…,y W Vector y 1 ,y 2 ,…,y W As output data rss.
Preferably, in the training of the anomaly detection model, the mean square error of the output data rss and the sub-sample ss is used as a loss function, and optimization iteration is performed in a gradient descent mode.
Preferably, the sequentially calculating the feature confidence corresponding to each feature dimension of the N feature dimensions according to the real-time subsample includes:
the feature confidence is computed for feature k as follows:
Figure BDA0003590324810000031
in the formula, phi k For feature confidence of feature k, k is 1,2, …, N, CM S (x S ) Classification model CM trained for training subsets S corresponding to feature subsets that do not contain feature k S At subsample x S The output result is 0 or 1, the subsample x S For sample data extracted from the real-time subsamples that are identical to the features contained in the feature subset S, CM S∪{k} (x S∪{k} ) Classification model CM trained using training subsets corresponding to feature subsets S { k } containing feature k S∪{k} At subsample x S∪{k} The output result is 0 or 1, the subsample x S∪{k} The sample data extracted from the real-time subsamples and identical to the features contained in the feature subset S { k },
Figure BDA0003590324810000032
representing a feature subset S that does not contain feature k.
Preferably, the diagnosing the abnormal feature according to the feature confidence includes:
firstly, normalizing the confidence degrees of all the features by adopting a Sigmoid function to obtain a weight score, wherein the absolute value of the weight score indicates the influence value of the features on the final detection result, and the detection result is explained by utilizing a SHAP explanation model based on the influence value.
The industrial system production abnormity detection and diagnosis method based on the multidimensional sensing data provided by the invention adopts an automatic coding machine to train an abnormity detection model in an unsupervised mode without providing an abnormity marking sample; the output of the anomaly detection model is used for training the supervised classification model, so that the interpretation and diagnosis of the anomaly detection result are realized on the basis, and the problem that the anomaly diagnosis is difficult to perform under the condition that the multi-dimensional sensing data anomaly detection is performed by using the black box model at present is solved.
Drawings
FIG. 1 is a flow chart of a multi-dimensional sensory data based industrial system production anomaly detection and diagnosis method of the present invention;
FIG. 2 is a diagram of a trellis diagram of the automatic coding machine of the present invention;
FIG. 3 is a schematic diagram of the setup of parameters of each layer of the automatic coding machine according to the present invention;
FIG. 4 is an explanatory diagram of anomaly detection for an anomalous subsample output in accordance with the present invention;
FIG. 5 is an explanatory diagram of the anomaly detection of the normal sub-sample output according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.
In order to solve the problem that abnormality diagnosis is difficult to perform under the condition of multi-dimensional sensing data abnormality detection by using a black box model in the prior art, the embodiment provides an industrial system production abnormality detection and diagnosis method based on multi-dimensional sensing data.
As shown in fig. 1, the embodiment provides a method for detecting and diagnosing production abnormality of an industrial system based on multidimensional sensing data, which includes the following steps:
s1, preprocessing the multi-dimensional sensing data sample, and dividing the preprocessed multi-dimensional sensing data sample into a plurality of sub-samples by adopting a sliding window, wherein the sub-samples comprise normal sub-samples and abnormal sub-samples.
In the embodiment, a given multidimensional sensing data sample s E R N×T S is a two-dimensional matrix, where N is the characteristic dimension of s, i.e., the number of devices included in the industrial system, and T is the data duration of s, i.e., the number of sampling points of the sensor. Therefore, the detailed operation of the present embodiment to preprocess the sample data is as follows:
1) data cleaning: and filling missing values in the multi-dimensional sensing data sample s by adopting the average value of the previous data and the next data.
2) Data normalization: the multidimensional sensing data sample s is normalized to have data in the range of [0,1 ].
In the embodiment, when data is divided, a sliding window with the window size W is used for dividing a multi-dimensional sensing data sample s to obtain continuous M sub-samples ss ∈ R N×W 。
It should be noted that, in the embodiment, since the normal subsample is required to be used for training the anomaly detection model, the subsamples in the embodiment include the normal subsample and the anomaly subsample, but the subsamples obtained by dividing in step 1 are not labeled, and the normal and anomaly differences correspond to the normal and anomaly of the original data.
And S2, obtaining an abnormal detection model based on normal sub-sample training in an unsupervised training mode by adopting an automatic coding machine.
In the embodiment, an automatic coding machine is adopted to train an anomaly detection model AM, the input of the automatic coding machine is an original sub-sample ss, the original sub-sample is converted into a low-dimensional feature space through an encoder, the low-dimensional feature is output as a reconstructed sub-sample rss through a decoder, and the training aim is to enable ss and rss to be as close as possible. Referring to fig. 2, the network structure of the employed automatic coding machine is as follows:
an input layer: the input is subsample ss ∈ R N×W 。
And (3) coding layer: using two layers of LSTM as the encoder, the N-dimensional feature vector x for each time instant (W time instants in total) in the subsample ss 1 ,x 2 ,…,x W (for a total of W times, i.e. x 1 For an N-dimensional feature vector, other same understanding) are sequentially inputted into each unit of the first layer of LSTM, and the obtained W hidden vectors are sequentially inputted into each unit of the second layer of LSTM, so as to obtain W hidden vectors h 1 ,h 2 ,…,h W 。
And (3) semantic layer: taking a hidden vector h W As encoded low-dimensional semantic vectors.
A decoding layer: two layers of LSTM are adopted as a decoder, and an implicit vector h is generated W Repeating W times and inputting each unit of the first layer of LSTM in sequence to obtain W hidden vectors, and inputting each unit of the second layer of LSTM in sequence to obtain W hidden vectors g 1 ,g 2 ,…,g w 。
An output layer: adopting a full connection layer to hide W vectors g 1 ,g 2 ,…,g w Conversion into a vector y conforming to the dimension ss of the subsamples 1 ,y 2 ,…,y W Vector y 1 ,y 2 ,…,y W As output data rss.
In the training process of the anomaly detection model AM, on one hand, in order to minimize the difference between ss and rss, the mean square error of output data rss and a subsample ss is used as a loss function, and on the basis, the model is optimally trained in a gradient descent mode; on the other hand, in order to make the model learn the pattern of the normal subsample, the normal subsample is used for training. The parameter settings of each layer of the automatic coding machine adopted in the present embodiment are shown in fig. 3.
After the training is finished, a real-time sub-sample ss is given and input into a trained automatic coding machine (namely an anomaly detection model AM), and an output reconstructed sub-sample rss is obtained. And calculating the mean square error of ss and rss, if the mean square error is larger than a predefined threshold value, judging the subsample to be abnormal, otherwise, judging the subsample to be normal.
S3, training a classification model according to the abnormality detection model, including:
step 31, constructing a labeled sub-sample set: and detecting the subsamples containing the normal subsamples and the abnormal subsamples by using an abnormal detection model, adding marks to the subsamples according to a detection result, marking the detected abnormal subsamples as 1, marking the normal subsamples as 0, and obtaining a marked subsample set LSS.
Step 32, constructing a classification model: assuming that F is a set of N features, taking N features in the set F each time according to the combination of the features to obtain 2 N -1 feature subset S, N-1, 2, …, N, generating a training subset comprising only features in the feature subset from the labeled subsample set LSS according to each feature subset, training a classification model on each training subset in a supervised manner using an XGBoost classifier, and obtaining 2 in total N -1 classification model.
The number of the subsamples in each training subset is the same as that in the LSS labeled subsample set, and the labels of the subsamples are unchanged.
S4, real-time detection and diagnosis are carried out on the production abnormity of the industrial system based on the abnormity detection model and the classification model, and the method comprises the following steps:
s41, acquiring a real-time subsample to be detected, and ending if the detection result of the abnormal detection model on the real-time subsample is a normal subsample; otherwise, executing the next step;
and S42, sequentially calculating the feature confidence corresponding to each feature dimension in the N feature dimensions according to the real-time subsamples by using the classification model.
The embodiment calculates the feature confidence corresponding to each feature when the real-time sub-sample is abnormal so as to calculate the feature of the feature kTo illustrate the confidence level, given a feature k, (a feature is one of N features) the confidence level φ for a feature k is evaluated by computing the difference between a classification model using the feature k and a classification model not using the feature k k Degree of confidence phi k The larger the feature k is, the higher the importance degree is.
The feature confidence is computed for feature k as follows:
Figure BDA0003590324810000071
in the formula, phi k For feature confidence of feature k, k is 1,2, …, N, CM S (x S ) Classification model CM trained for training subsets S corresponding to feature subsets that do not contain feature k S At subsample x S The output result is 0 or 1, the subsample x S For sample data extracted from the real-time subsamples that are identical to the features contained in the feature subset S, CM S∪{k} (x S∪{k} ) Classification model CM trained using training subsets corresponding to feature subsets S { k } containing feature k S∪{k} At subsample x S∪{k} The output result is 0 or 1, the subsample x S∪{k} The sample data extracted from the real-time subsamples and identical to the features contained in the feature subset S { k },
Figure BDA0003590324810000072
representing a feature subset S that does not contain feature k.
And S43, diagnosing abnormal features according to the feature confidence, namely positioning abnormal equipment in the industrial system.
Firstly, normalizing the confidence degrees of all the features by adopting a Sigmoid function to obtain a weight score, wherein the absolute value of the weight score indicates the influence value of the features on the final detection result, and the detection result is explained by utilizing a SHAP explanation model based on the influence value.
The abnormity judgment is established on the SHAP interpretation model, and the SHAP interpretation model is realized based on the Shapley value, so the calculation of the feature confidence coefficient in the embodiment is equivalent to the calculation of the Shapley value. And (4) obtaining abnormal characteristics (for example, the influence value is higher than a set threshold) based on the finally obtained influence value, and directly locating the equipment which can cause the industrial system to generate the abnormity because the characteristics correspond to the equipment.
For convenience of observation, the diagnosis result is further visualized in the present embodiment, and referring to fig. 4 and 5, in the figure, f (x) represents the probability that the output of the classification model is an abnormal result, the left side of f (x) represents positive correlation to the abnormal detection result, the right side represents negative correlation to the abnormal result, and the larger the feature region width is, the higher the weight score of the feature is, thereby diagnosing the cause of the abnormality. For example, in fig. 4, the influence values of the features (labels of corresponding devices) such as f4, f6, f1 are high, which indicates that the most probable cause of the abnormality is generated by the devices corresponding to the features.
The technical features of the embodiments described above may be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the embodiments described above are not described, but should be considered as being within the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above examples are merely illustrative of several embodiments of the present invention, and the description thereof is more specific and detailed, but not to be construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present invention should be subject to the appended claims.

Claims (7)

1. The industrial system production abnormity detection and diagnosis method based on the multidimensional sensing data is characterized by comprising the following steps:
s1, preprocessing the multi-dimensional sensing data sample, and dividing the preprocessed multi-dimensional sensing data sample into a plurality of sub-samples by adopting a sliding window, wherein the sub-samples comprise normal sub-samples and abnormal sub-samplesThen, the process is carried out; given multi-dimensional sensing data sample s ∈ R N×T S is a two-dimensional matrix, wherein N is the characteristic dimension of s, namely the number of devices contained in the industrial system, and T is the data duration of s, namely the number of sampling points of the sensor;
s2, obtaining an abnormal detection model based on normal sub-sample training in an unsupervised training mode by adopting an automatic coding machine;
S3, training a classification model according to the abnormality detection model, including:
step 31, detecting the subsamples containing the normal subsamples and the abnormal subsamples by using an abnormal detection model, and adding marks to the subsamples according to a detection result to obtain a marked subsample set;
step 32, assuming that F is a set of N features, and obtaining 2 by taking N features in the set F each time according to the combination of the features N -1 feature subset S, N-1, 2, …, N, generating a training subset from the set of labeled sub-samples according to each feature subset, the training subset comprising only features of the feature subset, training a classification model on each training subset in a supervised manner using XGBoost classifier, and obtaining 2 in total N -1 classification model;
s4, real-time detection and diagnosis are carried out on the production abnormity of the industrial system based on the abnormity detection model and the classification model, and the method comprises the following steps:
acquiring a real-time subsample to be detected, and finishing if the detection result of the abnormal detection model on the real-time subsample is a normal subsample; otherwise, the classification model is used for sequentially calculating the feature confidence coefficient corresponding to each feature dimension in the N feature dimensions according to the real-time subsamples, and abnormal features are diagnosed according to the feature confidence coefficients, namely, abnormal equipment in the industrial system is positioned.
2. The method for detecting and diagnosing production anomalies in an industrial system based on multidimensional sensed data as recited in claim 1, wherein the preprocessing the multidimensional sensed data samples comprises:
filling missing values in a multi-dimensional sensing data sample s by adopting an average value of front and back data;
the multidimensional sensing data sample s is normalized to have data in the range of [0,1 ].
3. The method for detecting and diagnosing production abnormality of industrial system based on multi-dimensional sensing data as claimed in claim 1, wherein the dividing the preprocessed multi-dimensional sensing data sample into a plurality of sub-samples by using a sliding window comprises:
dividing the multi-dimensional sensing data sample s by using a sliding window with the window size W to obtain continuous M sub-samples ss ∈ R N×W 。
4. The method of claim 3, wherein the network structure of the automatic coding machine comprises an input layer, a coding layer, a semantic layer, a decoding layer and an output layer, wherein:
the input layer: the input is subsample ss ∈ R N×W ；
The coding layer: two layers of LSTM are used as an encoder, and N-dimensional feature vectors x of W time instants in a subsample ss 1 ,x 2 ,…,x W Sequentially inputting each unit of the first layer of LSTM, sequentially inputting the obtained W hidden vectors into each unit of the second layer of LSTM, and obtaining W hidden vectors h 1 ,h 2 ,…,h W ；
The semantic layer: taking a hidden vector h W As a low-dimensional semantic vector after encoding;
the decoding layer: two layers of LSTM are adopted as a decoder, and an implicit vector h is generated W Repeating W times and inputting each unit of the first layer of LSTM in sequence to obtain W hidden vectors, and inputting each unit of the second layer of LSTM in sequence to obtain W hidden vectors g 1 ,g 2 ,…,g W ；
The output layer: adopting a full connection layer to hide W vectors g 1 ,g 2 ,…,g W Conversion into a vector y conforming to the dimension ss of the subsamples 1 ,y 2 ,…,y W Vector y 1 ,y 2 ,…,y W As output data rss.
5. The method for detecting and diagnosing production abnormalities in an industrial system based on multidimensional sensed data as set forth in claim 4, wherein the mean square error of the output data rss and the subsample ss is used as a loss function in the training of the abnormality detection model, and the optimization iteration is performed in a gradient descent manner.
6. The method for detecting and diagnosing production anomalies of an industrial system based on multidimensional sensed data as recited in claim 1, wherein the sequentially calculating feature confidence corresponding to each of the N feature dimensions from real-time subsamples comprises:
Feature confidence is computed for feature k as follows:
Figure FDA0003590324800000021
in the formula, phi k For feature confidence of feature k, k is 1,2, …, N, CM S (x S ) Classification model CM trained for training subsets S corresponding to feature subsets that do not contain feature k S At subsample x S The output result is 0 or 1, the subsample x S For sample data extracted from the real-time subsamples that are identical to the features contained in the feature subset S, CM S∪{k} (x S∪{k} ) Classification model CM trained using training subsets corresponding to feature subsets S { k } containing feature k S∪{k} At subsample x S∪{k} The output result is 0 or 1, the subsample x S∪{k} The sample data extracted from the real-time subsamples and identical to the features contained in the feature subset S { k },
Figure FDA0003590324800000022
representing a feature subset S that does not contain feature k.
7. The method of claim 1, wherein the diagnosing abnormal features based on feature confidence comprises:
firstly, normalizing the confidence degrees of all the features by adopting a Sigmoid function to obtain a weight score, wherein the absolute value of the weight score indicates the influence value of the features on the final detection result, and the detection result is explained by utilizing a SHAP explanation model based on the influence value.
CN202210375778.5A 2022-04-11 2022-04-11 Industrial system production abnormity detection and diagnosis method based on multi-dimensional sensing data Pending CN114841250A (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202210375778.5A CN114841250A (en) 2022-04-11 2022-04-11 Industrial system production abnormity detection and diagnosis method based on multi-dimensional sensing data
PCT/CN2022/135714 WO2023197617A1 (en) 2022-04-11 2022-11-30 Method for detecting and diagnosing production abnormality of industrial system on basis of multi-dimensional sensing data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210375778.5A CN114841250A (en) 2022-04-11 2022-04-11 Industrial system production abnormity detection and diagnosis method based on multi-dimensional sensing data

Publications (1)

Publication Number Publication Date
CN114841250A true CN114841250A (en) 2022-08-02

Family

ID=82563882

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210375778.5A Pending CN114841250A (en) 2022-04-11 2022-04-11 Industrial system production abnormity detection and diagnosis method based on multi-dimensional sensing data

Country Status (2)

Country Link
CN (1) CN114841250A (en)
WO (1) WO2023197617A1 (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115358348A (en) * 2022-10-19 2022-11-18 成都数之联科技股份有限公司 Vehicle straight-through rate influence characteristic determination method, device, equipment and medium
CN115474108A (en) * 2022-09-16 2022-12-13 南京品淳通信科技有限公司 Event monitoring system and method based on edge calculation
CN116430817A (en) * 2023-04-26 2023-07-14 同心县启胜新能源科技有限公司 Data acquisition processing method and system applied to photovoltaic module production system
WO2023197617A1 (en) * 2022-04-11 2023-10-19 浙江工业大学 Method for detecting and diagnosing production abnormality of industrial system on basis of multi-dimensional sensing data

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2012137934A (en) * 2010-12-27 2012-07-19 Hitachi Ltd Abnormality detection/diagnostic method, abnormality detection/diagnostic system, abnormality detection/diagnostic program and company asset management/facility asset management system
CN113051822A (en) * 2021-03-25 2021-06-29 浙江工业大学 Industrial system anomaly detection method based on graph attention network and LSTM automatic coding model
CN113255733B (en) * 2021-04-29 2023-04-07 西安交通大学 Unsupervised anomaly detection method under multi-modal data loss
CN114065613B (en) * 2021-10-27 2022-12-09 中国华能集团清洁能源技术研究院有限公司 Multi-working-condition process industrial fault detection and diagnosis method based on deep migration learning
CN114090396B (en) * 2022-01-24 2022-06-14 华南理工大学 Cloud environment multi-index unsupervised anomaly detection and root cause analysis method
CN114841250A (en) * 2022-04-11 2022-08-02 浙江工业大学 Industrial system production abnormity detection and diagnosis method based on multi-dimensional sensing data

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023197617A1 (en) * 2022-04-11 2023-10-19 浙江工业大学 Method for detecting and diagnosing production abnormality of industrial system on basis of multi-dimensional sensing data
CN115474108A (en) * 2022-09-16 2022-12-13 南京品淳通信科技有限公司 Event monitoring system and method based on edge calculation
CN115474108B (en) * 2022-09-16 2023-09-22 南京品淳通信科技有限公司 Event monitoring system and method based on edge calculation
CN115358348A (en) * 2022-10-19 2022-11-18 成都数之联科技股份有限公司 Vehicle straight-through rate influence characteristic determination method, device, equipment and medium
CN116430817A (en) * 2023-04-26 2023-07-14 同心县启胜新能源科技有限公司 Data acquisition processing method and system applied to photovoltaic module production system
CN116430817B (en) * 2023-04-26 2023-09-29 同心县启胜新能源科技有限公司 Data acquisition processing method and system applied to photovoltaic module production system

Also Published As

Publication number Publication date
WO2023197617A1 (en) 2023-10-19

Similar Documents

Publication Publication Date Title
CN114841250A (en) Industrial system production abnormity detection and diagnosis method based on multi-dimensional sensing data
CN111504676B (en) Equipment fault diagnosis method, device and system based on multi-source monitoring data fusion
CN107657250B (en) Bearing fault detection and positioning method and detection and positioning model implementation system and method
CN109146246B (en) Fault detection method based on automatic encoder and Bayesian network
CN111428789A (en) Network traffic anomaly detection method based on deep learning
CN111967486A (en) Complex equipment fault diagnosis method based on multi-sensor fusion
CN111460728A (en) Method and device for predicting residual life of industrial equipment, storage medium and equipment
CN113051822A (en) Industrial system anomaly detection method based on graph attention network and LSTM automatic coding model
Azzalini et al. A minimally supervised approach based on variational autoencoders for anomaly detection in autonomous robots
CN115293280A (en) Power equipment system anomaly detection method based on space-time feature segmentation reconstruction
CN115484102A (en) Industrial control system-oriented anomaly detection system and method
CN117290800B (en) Timing sequence anomaly detection method and system based on hypergraph attention network
CN112685476A (en) Periodic multivariate time series anomaly detection method and system
CN114363195A (en) Network flow prediction early warning method for time and spectrum residual convolution network
CN116150901A (en) Rolling bearing residual life prediction method based on attention-enhanced time-frequency converter
CN115903741A (en) Data anomaly detection method for industrial control system
CN112115922A (en) Rotary machine fault diagnosis method for enhancing deep feature learning
CN116680643A (en) Abnormality detection method combining deep contrast learning and density clustering
CN114707577A (en) Anomaly detection method and system based on self-confrontation variational self-encoder
CN116977708B (en) Bearing intelligent diagnosis method and system based on self-adaptive aggregation visual view
Wu et al. Multiscale reduction clustering of vibration signals for unsupervised diagnosis of machine faults
CN111858343A (en) Countermeasure sample generation method based on attack capability
CN116821619A (en) Time sequence anomaly detection method based on multi-element time sequence relation learning
CN116680639A (en) Deep-learning-based anomaly detection method for sensor data of deep-sea submersible
CN116167008A (en) Abnormal positioning method for internet of things sensing cloud data center based on data enhancement

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