CN113671917B - Detection method, system and equipment for abnormal state of multi-modal industrial process - Google Patents
Detection method, system and equipment for abnormal state of multi-modal industrial process Download PDFInfo
- Publication number
- CN113671917B CN113671917B CN202110955882.7A CN202110955882A CN113671917B CN 113671917 B CN113671917 B CN 113671917B CN 202110955882 A CN202110955882 A CN 202110955882A CN 113671917 B CN113671917 B CN 113671917B
- Authority
- CN
- China
- Prior art keywords
- matrix
- reconstruction
- data
- module
- error
- 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
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 96
- 238000001514 detection method Methods 0.000 title claims abstract description 89
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 74
- 239000011159 matrix material Substances 0.000 claims abstract description 150
- 238000000034 method Methods 0.000 claims abstract description 79
- 238000012544 monitoring process Methods 0.000 claims abstract description 78
- 230000008569 process Effects 0.000 claims abstract description 35
- 230000003416 augmentation Effects 0.000 claims abstract description 22
- 238000007781 pre-processing Methods 0.000 claims abstract description 19
- 238000001914 filtration Methods 0.000 claims abstract description 10
- 230000006870 function Effects 0.000 claims description 22
- 238000012549 training Methods 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 14
- 238000003860 storage Methods 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 10
- 238000004140 cleaning Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 4
- 230000002547 anomalous effect Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 230000003190 augmentative effect Effects 0.000 claims description 2
- 230000011218 segmentation Effects 0.000 claims description 2
- 239000000126 substance Substances 0.000 claims description 2
- 230000005856 abnormality Effects 0.000 abstract description 5
- 238000009776 industrial production Methods 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 13
- 238000009826 distribution Methods 0.000 description 12
- 238000004590 computer program Methods 0.000 description 8
- 238000012360 testing method Methods 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 4
- 239000010410 layer Substances 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000005406 washing Methods 0.000 description 3
- 230000007423 decrease Effects 0.000 description 2
- 238000009795 derivation Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000011049 filling Methods 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 239000002356 single layer Substances 0.000 description 2
- 238000013179 statistical model Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 238000012371 Aseptic Filling Methods 0.000 description 1
- 239000002253 acid Substances 0.000 description 1
- 239000003513 alkali Substances 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 239000002585 base Substances 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010924 continuous production Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 238000004540 process dynamic Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012859 sterile filling Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
- G05B19/41875—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by quality surveillance of production
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/31—From computer integrated manufacturing till monitoring
- G05B2219/31359—Object oriented model for fault, quality control
-
- 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)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Pure & Applied Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Algebra (AREA)
- Computing Systems (AREA)
- Manufacturing & Machinery (AREA)
- Quality & Reliability (AREA)
- Automation & Control Theory (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention belongs to the field of process monitoring, and particularly relates to a method, a system and equipment for detecting abnormal states of a multi-modal industrial process, aiming at solving the problem of insufficient detection precision of abnormal production states reflected by abnormal change trends of monitored multi-dimensional variables in a complex industrial production process. The method comprises acquiring state monitoring data; preprocessing the state monitoring data to obtain preprocessed data; segmenting the preprocessed data, and constructing an augmentation matrix used for describing the state of the production process at the moment t as a first matrix; based on the error between the matrixes, the set abnormality index s is calculated et 、s bt The following detection results; filtering the obtained detection result; and if the value corresponding to the filtered detection result is larger than any one of the two pre-acquired alarm thresholds, early warning is carried out. The invention improves the detection precision of abnormal state detection in the multi-mode industrial process.
Description
Technical Field
The invention belongs to the field of process monitoring, and particularly relates to a method, a system and equipment for detecting abnormal states of a multi-mode industrial process.
Background
Modern industrial production lines are increasingly complex, and if abnormal phenomena in the production process cannot be timely found, diagnosed and eliminated, the abnormal phenomena can cause damage to equipment, reduction in product quality and even serious production accidents and casualties. Today, a large number of sensors are used for monitoring the production process, and sufficient monitoring data can be collected from the production process. The data can be used for constructing a model based on data driving, and the real-time monitoring of the production process and the automatic detection of abnormal states are realized. The related research has great research significance and application value.
Most of the existing process monitoring algorithms are directed to single-mode production processes. The single mode production process has a single operation mode, and the data distribution is relatively simple, and can generally represent the data distribution of a single peak. One of the most typical representatives is the gaussian distribution. However, data collected from real-world production processes often fail to satisfy the above assumptions. Influenced by market demands, the production process can be switched among multiple sets of parameter settings; the production process can be switched among different procedures according to the process requirements, so that the monitoring data are in multimodal distribution. At this time, the single-modal process anomaly detection algorithm based on the unimodal distribution assumption and the independent sample assumption is often difficult to meet the task requirement. Therefore, it is necessary to construct an abnormality detection method of the multimodal process.
Although many methods have been proposed in recent years for multi-modal anomaly detection. Most methods analyze objects in a relatively single data format. In the production process, each operation mode usually corresponds to a group of fixed parameter settings, and the acquired data have significant statistical characteristics. Most current anomaly detection algorithms based on independent sample assumptions and based on statistical features of data can detect these anomalous states with anomalous data distributions. However, the actual production process is often a continuous process, and most statistical models cannot effectively reflect abnormal variation trend in time sequence. Furthermore, in addition to switching between sets of operating modes for a real-world production process, the monitored variables within a single operating modality may also be time-varying. Since an industrial process may be cyclically switched among a plurality of processes, the monitored data in the same operation mode may show various changing trends. However, it is difficult to efficiently model these features using only statistical models.
Therefore, there is a need in the art for an efficient modeling method for multi-modal industrial processes with multiple timing characteristics, and further enabling the detection of abnormal states in multi-modal processes.
Disclosure of Invention
In order to solve the above-mentioned problems in the prior art, that is, to solve the problem of insufficient detection accuracy of abnormal production states reflected by abnormal variation trends of monitored multidimensional variables in a complex industrial process including a plurality of operation modalities, a first aspect of the present invention proposes a method for detecting abnormal states of a multimodal industrial process, the method comprising:
s10, collecting the state monitoring data of the production line in the production process through a plurality of groups of sensors;
s20, preprocessing the state monitoring data to obtain preprocessed data;
s30, adopting sliding window strategy, self-defining window size n and sliding step length n s Segmenting the preprocessed data; after the division, constructing an augmentation matrix used for describing the state of the production process at the time t by utilizing the preprocessed data from the time t-n +1 to the time t, and using the augmentation matrix as a first matrix;
s40, reconstructing the first matrix as a second matrix by using the trained self-encoder model, and reconstructing a residual error between the first matrix and the second matrix by using the self-encoder model; summing the second matrix and the residual error to obtain a final reconstruction result, calculating a reconstruction error between the reconstruction result and the first matrix data, and calculating an abnormal index s of the preprocessed data in a set manner based on the reconstruction error et 、s bt The following detection results;
s50, combining the historical detection results, and filtering the detection results obtained in S40; if the value corresponding to the filtered detection result is larger than any one of two pre-acquired alarm thresholds, early warning is carried out;
the self-encoder model consists of three sub-modules, namely a basic reconstruction module, a residual reconstruction module and a reconstruction error prediction module;
the basic reconstruction module and the residual error reconstruction module are constructed based on a self-coding network; the basic reconstruction module is used for reconstructing the first matrix to obtain a second matrix; the residual error reconstruction module is used for reconstructing residual errors between the first matrix and the second matrix; the reconstruction error prediction module is constructed based on a feed-forward network and is used for estimating the reconstruction error of the self-encoder model.
In some preferred embodiments, "pre-processing the condition monitoring data" is performed by:
cleaning the state monitoring data: setting upper and lower thresholds of a sensor signal according to a set normal working range of a sensor, limiting sensor data which exceed an upper threshold limit, namely state monitoring data, to an upper threshold limit, and limiting state monitoring data which are lower than a lower threshold limit to a lower threshold limit;
carrying out normalization processing on the cleaned state monitoring data, wherein the normalization processing process comprises the following steps:
x n ＝S(x-L l )/(L h -L l )
where x is the raw sensor signal, i.e., the unnormalized state monitoring data, L h And L l Respectively, the upper threshold and the lower threshold of the state monitoring data, S is a scaling factor, and the value is [0,1 ]]X is n The data is the state monitoring data after normalization processing.
In some preferred embodiments, the self-encoder model is trained by:
a10, collecting historical state monitoring data of a production line in the production process through a plurality of groups of monitoring sensors to serve as training data;
a20, preprocessing the training data to obtain preprocessed data;
a30, adopting sliding window strategy, by self-defining window size n and sliding step length n s Dividing the preprocessed data into a plurality of data to be detected; after the division, state detection data from the time t-n +1 to the time t are used for constructing augmentation used for describing the state of the production process at the time tA matrix as a first matrix;
a40, reconstructing the first matrix as the second matrix through the basic reconstruction module of the self-encoder model, and reconstructing the residual error between the first matrix and the second matrix through the residual error reconstruction module; summing the second matrix and the residual error to obtain a final reconstruction result, and calculating a reconstruction error between the reconstruction result and the first matrix data through a reconstruction error prediction module;
a50, based on the historical state monitoring data in the first matrix, combining the historical state monitoring data in the second matrix, the reconstruction result of the residual between the first matrix and the second matrix, and the reconstruction error of the self-encoder model, calculating the loss values corresponding to the basic reconstruction module, the residual reconstruction module and the reconstruction error prediction module through a pre-constructed loss function, and updating parameters;
and A60, circularly executing the steps A10-A50 until a trained self-encoder model is obtained.
In some preferred embodiments, the base reconstruction module, the residual reconstruction module, and the reconstruction error prediction module have a loss function during training as follows:
wherein L is B ，L R And L P Respectively is a loss function of a basic reconstruction module, a residual reconstruction module and a reconstruction error prediction module, n is the length of a data sample, namely each augmentation matrix comprises data of m sensors at continuous n sampling moments,
historical state monitoring data X of time i by using basic reconstruction module i The result of the reconstruction of (a) is,
x for residual error reconstruction module to time i i And
residual error of
Reconstruction result of e Pi Is a pair of reconstruction error prediction modules X i Final reconstruction error of
As a result of the estimation of (a),
in some preferred embodiments, the set abnormality index s et 、s bt The calculation method comprises the following steps:
wherein e is et And e bt Respectively representing the historical state monitoring of the self-encoder model at the t momentStatistical errors and statistical offsets between the reconstructed sequence of test data and the input data,
is the model for t-i time data X t-i The result of the reconstruction of (a) is,
is the corresponding reconstruction error, s, predicted by the reconstruction error prediction module et And s bt The abnormal index of the set one dimension respectively represents the statistical error and the statistical deviation of the reconstructed matrix, mu 1 ，∑ 1 And mu 2 ，∑ 2 Is e et And e bt The corresponding mean vector and covariance matrix.
In some preferred embodiments, the alarm threshold T is set at two anomaly indicators e And T b The method is obtained by calculation by adopting a nuclear density estimation method, and the calculation method comprises the following steps:
wherein s is i Representing the abnormal score of the ith sample, N is the total number of samples participating in calculation, h is bandwidth, K (-) is a kernel function, a Gaussian kernel function is generally adopted, s represents the abnormal score, p(s) represents the probability density corresponding to the abnormal score s, alpha represents confidence coefficient, T (T) represents the probability density corresponding to the abnormal score s s And (alpha) is an alarm threshold value when the confidence coefficient is set to be alpha.
In a second aspect of the present invention, a system for detecting abnormal status of a multi-modal industrial process is provided, the system comprising: the device comprises a data acquisition module, a preprocessing module, a matrix construction module, a state detection module and a trigger alarm module;
the data acquisition module is configured to acquire state monitoring data of the production line in the production process through a plurality of groups of sensors;
the preprocessing module is configured to preprocess the state monitoring data to obtain preprocessed data;
the matrix building module is configured to adopt a sliding window strategy and customize the window size n and the sliding step length n s Segmenting the preprocessed data; after the division, constructing an augmentation matrix used for describing the state of the production process at the time t by utilizing the preprocessed data from the time t-n +1 to the time t, and using the augmentation matrix as a first matrix;
the state detection module is configured to reconstruct a first matrix as a second matrix by using the trained self-encoder model, and reconstruct a residual error between the first matrix and the second matrix by using the self-encoder model; summing the second matrix and the residual error to obtain a final reconstruction result, calculating a reconstruction error between the reconstruction result and the first matrix data, and calculating an abnormal index s of the preprocessed data in a set state based on the reconstruction error et 、s bt The following detection results;
the trigger alarm module is configured to perform filtering processing on the detection result acquired from the state detection module in combination with the historical detection result; if the value corresponding to the filtered detection result is larger than any one of two pre-acquired alarm thresholds, early warning is carried out;
the self-encoder model consists of three sub-modules, namely a basic reconstruction module, a residual reconstruction module and a reconstruction error prediction module;
the basic reconstruction module and the residual error reconstruction module are constructed based on a self-coding network; the basic reconstruction module is used for reconstructing the first matrix to obtain a second matrix; the residual error reconstruction module is used for reconstructing residual errors between the first matrix and the second matrix; the reconstruction error prediction module is constructed based on a feed-forward network and is used for estimating the reconstruction error of the self-encoder model.
In a third aspect of the invention, an apparatus is presented, at least one processor; and a memory communicatively coupled to at least one of the processors; wherein the memory stores instructions executable by the processor for execution by the processor to implement the claimed method for detecting an abnormal condition of a multimodal industrial process.
In a fourth aspect of the present invention, a computer-readable storage medium is provided, which stores computer instructions for being executed by the computer to implement the claimed method for detecting an abnormal state of a multi-modal industrial process.
The invention has the beneficial effects that:
the invention improves the detection precision of abnormal states in the multi-mode industrial process.
1) Based on the GRU double-self-coding network structure, one self-coding network learns the macroscopic trend of filling monitoring data, and the other self-coding network further learns the jagged detailed characteristics on the basis of the first self-coding network, so that the modeling of a multi-mode dynamic process is realized, the overall modeling precision of a self-coding model is improved, and compared with a single sequence-to-sequence network structure, the model can realize higher modeling precision in a smaller network scale.
2) According to the invention, two abnormal indexes are designed, the model can effectively detect the abnormality in the multi-mode production process based on the two indexes, and compared with the Mahalanobis distance index which is widely adopted at present, the two indexes can more effectively detect possible abnormal trends. With the development of the modern industrial production towards large scale and integration, the scale of the sensor network for monitoring the production process is gradually enlarged, and the method can effectively utilize mass monitoring data to carry out modeling and realize automatic anomaly detection on multi-path monitoring data; meanwhile, with the intelligent and personalized development of industrial production, the production process becomes more complex, the method can effectively detect the abnormal state of the industrial process with complex process dynamics, and has considerable application prospect.
Drawings
Other features, objects and advantages of the present application will become more apparent upon reading of the following detailed description of non-limiting embodiments thereof, made with reference to the accompanying drawings.
FIG. 1 is a flow diagram of a method for detecting an abnormal condition of a multimodal industrial process in accordance with one embodiment of the present invention;
FIG. 2 is a block diagram of a detection system for multimodal industrial process anomaly status according to one embodiment of the present invention;
FIG. 3 is a graphical illustration of the trend of historical status inspection data during normal operating conditions for a CIP process for an aseptic filling line balance tank in accordance with an embodiment of the present invention;
FIG. 4 is a diagram of the basic structure of a self-coding model according to one embodiment of the invention;
FIG. 5 is a block diagram of the basic reconstruction module and the residual reconstruction module of the self-coding model according to an embodiment of the present invention;
FIG. 6 is a reconstruction of a portion of normal state detection data according to an embodiment of the present invention;
FIG. 7 shows a comparison of fitting accuracy of the self-coding model to process data at different network scales according to an embodiment of the present invention;
FIG. 8 is a diagram illustrating the detection of two abnormal states in accordance with one embodiment of the present invention;
fig. 8 (a): a record of abnormal dynamically changing processes (normalized values);
fig. 8 (b): results of detection (S) of the data described in FIG. 8(a) by the model e ) The dotted data in the graph represents the actual detection result;
fig. 8 c): results of detection (S) of the data described in FIG. 8(a) by the model b )；
Fig. 8 (d): abnormal data migration and records of abnormal dynamic processes (normalized values);
fig. 8 (e): results of detection (S) of the data described in FIG. 8(b) by the model e )；
Fig. 8 (f): results of detection (S) of the data described in FIG. 8(b) by the model b )；
Fig. 9 is a schematic structural diagram of a computer system suitable for implementing an electronic device according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some, but not all embodiments of the present invention. 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.
The present application will be described in further detail with reference to the following drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the relevant invention and not restrictive of the invention. It should be noted that, for convenience of description, only the portions related to the related invention are shown in the drawings.
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict.
The invention relates to a detection method for abnormal states of a multi-modal industrial process, which comprises the following steps:
s10, acquiring state monitoring data of the production line in the production process through a plurality of groups of sensors;
s20, preprocessing the state monitoring data to obtain preprocessed data;
s30, adopting sliding window strategy, self-defining window size n and sliding step length n s Segmenting the preprocessed data; after the division, constructing an augmentation matrix used for describing the state of the production process at the time t by utilizing the preprocessed data from the time t-n + 1 to the time t, and using the augmentation matrix as a first matrix;
s40, reconstructing the first matrix as a second matrix by using the trained self-encoder model, and reconstructing a residual error between the first matrix and the second matrix by using the self-encoder model; summing the second matrix and the residual error to obtain a final reconstruction result, calculating a reconstruction error between the reconstruction result and the first matrix data, and calculating an abnormal index s of the preprocessed data in a set state based on the reconstruction error et 、s bt The following detection results;
s50, combining the historical detection results, and filtering the detection results obtained in S40; if the value corresponding to the filtered detection result is larger than any one of two pre-acquired alarm thresholds, early warning is carried out;
the self-encoder model consists of three sub-modules, namely a basic reconstruction module, a residual reconstruction module and a reconstruction error prediction module;
the basic reconstruction module and the residual error reconstruction module are constructed based on a self-coding network; the basic reconstruction module is used for reconstructing the first matrix to obtain a second matrix; the residual error reconstruction module is used for reconstructing residual errors between the first matrix and the second matrix; the reconstruction error prediction module is constructed based on a feed-forward network and is used for estimating the reconstruction error of the self-encoder model.
In order to more clearly describe the method for detecting an abnormal state of a multi-modal industrial process according to the present invention, the following description will discuss the steps in an embodiment of the method according to the present invention with reference to fig. 1.
In the following embodiments, a training process of the self-encoder model is described first, and then a process of obtaining a detection result of an abnormal state of a multi-modal industrial process by using a detection method of an abnormal state of a multi-modal industrial process is described in detail.
1. Training process for autoencoder model
A10, collecting historical state monitoring data of a production line in the production process through a plurality of groups of monitoring sensors to serve as training data;
in this embodiment, historical state monitoring data under normal operating conditions of the production line is collected. The invention is a data-driven anomaly detection method, which requires modeling of the normal production process; when the online application is carried out, the current abnormal state of the production line is measured according to the deviation degree of the state monitoring data and the normal working state of the production line. Therefore, the collected historical state monitoring data needs to cover all normal states as much as possible. Fig. 3 is historical state monitoring data of a CIP process of a balance tank of the sterile filling line under normal working conditions, and relates to data of 4-way sensors. In the figure, the process involves three kinds of cleaning processes in total: the three records of different value ranges are respectively corresponded to 'water washing', 'alkali washing' and 'acid washing'.
A20, preprocessing the training data to obtain preprocessed data;
in this embodiment, the preprocessing includes cleaning and normalizing the historical state monitoring data, which is specifically as follows:
a21, cleaning the state monitoring data: due to sensor disturbance, etc., the data collected by the sensor (status monitoring data) sometimes deviates significantly from the normal range, and the data needs to be corrected. And setting the upper and lower thresholds of each path of sensor signal by using the normal working range set by the sensor. Limiting data exceeding the upper threshold limit to the upper threshold limit; similarly, values below the lower threshold limit are limited to the lower threshold limit.
A22, performing normalization processing on the cleaned state monitoring data, wherein the normalization processing process is as follows:
x n ＝S(x-L l )/(L h -L l ) (1)
where x is the raw sensor signal, i.e., the unnormalized state monitoring data, L h And L l Respectively an upper threshold limit and a lower threshold limit of sensor data, S is a scaling factor, and the value is [0,1 ]]In the present invention, it is preferably set to 0.8, x n And monitoring data for the historical state after normalization processing.
A30, adopting sliding window strategy, by self-defining window size n and sliding step length n s Dividing the preprocessed data into a plurality of data to be detected; after the division, an augmentation matrix used for describing the state of the production process at the time t is constructed by using the state detection data from the time t-n + 1 to the time t and used as a first matrix;
condition detection data collected from industrial sites is often accompanied by varying degrees of disturbance. Due to noise, the data collected by the sensor network at the current moment may deviate from the actual production state. Based on this, in the present embodiment, a sliding window strategy is adopted for preprocessing data acquired in a20And (3) performing data segmentation: the state monitoring data of the equipment at the time t is X t (X t ＝[x t1 ，x t2 ，...，x tm ]，x t1 To x tm Representing the condition monitoring data collected by the m sensors at time t). And intercepting the data segment by adopting a sliding window strategy and a sliding window with the window length of n. Augmented matrix S constructed from t-n +1 time to t time t ＝[X t-n+1 ，X t-n+2 ，...，X t ](S t ∈R m×n ) The state of the production process at time t is described.
Adjacent data samples S t ，S t+s Time interval s (corresponding to the sliding step n) s ) Can be set according to actual requirements, in the embodiment, n s Set to 3.
A40, reconstructing the first matrix as the second matrix through the basic reconstruction module of the self-encoder model, and reconstructing the residual error between the first matrix and the second matrix through the residual error reconstruction module; summing the second matrix and the residual error to obtain a final reconstruction result, and calculating a reconstruction error between the reconstruction result and the first matrix data through a reconstruction error prediction module;
in this embodiment, the self-encoder model is composed of three sub-modules, i.e., a basic reconstruction module, a residual reconstruction module and a reconstruction error prediction module, and fig. 4 shows an overall structure of the self-encoder model. The basic reconstruction module and the residual reconstruction module are both self-coding networks, and the basic reconstruction module is used for reconstructing the first matrix S t To obtain a second matrix
The residual error reconstruction module is used for reconstructing a first matrix S t Second matrix output by the basic reconstruction module
The residual error between. Final reconstruction result from coding model
Based on the output of the reconstruction module
Residual from output of residual reconstruction module
And (4) summing. The reconstruction error prediction module is a feedforward network and is used for estimating the data X of the model at each sampling moment t Reconstruction error of
The data distribution under different modes may have large difference, and it is relatively difficult to adopt a single self-coding model while considering the characteristics of large difference of data distribution and variation range during modeling. The self-coding model described in fig. 4 distributes the modeling task to two network modules: the basic reconstruction module is used for modeling macroscopic data characteristics of the data; after the macroscopic features are subtracted, the original local features which are small in value range and not obvious compared with the macroscopic features can be recorded in the residual sequence more obviously
The residual reconstruction module is thus able to model these local features relatively easily. Compared with a single self-coding network, the structure shown in fig. 4 can achieve the effect of realizing more accurate modeling in a smaller network scale. Fig. 6 is a result of reconstruction of a portion of normal data from the coding model. As can be seen from the figure, the basic reconstruction module learns the macroscopic trend of the data, and the residual error reconstruction module further learns the jagged detailed characteristics on the basis of the data of the basic reconstruction module, so that the overall modeling precision of the model is improved. Fig. 7 shows the reconstruction accuracy of the basic reconstruction module and the entire model for process data at different network scales (specifically, different hidden layer scales), which can be obtained from the comparison result. The method has positive significance for modeling complex industrial processes under the condition of limited computing resources. A reconstruction error prediction module forThe final reconstruction error is predicted. By subtracting prediction errors under different modes, the problem of interference of model on data modeling precision difference of different modes on abnormal detection is relieved. The hidden layer dimension of the self-coding model should be smaller than the scale of the input augmentation matrix (i.e., n × m).
Because the model needs to model the dynamic characteristics of the process, preferably, both the basic reconstruction module and the residual reconstruction module use a sequence-to-sequence GRU network. Fig. 5 is a schematic diagram of the working principle of the network. To explain the working principle of the module more intuitively, fig. 5 sequentially expands the calculation flow of the adjacent n-time data in a single data sample according to time sequence. The G module in the figure is a GRU network (coding network); the "D" module is also another GRU network (decoding network). Encoding network forward computing input sequence [ X ] t-n+1 ，X t-n+2 ，...，X t ]State h at each time i i Decoding network using state h at time t t Stepwise backward derivation of S t . Wh + b in the figure is a single-layer feedforward network, and the effect of the single-layer feedforward network is to reversely deduce the process state of a decoding network at the moment i
Mapping to reconstructed data
The input and output of each module of the model are described as follows: the encoder portion input of the basis reconstruction module is S t The initial state input is 0, and the initial state of the decoder is the state of the encoder at the last time t
First reconstruction data
Based on a feed-forward network
And (6) directly calculating and obtaining. In subsequent derivation, the decoding network is selected fromSuccessive use state from time t to time t-n + 1
And reconstructing the result
And deducing the state and reconstruction result at the moment i. The coding network input of the residual error reconstruction module is
The input of the decoding network at the ith moment is
And
([·]representing the concatenation of data in the feature dimension), and the rest of the settings are the same as those of the basic reconstruction module. The input of the reconstruction error prediction module is
In the preferred embodiment, the analysis object is 4 temperature signals, the length of the data sample is 40, the input dimension of the model is 4, the number of network layers is 2, and the hidden layer dimension is 72.
A50, based on the historical state monitoring data in the first matrix, combining the historical state monitoring data in the second matrix, the reconstruction result of the residual between the first matrix and the second matrix, and the reconstruction error of the self-encoder model, calculating the loss values corresponding to the basic reconstruction module, the residual reconstruction module and the reconstruction error prediction module through a pre-constructed loss function, and updating parameters;
in the present embodiment, the history record under the normal working condition of the production line processed in step S20 is divided into two parts, namely a training set and a verification set: the training set is used for model training; the validation set data is used to validate the model effect and prevent over-fitting of the model. The three sub-module training all adopt root Mean Square Error (MSE), and the specific loss function is as follows:
wherein L is B ，L R And L P Respectively is a loss function of a basic reconstruction module, a residual reconstruction module and a reconstruction error prediction module, n is the length of a data sample, namely each augmentation matrix comprises data of m sensors at continuous n sampling moments,
historical state monitoring data X of time i by using basic reconstruction module i The result of the reconstruction of (a) is,
x for residual error reconstruction module to time i i And
residual error of
Reconstruction result of e Pi Is a pair of reconstruction error prediction modules X i Final reconstruction error of
The estimation result of (2).
And A60, circularly executing the steps A10-A50 until a trained self-encoder model is obtained.
In this embodiment, the auto encoder model is trained in a loop until a preset training precision is reached or a set training number is reached, so as to obtain the auto encoder model.
In addition, after the self-coding model is trained, the data of the verification set are calculated in two abnormal indexes s according to the reconstruction error et And s bt The following values, i.e. the detection results, are specifically:
wherein e is et And e bt Respectively representing the statistical error and the statistical offset between the reconstructed sequence of the historical state monitoring data at the time t and the input data from the encoder model,
is the model for t-i time data X t-i The result of the reconstruction of (a) is,
is the corresponding reconstruction error, s, predicted by the reconstruction error prediction module et And s bt The abnormal index of the set one dimension respectively represents the statistical error and the statistical deviation of the reconstructed matrix, mu 1 ，∑ 1 And mu 2 ，∑ 2 Is e et And e bt The corresponding mean vector and covariance matrix.
When the single model is adopted to reconstruct the multi-modal data, the reconstruction errors of the model to the data of different modes have difference, and the difference can influence the subsequent reconstruction errorsAlarm threshold estimation and anomaly detection procedures, therefore in calculating e et By subtracting the estimation error e p This difference is suppressed.
After the above-mentioned process, the distribution of the detection results basically belongs to unimodal distribution, and the corresponding alarm threshold can be estimated by adopting statistical method. Considering that data does not necessarily conform to common distribution such as Gaussian distribution, the alarm threshold T under two indexes is calculated by adopting a nuclear density estimation method e And T b . The specific calculation formula is as follows:
wherein s is i Representing the abnormal score of the ith sample, N is the total number of samples participating in calculation, h is bandwidth, K (·) is a kernel function, a Gaussian kernel function is generally adopted, s represents the abnormal score, p(s) represents the probability density corresponding to the abnormal score s, and alpha represents confidence. T is s And (alpha) is an alarm threshold value when the confidence coefficient is set to be alpha. When calculating specifically, s in the training data is respectively calculated et 、s bt Carry-in s i Calculating a threshold value s corresponding to T when the confidence coefficient reaches alpha e And T b . In the present invention, α is preferably set to 0.99.
2. Detection method for abnormal state of multi-modal industrial process
S10, acquiring state monitoring data of the production line in the production process through a plurality of groups of sensors;
in this embodiment, the state monitoring data of the production line during the production process is acquired as input data.
S20, preprocessing the state monitoring data to obtain preprocessed data;
s30, adopting sliding window strategy, self-defining window size n and sliding step length n s Dividing the preprocessed data into a plurality of data to be processedDetecting data; after the division, an augmentation matrix used for describing the state of the production process at the time t is constructed by using the state detection data from the time t-n + 1 to the time t and used as a first matrix;
in this embodiment, the status monitoring data is preprocessed to obtain preprocessed data, the preprocessed data is divided, and an augmentation matrix describing the status of the production process at time t is used as a component of the preprocessed data, and as the first matrix, the preprocessing and the division of the preprocessed data refer to the methods in steps a20 and a30, which are not described one by one here.
S40, reconstructing the first matrix as a second matrix by using the trained self-encoder model, and reconstructing a residual error between the first matrix and the second matrix by using the self-encoder model; summing the second matrix and the residual error to obtain a final reconstruction result, calculating a reconstruction error between the reconstruction result and the first matrix data, and calculating an abnormal index s of the preprocessed data in a set state based on the reconstruction error et 、s bt The following detection results;
in this embodiment, a first matrix is reconstructed by the trained self-coding model, and a residual error of the first matrix before and after reconstruction is calculated; and then reconstructing the residual matrix by adopting a self-encoder model. And combining the two reconstruction results to obtain final reconstruction data, and calculating the error between the final reconstruction data and the original data. Based on the error, the detection result of the filling monitoring data under the set abnormal index is calculated.
S50, combining the historical detection results, and filtering the detection results obtained in S40; if the value corresponding to the filtered detection result is larger than any one of two pre-acquired alarm thresholds, early warning is carried out;
because the data of field collection have noise of different degrees, for providing more stable testing result. In this embodiment, a filtering strategy is adopted, and a filtering algorithm is adopted to calculate a more stable detection result according to the current detection result and the adjacent l historical detection results. The specific calculation formula is as follows:
wherein the content of the first and second substances,
for the final anomaly score, l is the length of the filter window and c is the decay constant, preferably in the range of [0,1 ]]。
An abnormality is determined when
And
any one of the indices exceeds the corresponding threshold value (
Or S b ＞pT b ) An alarm is triggered.
Where p is a sensitivity constant greater than 1, the specific data value can be adjusted according to actual requirements. The sensitivity of the model decreases with increasing p, but the corresponding indication is more reliable. In the present invention, l is preferably set to 20, c is preferably set to 0.9, and p is preferably set to 1.0.
FIG. 8 shows the results of the present method for two types of tests that include an anomaly record. FIG. 8(a) depicts the transition from the first "water wash" to the "alkaline wash" anomaly, and FIGS. 8(b), (c) are the corresponding test results, with the dots representing the actual test results; fig. 8(d) includes the abnormal variation trend in the "alkaline cleaning" stage and the abnormal decrease and variation trend of the data after the "alkaline cleaning" is finished, and fig. 8(e) and (f) are the corresponding detection results. From the figure, combine S e And S b These abnormal states can be indicated well as a result of detection under the index.
As can be seen from the above description and result diagram, in the preferred embodiment of the present invention, the anomaly detection method for the multi-modal dynamic process mainly includes two main links of offline modeling and online monitoring: (1) modeling under a line: collecting historical monitoring data of a normal production process; carrying out data cleaning and data normalization; intercepting the augmentation matrix by adopting a sliding window strategy, and representing the current state by using adjacent data segments including current time data; and training a description model in the normal production process, and respectively calculating alarm thresholds corresponding to the two abnormal indexes according to the reconstruction result of the model on the normal data. When the deployment is carried out on line: constructing an augmentation matrix by using the data at the current moment and the adjacent historical records to describe the current state; reconstructing the augmentation matrix by using the trained model, and calculating detection values under two abnormal indexes according to reconstruction errors; and if the detection value exceeds the corresponding alarm threshold value, giving out an early warning.
A second embodiment of the present invention provides a system for detecting an abnormal state of a multi-modal industrial process, as shown in fig. 2, specifically including: the system comprises a data acquisition module 100, a preprocessing module 200, a matrix construction module 300, a state detection module 400 and a trigger alarm module 500;
the data acquisition module 100 is configured to acquire state monitoring data of the production line in the production process through a plurality of groups of sensors;
the preprocessing module 200 is configured to preprocess the state monitoring data to obtain preprocessed data;
the matrix building block 300 is configured to employ a sliding window strategy by customizing window size n and sliding step length n s Segmenting the preprocessed data; after the division, constructing an augmentation matrix used for describing the state of the production process at the time t by utilizing the preprocessed data from the time t-n + 1 to the time t, and using the augmentation matrix as a first matrix;
the state detection module 400 is configured to reconstruct a first matrix as a second matrix by using the trained self-encoder model, and reconstruct a residual error between the first matrix and the second matrix by using the self-encoder model; summing the second matrix and the residual error as a final reconstruction result, and calculating the reconstruction result and the first momentThe reconstruction error between the array data is calculated, and the abnormal index s of the preprocessed data in the set state is calculated based on the reconstruction error et 、s bt The following detection results;
the trigger alarm module 500 is configured to perform filtering processing on the detection result obtained in the state detection module in combination with the historical detection result; if the value corresponding to the filtered detection result is larger than any one of two pre-acquired alarm thresholds, early warning is carried out;
the self-encoder model consists of three sub-modules, namely a basic reconstruction module, a residual reconstruction module and a reconstruction error prediction module;
the basic reconstruction module and the residual error reconstruction module are constructed based on a self-coding network; the basic reconstruction module is used for reconstructing the first matrix to obtain a second matrix; the residual error reconstruction module is used for reconstructing residual errors between the first matrix and the second matrix; the reconstruction error prediction module is constructed based on a feed-forward network and is used for estimating the reconstruction error of the self-encoder model.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working process and related description of the system described above may refer to the corresponding process in the foregoing method embodiment, and details are not described herein again.
It should be noted that, the detection system for the abnormal state of the multi-modal industrial process provided in the foregoing embodiment is only illustrated by the division of the above functional modules, and in practical applications, the functions may be allocated to different functional modules according to needs, that is, the modules or steps in the embodiment of the present invention are further decomposed or combined, for example, the modules in the foregoing embodiment may be combined into one module, or may be further split into multiple sub-modules, so as to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not to be construed as unduly limiting the present invention.
An apparatus of a third embodiment of the invention, at least one processor; and a memory communicatively coupled to at least one of the processors; wherein the memory stores instructions executable by the processor for execution by the processor to implement the claimed method for detecting an abnormal condition of a multimodal industrial process.
A computer-readable storage medium of a fourth embodiment of the present invention stores computer instructions for execution by the computer to implement the claimed method for detecting an abnormal state of a multi-modal industrial process.
It is clear to those skilled in the art that, for convenience and brevity not described, the specific working processes and related descriptions of the above-described apparatuses and computer-readable storage media may refer to the corresponding processes in the foregoing method examples, and are not described herein again.
Referring now to FIG. 9, there is illustrated a block diagram of a computer system suitable for use as a server in implementing embodiments of the subject systems, methods, and devices. The server shown in fig. 9 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present application.
As shown in fig. 9, the computer system includes a Central Processing Unit (CPU) 901, which can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM) 902 or a program loaded from a storage portion 908 into a Random Access Memory (RAM) 903. In the RAM903, various programs and data necessary for system operation are also stored. The CPU901, ROM 902, and RAM903 are connected to each other via a bus 904. An Input/Output (I/O) interface 905 is also connected to bus 904.
The following components are connected to the I/O interface 905: an input portion 906 including a keyboard, a mouse, and the like; an output portion 907 including components such as a cathode ray tube, a liquid crystal display, and the like, and a speaker; a storage portion 908 including a hard disk and the like; and a communication section 909 including a network interface card such as a local area network card, a modem, or the like. The communication section 909 performs communication processing via a network such as the internet. A drive 910 is also connected to the I/O interface 905 as needed. A removable medium 911 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 910 as necessary, so that a computer program read out therefrom is mounted into the storage section 908 as necessary.
In particular, according to an embodiment of the present disclosure, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method illustrated in the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network via the communication section 909, and/or installed from the removable medium 911. The computer program, when executed by the CPU901, performs the above-described functions defined in the method of the present application. It should be noted that the computer readable medium mentioned above in the present application may be a computer readable signal medium or a computer readable storage medium or any combination of the two. The computer-readable storage medium may be, for example but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of the computer readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a RAM, a ROM, an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present application, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In this application, however, a computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, etc., or any suitable combination of the foregoing.
Computer program code for carrying out operations for aspects of the present application may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C + + or the like and conventional procedural programming languages, such as the C language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network or a wide area network, or the connection may be made to an external computer (for example, through the Internet using an Internet service provider).
The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
The terms "first," "second," and the like are used for distinguishing between similar elements and not necessarily for describing or implying a particular order or sequence.
The terms "comprises," "comprising," or any other similar term are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
So far, the technical solutions of the present invention have been described in connection with the preferred embodiments shown in the drawings, but it is easily understood by those skilled in the art that the scope of the present invention is obviously not limited to these specific embodiments. Equivalent changes or substitutions of related technical features can be made by those skilled in the art without departing from the principle of the invention, and the technical scheme after the changes or substitutions can fall into the protection scope of the invention.
Claims (9)
1. A method for detecting an abnormal state of a multi-modal industrial process, the method comprising the steps of:
s10, collecting the state monitoring data of the production line in the production process through a plurality of groups of sensors;
s20, preprocessing the state monitoring data to obtain preprocessed data;
s30, adopting sliding window strategy, self-defining window size n and sliding step length n s Dividing the preprocessed data into a plurality of data to be detected; after the division, an augmentation matrix used for describing the state of the production process at the time t is constructed by using the state detection data from the time t-n +1 to the time t and used as a first matrix;
s40, reconstructing the first matrix by using the trained self-encoder modelAs a second matrix, reconstructing a residual error between the first matrix and the second matrix by using the self-encoder model; summing the second matrix and the residual error to obtain a final reconstruction result, calculating a reconstruction error between the reconstruction result and the first matrix data, and calculating an abnormal index s of the preprocessed data in a set state based on the reconstruction error et 、s bt The following detection results;
s50, combining the historical detection results, and filtering the detection results obtained in S40; if the value corresponding to the filtered detection result is larger than any one of two pre-acquired alarm thresholds, early warning is carried out;
the self-encoder model consists of three sub-modules, namely a basic reconstruction module, a residual reconstruction module and a reconstruction error prediction module;
the basic reconstruction module and the residual error reconstruction module are constructed based on a self-coding network; the basic reconstruction module is used for reconstructing the first matrix to obtain a second matrix; the residual error reconstruction module is used for reconstructing residual errors between the first matrix and the second matrix; the reconstruction error prediction module is constructed based on a feed-forward network and is used for estimating the reconstruction error of the self-encoder model.
2. The method for detecting abnormal conditions of multimodal industrial processes according to claim 1, wherein the condition monitoring data is preprocessed by:
cleaning the state monitoring data: setting upper and lower thresholds of a sensor signal according to a set normal working range of a sensor, limiting sensor data which exceed an upper threshold limit, namely state monitoring data, to an upper threshold limit, and limiting state monitoring data which are lower than a lower threshold limit to a lower threshold limit;
carrying out normalization processing on the cleaned state monitoring data, wherein the normalization processing process comprises the following steps:
x n ＝S(x-L l )/(L h -L l )
where x is the raw sensor signal, i.e., the unnormalized state monitoring data, L h And L l Respectively, the upper threshold and the lower threshold of the state monitoring data, S is a scaling factor, and the value is [0,1 ]]X is n The data is the state monitoring data after normalization processing.
3. The method for detecting the abnormal state of the multi-modal industrial process according to claim 1, wherein the self-encoder model is trained by:
a10, collecting historical state monitoring data of a production line in the production process through a plurality of groups of monitoring sensors to serve as training data;
a20, preprocessing the training data to obtain preprocessed data;
a30, adopting sliding window strategy, by self-defining window size n and sliding step length n s Dividing the preprocessed data into a plurality of data to be detected; after the division, an augmentation matrix used for describing the state of the production process at the time t is constructed by using state detection data from the time t-n +1 to the time t and used as a first matrix;
a40, reconstructing the first matrix as the second matrix through the basic reconstruction module of the self-encoder model, and reconstructing the residual error between the first matrix and the second matrix through the residual error reconstruction module; summing the second matrix and the residual error to obtain a final reconstruction result, and calculating a reconstruction error between the reconstruction result and the first matrix data through a reconstruction error prediction module;
a50, based on the historical state monitoring data in the first matrix, combining the historical state monitoring data in the second matrix, the reconstruction result of the residual between the first matrix and the second matrix, and the reconstruction error of the self-encoder model, calculating the loss values corresponding to the basic reconstruction module, the residual reconstruction module and the reconstruction error prediction module through a pre-constructed loss function, and updating parameters;
and A60, circularly executing the steps A10-A50 until a trained self-encoder model is obtained.
4. The method of claim 3, wherein the basic reconstruction module, the residual reconstruction module and the reconstruction error prediction module have a loss function corresponding to:
wherein L is B ，L R And I P Respectively is a loss function of a basic reconstruction module, a residual reconstruction module and a reconstruction error prediction module, n is the length of a data sample, namely each augmentation matrix comprises data of m sensors at continuous n sampling moments,
historical state monitoring data X of time i by using basic reconstruction module i The result of the reconstruction of (a) is,
x for residual error reconstruction module to time i i And
residual error of
Reconstruction result of e Pi Is a pair of reconstruction error prediction modules X i Final reconstruction error of
As a result of the estimation of (a),
5. the method for detecting the abnormal state of the multimodal industrial process as claimed in claim 4, wherein the set abnormal index s et 、s bt The calculation method comprises the following steps:
wherein e is et And e bt Respectively representing the statistical error and the statistical offset between the reconstructed sequence of the historical state monitoring data at the time t and the input data from the encoder model,
is the model for t-i time data X t-i The result of the reconstruction of (a) is,
is the corresponding reconstruction error, s, predicted by the reconstruction error prediction module et And s bt The abnormal index of the set one dimension respectively represents the statistical error and the statistical deviation of the reconstructed matrix, mu 1 ，Σ 1 And mu 2 ，Σ 2 Is e et And e bt The corresponding mean vector and covariance matrix.
6. Method for the detection of anomalous states in multimodal industrial processes according to claim 1, characterised in that said alarm threshold T is lower for two anomaly indicators e And T b The method is obtained by calculation by adopting a nuclear density estimation method, and the calculation method comprises the following steps:
wherein s is i Representing the abnormal score of the ith sample, N being the total number of samples participating in calculation, h being bandwidth, K (-) being a kernel function, adopting a Gaussian kernel function, s representing the abnormal score, p(s) representing the probability density corresponding to the abnormal score s, alpha representing confidence, T s And (alpha) is an alarm threshold value when the confidence coefficient is set to be alpha.
7. A system for detecting an abnormal condition of a multimodal industrial process, the system comprising: the device comprises a data acquisition module, a preprocessing module, a matrix construction module, a state detection module and a trigger alarm module;
the data acquisition module is configured to acquire state monitoring data of the production line in the production process through a plurality of groups of sensors;
the preprocessing module is configured to preprocess the state monitoring data to obtain preprocessed data;
the matrix building module is configured to adopt a sliding window strategy and customize the window size n and the sliding step length n s Segmenting the preprocessed data; after segmentation, preprocessing data from the time t-n +1 to the time t are utilized to construct a description for the time tAn augmented matrix of states of the production process as a first matrix;
the state detection module is configured to reconstruct a first matrix as a second matrix by using the trained self-encoder model, and reconstruct a residual error between the first matrix and the second matrix by using the self-encoder model; summing the second matrix and the residual error to obtain a final reconstruction result, calculating a reconstruction error between the reconstruction result and the first matrix data, and calculating an abnormal index s of the preprocessed data in a set state based on the reconstruction error et 、s bt The following detection results;
the trigger alarm module is configured to perform filtering processing on the detection result acquired from the state detection module in combination with the historical detection result; if the value corresponding to the filtered detection result is larger than any one of two pre-acquired alarm thresholds, early warning is carried out;
the self-encoder model consists of three sub-modules, namely a basic reconstruction module, a residual reconstruction module and a reconstruction error prediction module;
the basic reconstruction module and the residual error reconstruction module are constructed based on a self-coding network; the basic reconstruction module is used for reconstructing the first matrix to obtain a second matrix; the residual error reconstruction module is used for reconstructing residual errors between the first matrix and the second matrix; the reconstruction error prediction module is constructed based on a feed-forward network and is used for estimating the reconstruction error of the self-encoder model.
8. A detection device for an abnormal state of a multimodal industrial process, comprising:
at least one processor; and
a memory communicatively coupled to at least one of the processors; wherein the content of the first and second substances,
the memory stores instructions executable by the processor to implement the method for detecting an abnormal condition of a multimodal industrial process as recited in any one of claims 1-6.
9. A computer-readable storage medium storing computer instructions for execution by the computer to implement the method for detecting an abnormal state of a multimodal industrial process of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110955882.7A CN113671917B (en) | 2021-08-19 | 2021-08-19 | Detection method, system and equipment for abnormal state of multi-modal industrial process |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110955882.7A CN113671917B (en) | 2021-08-19 | 2021-08-19 | Detection method, system and equipment for abnormal state of multi-modal industrial process |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113671917A CN113671917A (en) | 2021-11-19 |
| CN113671917B true CN113671917B (en) | 2022-08-02 |
Family
ID=78544141
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110955882.7A Active CN113671917B (en) | 2021-08-19 | 2021-08-19 | Detection method, system and equipment for abnormal state of multi-modal industrial process |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113671917B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114218197A (en) * | 2021-12-17 | 2022-03-22 | 上海繁易信息科技股份有限公司 | Object-oriented industrial Internet of things data modeling method and system |
| CN114519696B (en) * | 2021-12-31 | 2022-11-29 | 扬州盛强薄膜材料有限公司 | PVC heat shrinkage film detection method and system based on optical intelligence |
| EP4220321A1 (en) * | 2022-01-26 | 2023-08-02 | Siemens Aktiengesellschaft | Computer-implemented method for detecting deviations in a manufacturing process |
| CN114676593B (en) * | 2022-04-18 | 2023-05-23 | 无锡物联网创新中心有限公司 | Abnormality detection method and related device for textile equipment |
| WO2023220931A1 (en) * | 2022-05-17 | 2023-11-23 | Applied Materials, Inc. | Analysis of multi-run cyclic processing procedures |
| CN117746323B (en) * | 2023-12-19 | 2024-06-04 | 国网湖北省电力有限公司黄龙滩水力发电厂 | Hydropower station auxiliary equipment fault detection method based on space-time state diagram |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689203A (en) * | 2019-09-30 | 2020-01-14 | 国网山东省电力公司电力科学研究院 | Self-encoder-based primary fan fault early warning method for thermal power plant |
| CN111669373A (en) * | 2020-05-25 | 2020-09-15 | 山东理工大学 | Network anomaly detection method and system based on space-time convolutional network and topology perception |
| CN111720296A (en) * | 2020-06-11 | 2020-09-29 | 浙江浙能技术研究院有限公司 | Steam feed pump equipment abnormal state early warning method based on self-encoder |
| CN112101400A (en) * | 2019-12-19 | 2020-12-18 | 国网江西省电力有限公司电力科学研究院 | Industrial control system abnormality detection method, equipment, server and storage medium |
| CN112101636A (en) * | 2020-08-26 | 2020-12-18 | 北京航空航天大学 | GRU and GARCH-based satellite long-period variance-variance degradation prediction evaluation method |
| CN112257918A (en) * | 2020-10-19 | 2021-01-22 | 中国科学院自动化研究所 | Traffic flow prediction method based on circulating neural network with embedded attention mechanism |
| CN112257263A (en) * | 2020-10-22 | 2021-01-22 | 中国科学院自动化研究所 | Equipment residual life prediction system based on self-attention mechanism |
| CN112329865A (en) * | 2020-11-09 | 2021-02-05 | 平安科技（深圳）有限公司 | Data anomaly identification method and device based on self-encoder and computer equipment |
| CN112667808A (en) * | 2020-12-23 | 2021-04-16 | 沈阳新松机器人自动化股份有限公司 | BERT model-based relationship extraction method and system |
| WO2021098657A1 (en) * | 2019-11-18 | 2021-05-27 | 中国科学院深圳先进技术研究院 | Video detection method and apparatus, terminal device, and readable storage medium |
| CN113255835A (en) * | 2021-06-28 | 2021-08-13 | 国能大渡河大数据服务有限公司 | Hydropower station pump equipment anomaly detection method |
| CN113255733A (en) * | 2021-04-29 | 2021-08-13 | 西安交通大学 | Unsupervised anomaly detection method under multi-modal data loss |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR102061810B1 (en) * | 2018-03-23 | 2020-01-02 | 단국대학교 산학협력단 | System and Method for Processing Multi type Sensor Signal Based on Multi modal Deep Learning |
| US11410048B2 (en) * | 2019-05-17 | 2022-08-09 | Honda Motor Co., Ltd. | Systems and methods for anomalous event detection |
-
2021
- 2021-08-19 CN CN202110955882.7A patent/CN113671917B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689203A (en) * | 2019-09-30 | 2020-01-14 | 国网山东省电力公司电力科学研究院 | Self-encoder-based primary fan fault early warning method for thermal power plant |
| WO2021098657A1 (en) * | 2019-11-18 | 2021-05-27 | 中国科学院深圳先进技术研究院 | Video detection method and apparatus, terminal device, and readable storage medium |
| CN112101400A (en) * | 2019-12-19 | 2020-12-18 | 国网江西省电力有限公司电力科学研究院 | Industrial control system abnormality detection method, equipment, server and storage medium |
| CN111669373A (en) * | 2020-05-25 | 2020-09-15 | 山东理工大学 | Network anomaly detection method and system based on space-time convolutional network and topology perception |
| CN111720296A (en) * | 2020-06-11 | 2020-09-29 | 浙江浙能技术研究院有限公司 | Steam feed pump equipment abnormal state early warning method based on self-encoder |
| CN112101636A (en) * | 2020-08-26 | 2020-12-18 | 北京航空航天大学 | GRU and GARCH-based satellite long-period variance-variance degradation prediction evaluation method |
| CN112257918A (en) * | 2020-10-19 | 2021-01-22 | 中国科学院自动化研究所 | Traffic flow prediction method based on circulating neural network with embedded attention mechanism |
| CN112257263A (en) * | 2020-10-22 | 2021-01-22 | 中国科学院自动化研究所 | Equipment residual life prediction system based on self-attention mechanism |
| CN112329865A (en) * | 2020-11-09 | 2021-02-05 | 平安科技（深圳）有限公司 | Data anomaly identification method and device based on self-encoder and computer equipment |
| CN112667808A (en) * | 2020-12-23 | 2021-04-16 | 沈阳新松机器人自动化股份有限公司 | BERT model-based relationship extraction method and system |
| CN113255733A (en) * | 2021-04-29 | 2021-08-13 | 西安交通大学 | Unsupervised anomaly detection method under multi-modal data loss |
| CN113255835A (en) * | 2021-06-28 | 2021-08-13 | 国能大渡河大数据服务有限公司 | Hydropower station pump equipment anomaly detection method |
Non-Patent Citations (3)
| Title |
|---|
| Bidirectional GRU Based Autoencoder for Dimensionality Reduction in Hyperspectral Images;Shivam Pande，Biplab Banerjee;《2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS》;20210716;第2731-2734页 * |
| 基于自编码器和门限循环单元神经网络的滚动轴承退化趋势预测;王鹏等;《振动与冲击》;20200930;第39卷(第17期);第106-111页 * |
| 徐歆尧,徐德.基于GRU 网络的工业过程监测数据的缺失数据填补.《 2020中国自动化大会》.2020, * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113671917A (en) | 2021-11-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113671917B (en) | Detection method, system and equipment for abnormal state of multi-modal industrial process | |
| US11334407B2 (en) | Abnormality detection system, abnormality detection method, abnormality detection program, and method for generating learned model | |
| CN111914883B (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
| CN111340238A (en) | Fault diagnosis method, device, equipment and storage medium of industrial system | |
| US8347144B2 (en) | False alarm mitigation | |
| US10852357B2 (en) | System and method for UPS battery monitoring and data analysis | |
| CN109086876B (en) | Method and device for detecting running state of equipment, computer equipment and storage medium | |
| CN112414694B (en) | Equipment multistage abnormal state identification method and device based on multivariate state estimation technology | |
| Russo et al. | Anomaly detection using deep autoencoders for in-situ wastewater systems monitoring data | |
| CN114298443B (en) | Industrial equipment predictive maintenance method and device based on health state index and electronic equipment | |
| CN111611294B (en) | Star sensor data anomaly detection method | |
| US12007745B2 (en) | Apparatus for predicting equipment damage | |
| CN110375983B (en) | Valve fault real-time diagnosis system and method based on time series analysis | |
| CN112581719B (en) | Semiconductor packaging process early warning method and device based on time sequence generation countermeasure network | |
| CN111612756B (en) | Coronary artery specificity calcification detection method and device | |
| CN115903741A (en) | Data anomaly detection method for industrial control system | |
| CN112487356A (en) | Structural health monitoring data enhancement method | |
| CN112416662A (en) | Multi-time series data anomaly detection method and device | |
| CN116304909A (en) | Abnormality detection model training method, fault scene positioning method and device | |
| CN115169430A (en) | Cloud network end resource multidimensional time sequence anomaly detection method based on multi-scale decoding | |
| CN115307943A (en) | Method and system for detecting abnormal operation state of rotating mechanical equipment under multiple working conditions | |
| KR20220132824A (en) | Distribution facility condition monitoring system and method | |
| CN116610996A (en) | Cable state detection method and device and computer equipment | |
| CN114676593B (en) | Abnormality detection method and related device for textile equipment | |
| KR20230075150A (en) | Method for managing system health |
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 |