A Distributed Detection System Based on Active Learning

Abstract

The present invention provides a distributed detection system based on active learning, including a distributed sensor cluster, an edge computing server cluster, a core computing server cluster, and an active learning server; the distributed sensor cluster is used for signal acquisition of the industrial site to provide various types of raw data for the system; the edge computing server cluster is used for signal fusion and feature extraction of the fused signals, and performs edge computing on the system; the core computing server cluster is used for synchronous fault prediction and anomaly classification of labeled data; the active learning server is used for machine annotation of unlabeled data. By implementing the embodiments of the present invention, it is possible to efficiently and accurately achieve fault prediction, anomaly monitoring, and classification of the industrial site.

Description

Technical Field

[0001] The present invention relates to the technical field of fault prediction, and in particular, to a distributed detection system based on active learning. Background Art

[0002] In the actual production process, the environment is usually relatively complex. The complex system and environment require us to control the system well. A fault or anomaly in a certain link will have a serious impact on the entire system. Therefore, it is necessary to perform fault prediction and anomaly monitoring on the production process system. At the same time, it is also necessary to classify the abnormal situations so as to perform abnormal processing according to the classification results. The traditional method mainly relies on manual to monitor the system for anomalies. For large and complex non-linear systems, it is difficult for manual empirical judgment to monitor the system efficiently and accurately, and it is impossible to realize fault prediction, anomaly monitoring and classification of the system, which is likely to cause unnecessary production operation failures or property losses. Summary of the Invention

[0003] The present invention provides a distributed detection system, a system, an electronic device and a storage medium based on active learning to solve the above technical problems, and can efficiently and accurately realize fault prediction, anomaly monitoring and classification of the industrial field.

[0004] The present invention provides a distributed detection system based on active learning, including a distributed sensor cluster, an edge computing server cluster, a core computing server cluster and an active learning server;

[0005] The distributed sensor cluster is used to collect the basic data generated in the industrial field and send the basic data to the edge computing server cluster;

[0006] The edge computing server cluster is used to perform wavelet transform on the received basic data by using a preset signal encryption algorithm, and then perform wavelet inverse transform by using a preset signal decryption algorithm to obtain the fused time series data and output it;

[0007] The core computing server cluster is used to input the fused time series data with labels into the trained fault prediction model to output the fault prediction result. At the same time, the fused time series data with labels is input into the trained anomaly classification model to output the anomaly classification result;

[0008] The active learning server is used to perform data annotation on the fused time series data based on the training conditions of the fault prediction model and the anomaly classification model.

[0009] In one embodiment, the basic data includes the signal data collected by all sensors respectively arranged on the production workshop, production equipment and production parts.

[0010] In one embodiment, the distributed sensor cluster determines the sampling frequency and sampling period of each sensor based on Shannon's theorem, so that each sensor performs signal acquisition according to the determined sampling frequency and sampling period.

[0011] In one embodiment, the edge computing server cluster includes:

[0012] A data fusion module for fusing the basic data to obtain fused signal data;

[0013] A signal transformation module for performing wavelet transform on the fused signal data by using a preset signal encryption algorithm to obtain the amplitude spectrum and phase spectrum of the signal, using the Monte Carlo random algorithm to evenly divide the amplitude spectrum to obtain a new amplitude spectrum, and then performing an element-wise product operation on the phase spectrum and the new amplitude spectrum to obtain a new frequency-domain signal;

[0014] A signal inverse transformation module for performing wavelet inverse transform on the new frequency-domain signal by using a preset signal decryption algorithm to obtain the fused time-series data and output it.

[0015] In one embodiment, the core computing server cluster includes:

[0016] A fault prediction module for inputting the labeled fused time-series data into a trained fault prediction model for autoregressive prediction to obtain a fault prediction result output by the fault prediction model;

[0017] An anomaly classification module for converting the labeled fused time-series data into an image, and then inputting the image into a trained anomaly classification model for classification to obtain an anomaly classification result output by the anomaly classification model.

[0018] In one embodiment, the core computing server cluster further includes:

[0019] A first model training module for training the fault prediction model according to pre-acquired first sample data and the corresponding fault monitoring results of the first sample data to obtain a trained fault prediction model;

[0020] A second model training model for training the anomaly classification model according to pre-acquired second sample data and the corresponding anomaly monitoring results of the second sample data to obtain a trained anomaly classification model.

[0021] In one embodiment, the active learning server includes:

[0022] A correlation operation module for performing a correlation operation based on the training conditions of the fault prediction model and the anomaly classification model to obtain a correlation evaluation result;

[0023] A data annotation module, configured to perform data annotation on the fused time-series data according to the correlation evaluation result.

[0024] In one embodiment, the active learning server further includes an annotation correction module, configured to correct the annotation of the fused time-series data according to the obtained manual correction instruction.

[0025] In one embodiment, the signal data includes the environmental data of the production workshop, the operation data of the production equipment, and the induction data of the production parts.

[0026] In one embodiment, the anomaly classification model is an anomaly classification model based on an attention mechanism.

[0027] The distributed detection system based on active learning provided by the present invention collects signals from the industrial site through a distributed sensor cluster, providing various types of raw data for the system; fuses various signals through an edge computing server cluster, extracts features from the fused signals, and performs edge computing on the system; synchronously performs fault prediction and anomaly classification on the labeled data through a core computing server cluster; performs machine annotation on the unlabeled data through an active learning server; thus, the embodiments of the present invention can efficiently and accurately implement fault prediction, anomaly monitoring, and classification of the industrial site. Description of the Drawings

[0028] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0029] Figure 1 is a schematic structural diagram of a distributed detection system based on active learning provided by an embodiment of the present invention;

[0030] Figure 2 is a schematic diagram of the signal preprocessing process of an edge computing server cluster provided by an embodiment of the present invention;

[0031] Figure 3 is a schematic diagram of the process of fault prediction and anomaly classification provided by an embodiment of the present invention;

[0032] Figure 4 is a schematic flowchart of signal preprocessing performed by an edge computing server cluster provided by an embodiment of the present invention;

[0033] Figure 5It is a schematic flow diagram of fault prediction and anomaly classification provided by an embodiment of the present invention. Detailed implementation manners

[0034] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0035] As Figure 1 shown, an embodiment of the present invention provides a distributed detection system based on active learning, including a distributed sensor cluster, an edge computing server cluster, a core computing server cluster, and an active learning server;

[0036] The distributed sensor cluster is configured to collect basic data generated in an industrial field and send the basic data to the edge computing server cluster;

[0037] The edge computing server cluster is configured to perform wavelet transform on the received basic data by using a preset signal encryption algorithm, and then perform inverse wavelet transform by using a preset signal decryption algorithm to obtain fused time-series data and output it;

[0038] The core computing server cluster is configured to input the fused time-series data with labels into a trained fault prediction model to output a fault prediction result. At the same time, the fused time-series data with labels is input into a trained anomaly classification model to output an anomaly classification result;

[0039] The active learning server is configured to perform data annotation on the fused time-series data based on the training conditions of the fault prediction model and the anomaly classification model.

[0040] It should be noted that the embodiment of the present invention adopts a distributed detection technology based on active learning: multiple types of sensors are used to collect production process data, the fused data is processed by the edge computing server cluster to generate multiple types of signal data, the core computing server cluster performs fault prediction and anomaly classification on multiple types of labeled data. At the same time, the active learning server performs machine annotation and manual correction on the original data, reducing the manpower and material costs required for data annotation. At the same time, accurate labels effectively improve the fault prediction ability and anomaly classification ability of the system.

[0041] In the embodiments of the present invention, a distributed detection system based on active learning is realized through sensor technology, deep learning technology, edge computing technology, and active learning technology. The sensor provides the signal acquisition ability for the system, the edge computing provides the signal fusion and data preprocessing ability for the system, and the active learning technology facilitates the data annotation of the system. While ensuring that the system obtains accurate labels, it reduces the waste of human and computing resources and continuously improves its fault prediction and anomaly classification capabilities.

[0042] In one embodiment, the basic data includes signal data collected by all sensors respectively arranged on the production workshop, production equipment, and production parts. Further, the signal data includes the environmental data of the production workshop, the operation data of the production equipment, and the induction data of the production parts.

[0043] It should be noted that the distributed sensor cluster is configured to: receive various different signals generated in the system, including signals such as pressure, temperature, humidity, light, production rhythm, and operation data of each node. For the above signal data, the sensor performs omnidirectional sampling on it. For different positions and different operating conditions, different sampling periods are adopted to obtain comprehensive data. The data obtained by sampling is sent to the edge computing server cluster.

[0044] In one embodiment, the distributed sensor cluster determines the sampling frequency and sampling period of each sensor based on Shannon's theorem, so that each sensor performs signal acquisition according to the determined sampling frequency and sampling period.

[0045] It should be noted that various different signals generated in the system include all signals on the production workshop, production equipment, and production parts, including physical signals such as sound, light, and electricity, various different forms of signals, and signals in normal working states, shutdown states, and abnormal occurrence situations are all collected. Further, this link determines the sampling frequency and period according to Shannon's theorem to better obtain signal data.

[0046] The distributed detection system based on active learning provided by the embodiments of the present invention can obtain various signals in the industrial field through sensors, so as to more comprehensively and accurately perform fault prediction, anomaly monitoring, and classification on the production system.

[0047] In one embodiment, the edge computing server cluster includes:

[0048] A data fusion module, configured to perform information fusion on the basic data to obtain fusion signal data;

[0049] A signal transformation module, configured to perform wavelet transform on the fused signal data by using a preset signal encryption algorithm to obtain the amplitude spectrum and phase spectrum of the signal, average-divide the amplitude spectrum by using a Monte Carlo random algorithm to obtain a new amplitude spectrum, and then perform an element-wise product operation on the phase spectrum and the new amplitude spectrum to obtain a new frequency-domain signal;

[0050] A signal inverse transformation module, configured to perform wavelet inverse transform on the new frequency-domain signal by using a preset signal decryption algorithm to obtain the fused time-series data and output it.

[0051] In an embodiment of the present invention, the edge computing server cluster is configured to: receive various signals sampled by a distributed sensor cluster, perform information fusion on the various signals to obtain a fused signal, perform edge computing analysis on the fused signal by using edge computing technology, convert the signal by performing wavelet transform on the signal through a signal encryption algorithm, extract features, and then convert the signal back by performing wavelet inverse transform on the signal through a signal decryption algorithm to obtain a processed signal, and input it into the core computing server cluster.

[0052] It should be noted that the edge computing server cluster may include two edge computing servers and an encryption-decryption program. The first edge computing server (including the signal transformation module) is mainly used to implement the signal encryption algorithm, and the second edge computing server (including the signal inverse transformation module) is mainly used to implement the signal decryption algorithm. Through the encryption and decryption processes, the time-domain and frequency-domain features of the signal are obtained.

[0053] Specifically, the main steps of the encryption-decryption process may include:

[0054] (1) Input various types of original signals sampled by the original sensor cluster into the edge computing server cluster for signal preprocessing;

[0055] (2) Perform information fusion on various signals to obtain a fused signal;

[0056] (3) Edge computing server 1 uses a signal encryption program for the original signal to obtain the amplitude spectrum and phase spectrum of the signal through wavelet transform;

[0057] (4) Average-divide the amplitude spectrum obtained in step (3) by using a Monte Carlo random algorithm to obtain a series of new amplitude spectra with equal lengths;

[0058] (5) Perform an element-wise product operation on the phase spectrum obtained in step (3) and the new amplitude spectrum obtained in step (4) to obtain a new frequency-domain signal;

[0059] (6) For the new frequency-domain signal obtained in step (5), perform inverse wavelet transform through the signal decryption program in the edge computing server 2 to obtain the processed time-series signal;

[0060] (7) Input the processed time-series signal obtained in step (6) into the core computing server cluster for fault diagnosis and anomaly classification.

[0061] The distributed detection system based on active learning provided by the embodiments of the present invention extracts features from the fusion signal through edge computing technology, which helps to more accurately realize the fault prediction, anomaly monitoring, and classification of the system.

[0062] In one embodiment, the core computing server cluster includes:

[0063] A fault prediction module, configured to input the labeled fusion time-series data into a trained fault prediction model for autoregressive prediction to obtain the fault prediction result output by the fault prediction model;

[0064] An anomaly classification module, configured to convert the labeled fusion time-series data into an image, and then input the image into a trained anomaly classification model for classification to obtain the anomaly classification result output by the anomaly classification model. Further, the anomaly classification model is an anomaly classification model based on the attention mechanism.

[0065] In the embodiments of the present invention, the core computing server cluster is configured to: receive the signal processed by the edge computing server cluster, perform further feature extraction on the labeled data, and first train the fault prediction model and the anomaly classification model. After the models are trained, perform autoregression on the time-series data in the fault prediction program to obtain the future fault prediction situation; in the anomaly classification program, first randomly segment the two-dimensional time-series signal into images, and then perform anomaly detection and classification on the images.

[0066] It should be noted that the core computing server cluster may include three core computing servers, and a time-series to image conversion program, a fault prediction program, and an anomaly classification program. The first core computing server is mainly used to execute the time-series to image conversion program, the second core computing server is mainly used to execute the fault prediction program, and the third core computing server is mainly used to execute the anomaly classification program to perform fault prediction and anomaly classification on the system. In other embodiments, the core computing server cluster may also include two core computing servers, one for implementing the function of the fault prediction module and one for implementing the function of the anomaly classification module.

[0067] Specifically, the main steps of the process of fault prediction and anomaly classification may include:

[0068] (1) Input the time series signals processed by the edge computing server cluster into the core computing server cluster for fault diagnosis and anomaly classification;

[0069] (2) The core computing server 1 uses a time series-image conversion program to convert the time series data into images;

[0070] (3) Input the images obtained in step (2) into the core computing server 3, and use an anomaly classification program to classify the images;

[0071] (4) Input the time series data in step (1) into the core computing server 2 for autoregressive prediction to achieve fault prediction.

[0072] In one embodiment, the core computing server cluster further includes:

[0073] A first model training module, configured to train the fault prediction model according to the pre-obtained first sample data and the corresponding fault monitoring results of the first sample data to obtain a trained fault prediction model;

[0074] A second model training model, configured to train the anomaly classification model according to the pre-obtained second sample data and the corresponding anomaly monitoring results of the second sample data to obtain a trained anomaly classification model.

[0075] It should be noted that before the core computing server cluster performs fault prediction and anomaly classification, it is necessary to train the corresponding models according to the sample data and the actual monitoring results of the sample data.

[0076] In one embodiment, the active learning server includes:

[0077] A correlation operation module, configured to perform a correlation operation based on the training status of the fault prediction model and the anomaly classification model to obtain a correlation evaluation result;

[0078] A data annotation module, configured to perform data annotation on the fused time series data according to the correlation evaluation result.

[0079] In one embodiment, the active learning server further includes an annotation correction module, configured to correct the annotation of the fused time series data according to the obtained manual correction instruction.

[0080] In the embodiment of the present invention, the active learning server is configured to: receive the signals processed by the edge computing server, perform a correlation evaluation according to the first-round training status, perform machine annotation and manual calibration on the unlabeled samples, so as to reduce the labor and material costs required for data annotation, and at the same time improve the fault prediction ability and anomaly classification ability of the system through accurate labels.

[0081] Specifically, the main steps of the active learning process may include:

[0082] (1) Perform a correlation operation based on the results obtained by the core computing server cluster;

[0083] (2) Use an active learning program to perform data annotation on the fused signals obtained by the edge computing server cluster according to the correlation operation results;

[0084] (3) Manually correct the data annotation results to obtain high-quality annotated data.

[0085] The distributed detection system based on active learning provided by the embodiments of the present invention effectively improves the accuracy of fault prediction, anomaly monitoring, and classification by using active learning technology to perform machine annotation and manual correction on the fused signals obtained by the edge computing server cluster, thereby obtaining accurate annotated data.

[0086] Based on the above solution, to better understand the distributed detection system based on active learning provided by the embodiments of the present invention, the following specific examples are listed for detailed description:

[0087] As Figure 1 shown, the schematic diagram of the manufacturing process anomaly detection system based on edge computing and continuous learning provided by the embodiments of the present invention includes: a distributed sensor cluster, an edge computing server cluster, a core computing server cluster, and an active learning server. First, the sensor cluster collects signals from the industrial site to provide various types of raw data for the system; then the edge computing server cluster fuses various signals, extracts features from the fused signals, and performs edge computing on the system; the core computing server cluster performs core computing on the system and synchronously performs fault prediction and anomaly classification on the labeled data; the active learning server performs machine annotation and manual repair on the unlabeled data.

[0088] As Figure 2 shown is the schematic diagram of the signal preprocessing process performed by the edge computing server cluster provided by the embodiments of the present invention, which includes: two edge computing servers, an encryption-decryption algorithm. Feature fusion and time-domain-frequency-domain-time-domain conversion are performed on the signals to extract deeper-level information.

[0089] As Figure 3The figure shows a schematic diagram of the fault prediction and anomaly classification process provided by an embodiment of the present invention, which includes: three core computing servers, one active learning server, a fault prediction program, an anomaly classification program, an active learning program, etc. The three core computing servers and one active learning server provide computing power for the system; the fault prediction program is used to predict faults for time series data; the anomaly classification program is used to classify and identify the faults occurring in the system; the active learning program is used to provide accurate annotation information for the system.

[0090] As Figure 4 The figure shows a flowchart of the signal preprocessing process of the edge computing server cluster provided by an embodiment of the present invention. The main steps may include:

[0091] Step 401: Input various types of original signals sampled by the original sensor cluster into the edge computing server cluster for signal preprocessing. For example, within time T, input a two-dimensional signal (matrix form of the two-dimensional signal: X=(x1,...,x N )T∈R N×T , N = m×n) into the edge computing server cluster for data processing;

[0092] Step 402: Perform information fusion on various types of signals to obtain the fused signal;

[0093] Step 403: Edge computing server 1 uses a signal encryption program for the original signal, and obtains the amplitude spectrum and phase spectrum of the signal through the wavelet transform algorithm (wavelet formula: where the scale b controls the stretching of the wavelet function, and the offset a controls the offset of the wavelet function);

[0094] Step 404: Use the Monte Carlo random algorithm to evenly divide the amplitude spectrum obtained in step 403 to obtain a series of new amplitude spectra with equal lengths;

[0095] Step 405: Perform an element-wise product operation on the phase spectrum obtained in step 403 and the new amplitude spectrum obtained in step 404 to obtain a new frequency-domain signal;

[0096] Step 406: For the new frequency-domain signal obtained in step 404, perform an inverse wavelet transform (inverse wavelet transform formula where b is the scale that controls the stretching of the wavelet function, and a is the offset that controls the offset of the wavelet function) through the signal decryption program in edge computing server 2 to obtain the processed time series signal (expression is );

[0097] Step 407: Input the processed time series signals obtained in Step 405 into the core computing server cluster for fault diagnosis and anomaly classification.

[0098] As Figure 5 shown in the flowchart of the fault prediction and anomaly classification process provided by the embodiment of the present invention, the main steps may include:

[0099] Step 501: Input the time series signals processed by the edge computing server cluster into the core computing server cluster for fault diagnosis and anomaly classification;

[0100] Step 502: The core computing server 1 uses a time series-image conversion program to convert time series data into images;

[0101] Step 503: Input the images obtained in Step 502 into the core computing server 3, and use an attention-based anomaly classification program to classify the images;

[0102] Step 504: Train the time series data in Step 501 using an autoregressive prediction algorithm to obtain a fault prediction result;

[0103] Step 505: Perform a correlation analysis based on the results obtained in Steps 503 and 504;

[0104] Step 506: Use an active learning program to perform data annotation on the fusion signals obtained by the edge computing server cluster, and use edge sampling to select those sample data that are extremely easy to be judged into two categories. Calculate according to the formula where and respectively represent the cases where the model predicts the most likely class and the second most likely class for the data x;

[0105] Step 507: Manually correct the data annotation results to obtain high-quality annotation data;

[0106] Step 508: Input the labeled data into the core computing server and go back to 501 for continuous training and updating.

[0107] Compared with the prior art, the embodiment of the present invention has the following advantages and characteristics:

[0108] (1) By setting up a distributed sensor cluster, comprehensively collect signals from the production process system, and obtain detailed production process data, which is beneficial to accurately and comprehensively predict faults, monitor and classify abnormal situations in the production process;

[0109] (2) Information fusion of various signals is carried out by the edge computing server cluster, fusing various different information, while preprocessing the signals on-site to reduce waste of computing resources, extracting deeper signal features, and obtaining data suitable for machine processing;

[0110] (3) Through the core computing server cluster, feature extraction is performed on the data and the model is trained to identify and classify abnormal situations of the system. At the same time, the autoregressive algorithm is used for fault prediction to predict and handle faults in advance;

[0111] (4) The active learning server performs machine annotation and manual calibration on unlabeled samples, thereby reducing the labor and material costs required for data annotation. At the same time, the fault prediction ability and abnormal classification ability of the system are effectively improved through the obtained accurate labels.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A distributed detection system based on active learning, characterized in that, It includes a distributed sensor cluster, an edge computing server cluster, a core computing server cluster, and an active learning server; The distributed sensor cluster is used to collect the basic data generated in the industrial field and send the basic data to the edge computing server cluster; The edge computing server cluster is used to perform wavelet transform on the received basic data by using a preset signal encryption algorithm, and then perform wavelet inverse transform by using a preset signal decryption algorithm to obtain the fused time-series data and output it; The core computing server cluster is used to input the fused time-series data with annotations into the trained fault prediction model to output the fault prediction result. At the same time, the fused time-series data with annotations is input into the trained anomaly classification model to output the anomaly classification result; The core computing server cluster includes: The fault prediction module is used to input the fused time-series data with annotations into the trained fault prediction model for autoregressive prediction to obtain the fault prediction result output by the fault prediction model; The anomaly classification module is used to convert the fused time-series data with annotations into an image, and then input the image into the trained anomaly classification model for classification to obtain the anomaly classification result output by the anomaly classification model; The active learning server is used to perform data annotation on the fused time-series data based on the training conditions of the fault prediction model and the anomaly classification model; The active learning server includes: The correlation operation module is used to perform correlation operation based on the training conditions of the fault prediction model and the anomaly classification model to obtain the correlation evaluation result; The data annotation module is used to perform data annotation on the fused time-series data according to the correlation evaluation result; The annotation correction module is used to correct the annotation of the fused time-series data according to the obtained manual correction instruction.

2. The distributed detection system based on active learning according to claim 1, wherein The basic data includes the signal data collected by all sensors respectively arranged on the production workshop, production equipment, and production parts.

3. The distributed detection system based on active learning according to claim 2, wherein The distributed sensor cluster determines the sampling frequency and sampling period of each sensor based on Shannon's theorem, so that each sensor performs signal acquisition according to the determined sampling frequency and sampling period.

4. The distributed detection system based on active learning according to claim 1, wherein The edge computing server cluster includes: The data fusion module is used to perform information fusion on the basic data to obtain fused signal data; The signal transformation module is used to perform wavelet transform on the fused signal data by using a preset signal encryption algorithm to obtain the amplitude spectrum and phase spectrum of the signal, use the Monte Carlo random algorithm to evenly divide the amplitude spectrum to obtain a new amplitude spectrum, and then perform element-wise product operation on the phase spectrum and the new amplitude spectrum to obtain a new frequency-domain signal; The signal inverse transformation module is used to perform wavelet inverse transform on the new frequency-domain signal by using a preset signal decryption algorithm to obtain the fused time-series data and output it.

5. The distributed detection system based on active learning according to claim 1, characterized in that The core computing server cluster further includes: The first model training module is used to train the fault prediction model according to the pre-obtained first sample data and the corresponding fault monitoring results to obtain the trained fault prediction model; The second model training model is used to train the anomaly classification model according to the pre-acquired second sample data and the corresponding anomaly monitoring results of the second sample data to obtain a trained anomaly classification model.

6. The distributed detection system based on active learning according to claim 2, wherein The signal data includes the environmental data of the production workshop, the operation data of the production equipment, and the induction data of the production parts.

7. The distributed detection system based on active learning according to claim 1, characterized in that The anomaly classification model is an anomaly classification model based on the attention mechanism.

Images