Manufacturing process abnormality detection method, electronic device, storage medium, and program product

Abstract

The application provides a manufacturing process anomaly detection method, an electronic device, a storage medium and a program product. The method comprises: acquiring signal data generated by a manufacturing process; performing image synthesis processing based on the signal data to obtain a synthesized image; inputting the synthesized image into a continuously learnable anomaly detection model to perform anomaly condition detection and anomaly classification, and obtaining an anomaly condition and an anomaly type output by the anomaly detection model. The anomaly detection model has a continuous learning capability. When a new anomaly condition occurs, the anomaly detection model can increase the detection and recognition capability of the new anomaly condition without losing the old anomaly detection and recognition capability of the anomaly detection model, thereby improving the anomaly detection performance of the anomaly detection model and ultimately improving the anomaly detection performance of the manufacturing process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of anomaly detection, and in particular to a manufacturing process anomaly detection method, an electronic device, a storage medium and a program product. BACKGROUND

[0002] The manufacturing process is complex in production and processing environment, and the complex system and environment require good control of the system. If a link fails or is abnormal, it will have a serious impact on the entire system. Therefore, it is necessary to detect anomalies in the manufacturing process to repair the system in advance and reduce unnecessary production operation failures or property losses according to the anomaly detection results.

[0003] At present, the anomaly detection of the system is performed by manpower. For large and complex nonlinear systems, the experience-based judgment of manpower cannot meet the requirements and is very wasteful in terms of labor cost. Therefore, artificial intelligence technology is used to replace manpower for anomaly detection. However, the existing deep learning method needs a large amount of training data for model training, which is very wasteful in terms of computing resources, resulting in ineffective anomaly detection. At the same time, when a new abnormal situation occurs, the model originally trained cannot detect the new abnormal situation, and the model needs to be retrained based on the new abnormal situation, resulting in the loss of the original old anomaly detection capability of the retrained model.

[0004] In summary, how to improve the anomaly detection performance of the manufacturing process is a problem to be solved at present. SUMMARY

[0005] The present application provides a manufacturing process anomaly detection method, an electronic device, a storage medium and a program product to solve the defects of the prior art that a large amount of training data is needed for model training and the model loses the original old anomaly detection capability, and to achieve high-performance anomaly detection.

[0006] The present application provides a manufacturing process anomaly detection method, comprising:

[0007] Obtaining signal data generated by a manufacturing process;

[0008] Performing image synthesis processing based on the signal data to obtain a synthesized image;

[0009] Inputting the synthesized image into a continuously learnable anomaly detection model to perform anomaly situation detection and anomaly classification, and obtaining an anomaly situation and an anomaly type output by the anomaly detection model.

[0010] According to the manufacturing process anomaly detection method provided by the present application, the image synthesis processing based on the signal data to obtain a synthesized image comprises:

[0011] Signal pre-processing is performed on the signal data to obtain a pre-processed time sequence signal.

[0012] Based on the time sequence signal, image synthesis processing is performed to obtain a synthesized image.

[0013] According to the manufacturing process anomaly detection method provided by the application, the signal pre-processing performed on the signal data to obtain a pre-processed time sequence signal comprises:

[0014] Frequency domain conversion is performed on the signal data to obtain an amplitude spectrum and a phase spectrum corresponding to the signal data.

[0015] Random shuffling processing is performed on the amplitude spectrum to obtain a new amplitude spectrum.

[0016] Element multiplication operation is performed on the new amplitude spectrum and the phase spectrum to obtain a frequency domain signal.

[0017] Time domain conversion is performed on the frequency domain signal to obtain a time sequence signal.

[0018] According to the manufacturing process anomaly detection method provided by the application, the signal pre-processing performed on the signal data to obtain a pre-processed time sequence signal comprises:

[0019] The time sequence signal is obtained by performing signal pre-processing on the signal data by the edge computing server cluster.

[0020] According to the manufacturing process anomaly detection method provided by the application, the image synthesis processing based on the time sequence signal to obtain a synthesized image comprises:

[0021] Random segmentation processing is performed on the time sequence signal to obtain a plurality of time sequence segments.

[0022] Correlation discrimination processing is performed on the plurality of time sequence segments to obtain correlation data.

[0023] Based on the correlation data, the plurality of time sequence segments are integrated according to correlation to obtain a synthesized sequence.

[0024] Based on the synthesized sequence, image synthesis processing is performed to obtain a synthesized image.

[0025] According to the manufacturing process anomaly detection method provided by the application, the method further comprises:

[0026] Obtaining a signal data training sample of a new abnormal situation;

[0027] Based on the signal data training sample, image synthesis processing is performed to obtain a new abnormal image.

[0028] Train the anomaly detection model based on the new abnormal image to obtain an updated anomaly detection model.

[0029] According to the manufacturing process anomaly detection method provided by the application, the anomaly detection model is trained based on the new abnormal image to obtain an updated anomaly detection model, which comprises the following steps:

[0030] Selecting a preset proportion of training sample sets from the original signal data training sample set;

[0031] Obtaining an image sample set corresponding to the training sample set;

[0032] Training the anomaly detection model based on the image sample set and the new abnormal image to obtain an updated anomaly detection model.

[0033] The application further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the manufacturing process anomaly detection method according to any one of the above.

[0034] The application further provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program is executable on a processor to implement the steps of the manufacturing process anomaly detection method according to any one of the above.

[0035] The application further provides a computer program product comprising a computer program, wherein the computer program is executable on a processor to implement the steps of the manufacturing process anomaly detection method according to any one of the above.

[0036] The manufacturing process anomaly detection method, the electronic device, the storage medium and the program product provided by the application obtain signal data generated in a manufacturing process, then perform image synthesis processing based on the signal data to obtain a synthesized image, and finally input the synthesized image into an anomaly detection model capable of continuous learning to perform anomaly detection and anomaly classification, thereby obtaining an anomaly situation and an anomaly type output by the anomaly detection model. The application automatically detects anomalies in the manufacturing process by using an anomaly detection model, that is, the manufacturing process is detected by using artificial intelligence technology, thereby improving the intelligence of the manufacturing process anomaly detection. Meanwhile, the anomaly detection model has the ability of continuous learning, and when a new anomaly situation occurs, the anomaly detection model can increase the detection and recognition ability of the new anomaly situation without losing the old anomaly detection and recognition ability of the anomaly detection model, thereby improving the anomaly detection performance of the anomaly detection model and finally improving the anomaly detection performance of the manufacturing process. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to make the technical solutions in the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those of ordinary skill in the art without any creative work on the basis of the drawings.

[0038] Figure 1 One of the flowcharts of the manufacturing process anomaly detection method provided by the present application;

[0039] Figure 2 The second flowchart of the manufacturing process anomaly detection method provided by the present application;

[0040] Figure 3 The signal preprocessing schematic diagram of the edge computing server cluster related to the embodiments of the present application;

[0041] Figure 4 The signal preprocessing schematic diagram of the center server cluster related to the embodiments of the present application;

[0042] Figure 5 The third flowchart of the manufacturing process anomaly detection method provided by the present application;

[0043] Figure 6 An example of the physical structure schematic diagram of an electronic device is shown. DETAILED DESCRIPTION

[0044] In order to make the technical solutions in the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those of ordinary skill in the art without any creative work on the basis of the drawings.

[0045] Figure 1 One of the flowcharts of the manufacturing process anomaly detection method provided by the present application, as shown in Figure 1 The manufacturing process anomaly detection method provided by the present application comprises:

[0046] Step 110, acquiring signal data generated by a manufacturing process;

[0047] In this embodiment, the signal data generated by the manufacturing process is a time series signal (time domain signal). The signal data can include at least one of the following: temperature, humidity, pressure, running data of each node, and running data generated by each device, etc. Correspondingly, the above signal data can be collected by sensor devices such as temperature sensors, humidity sensors, pressure sensors, etc.

[0048] In specific embodiments, signal data transmitted by the sensor cluster is received, the signal data being collected by the sensor cluster from the manufacturing process. The sensor cluster includes a plurality of sensors to collect a plurality of categories of signal data.

[0049] It can be understood that the sensor cluster is established to collect signals from all aspects of the manufacturing process, so as to obtain more detailed signal data, i.e., production process data, thereby improving the accuracy of manufacturing process anomaly detection, and ultimately improving the performance of manufacturing process anomaly detection.

[0050] In an embodiment, the signal data generated by the manufacturing process includes signals on industrial system or engineering system equipment. For example, vibration signals or frequency signals generated by each mechanical bearing part.

[0051] It should be noted that signal data in normal operation state and signal data in abnormal operation state are both acquired. Specifically, signals in normal operation state and signals in abnormal operation state are both sampled to obtain signal data.

[0052] In an embodiment, a sampling frequency or a sampling period is determined based on the Shannon theorem; and signals generated by the manufacturing process are sampled based on the sampling frequency or the sampling period to obtain signal data.

[0053] In step 120, image synthesis processing is performed based on the signal data to obtain a synthesized image.

[0054] In this embodiment, the signal data is converted into a synthesized image for better subsequent anomaly detection based on the synthesized image.

[0055] In specific embodiments, the manufacturing process anomaly detection method is applied to a central server cluster, and step 120 includes:

[0056] Signal data transmitted by the sensor cluster is received, and image synthesis processing is performed based on the signal data to obtain a synthesized image. The central server cluster can include one server device or a plurality of server devices.

[0057] In some embodiments, before the image synthesis processing, the signal data is preprocessed to obtain preprocessed time series signals, and the image synthesis processing is performed based on the preprocessed time series signals to obtain a synthesized image. The preprocessed time series signals are data suitable for machine processing. The specific execution process can be referred to the second embodiment described below, which will not be repeated here.

[0058] In an embodiment, image synthesis processing is directly performed based on the signal data or the time series signals after preprocessing to obtain a synthesized image. The synthesized image can include one or more.

[0059] In another embodiment, the signal data or the pre-processed time sequence signal is randomly segmented to obtain a plurality of time sequence segments; the plurality of time sequence segments are subjected to correlation discrimination processing to obtain correlation data; based on the correlation data, the plurality of time sequence segments are integrated according to correlation to obtain a synthetic sequence; and based on the synthetic sequence, image synthesis processing is performed to obtain a synthetic image. The specific execution process can refer to the second embodiment described below, which will not be described here.

[0060] In step 130, the synthetic image is input into the continuously learnable anomaly detection model to perform anomaly condition detection and anomaly classification, and the anomaly condition and its anomaly type output by the anomaly detection model are obtained.

[0061] In this embodiment, the anomaly detection model is a deep learning model, and the anomaly detection model based on deep learning construction and training can realize the functions of anomaly condition detection and anomaly classification.

[0062] At the same time, the anomaly detection model is a continuously learnable (continuous learning) model, and continuous learning provides incremental anomaly detection capability for the model, ensuring that the model continuously expands its detection and recognition capability for other anomalies while not losing its basic anomaly detection capability. That is, the anomaly detection model is trained again through continuous learning to obtain a new model with enhanced discrimination capability, so that the model retains the original anomaly detection performance while acquiring more anomaly detection and discrimination capability.

[0063] It can be understood that the anomaly detection model with continuous learning capability has stronger learning capability, and for new abnormal conditions, the anomaly detection model can increase the detection and recognition capability of new abnormal conditions without losing the old anomaly detection and recognition capability of the anomaly detection model.

[0064] It should be noted that anomaly condition detection is used to detect whether an abnormal condition occurs in the manufacturing process, and if an abnormal condition occurs, the abnormal condition is classified to obtain the abnormal condition and its abnormal type.

[0065] Among them, the abnormal condition can include one or more, that is, the manufacturing process can produce one abnormal condition, or it can produce multiple abnormal conditions. Each abnormal condition has its corresponding abnormal type.

[0066] In specific embodiments, the transformer-based anomaly detection and segmentation algorithm is used for anomaly detection and anomaly classification, thereby realizing anomaly detection of the manufacturing process.

[0067] To train the anomaly detection model, the anomaly detection model is obtained by the following method:

[0068] In the initial training, an original signal data training sample set is obtained, image synthesis processing is performed based on the original signal data training sample set to obtain an image sample set, and the model to be trained is trained based on the image sample set to obtain an anomaly detection model.

[0069] Specifically, in the initial training, all existing signal data is input into the model to be trained for deep learning training to obtain an anomaly detection model with good discrimination ability.

[0070] The original signal data training sample set includes a plurality of original signal data training samples, and each original signal data training sample includes training data corresponding to signal data and an anomaly detection label.

[0071] The image synthesis processing in the training process is basically the same as the image synthesis processing in the application process, and will not be described here.

[0072] The image sample set includes a plurality of image samples, that is, each original signal data training sample corresponds to an image sample.

[0073] In an embodiment, the initial training of the anomaly detection model can be performed on the application terminal of the manufacturing process anomaly detection method.

[0074] In another embodiment, to save the computing resources of the application terminal of the manufacturing process anomaly detection method, the initial training of the anomaly detection model can be performed on another training terminal, and after the anomaly detection model is trained, the anomaly detection model is deployed on the application terminal.

[0075] When a new anomaly occurs, a signal data training sample of the new anomaly is obtained, image synthesis processing is performed based on the signal data training sample to obtain a new anomaly image, and the anomaly detection model is trained based on the new anomaly image to obtain an updated anomaly detection model. The specific execution process is described in the third embodiment below, and will not be described here.

[0076] According to the manufacturing process anomaly detection method provided in the embodiments of the present application, signal data generated in a manufacturing process is acquired, then image synthesis processing is performed based on the signal data to obtain a synthesized image, and finally the synthesized image is input into a continuously learnable anomaly detection model to perform anomaly condition detection and anomaly classification, so that the anomaly condition and the anomaly type output by the anomaly detection model are obtained. The anomaly detection model is used to automatically perform anomaly detection on the manufacturing process, that is, the artificial intelligence technology is used to perform anomaly detection on the manufacturing process, so that the intelligence of the anomaly detection on the manufacturing process is improved. Meanwhile, the anomaly detection model has the capability of continuous learning, and when a new anomaly condition occurs, the anomaly detection model can increase the detection and recognition capability of the new anomaly condition without losing the old anomaly detection and recognition capability of the anomaly detection model, so that the anomaly detection performance of the anomaly detection model is improved, and finally the anomaly detection performance of the manufacturing process is improved.

[0077] Further, based on the first embodiment, a second embodiment of the manufacturing process anomaly detection method is provided. Figure 2 As shown in FIG. 2, in this embodiment, the step 120 includes: Figure 2

[0078] In the step 121, signal preprocessing is performed on the signal data to obtain a preprocessed time sequence signal.

[0079] In specific embodiments, a signal encryption algorithm is used to perform frequency domain conversion on the signal data to obtain a frequency domain signal, and a signal decryption algorithm is used to perform time domain conversion on the frequency domain signal to obtain a time sequence signal. The frequency domain conversion is the conversion from time domain to frequency domain, and the time domain conversion is the conversion from frequency domain to time domain.

[0080] Specifically, the step 121 includes:

[0081] In the step 1211, frequency domain conversion is performed on the signal data to obtain an amplitude spectrum and a phase spectrum corresponding to the signal data.

[0082] In this embodiment, Fourier transform is performed on the signal data to obtain an amplitude spectrum and a phase spectrum corresponding to the signal data.

[0083] In some embodiments, an edge computing server cluster is used to perform frequency domain conversion on the signal data to obtain an amplitude spectrum and a phase spectrum corresponding to the signal data. Specifically, a first edge computing server in the edge computing server cluster is used to perform frequency domain conversion on the signal data to obtain an amplitude spectrum and a phase spectrum corresponding to the signal data.

[0084] In the step 1212, random shuffling processing is performed on the amplitude spectrum to obtain a new amplitude spectrum.

[0085] ​In the embodiment, the random algorithm is used to randomly shuffle the amplitude spectrum to obtain a new amplitude spectrum. That is, the random algorithm is used to randomly shuffle the obtained amplitude spectrum to obtain a series of new amplitude spectra with different lengths.

[0086] In an embodiment, the Las Vegas algorithm is used to randomly shuffle the amplitude spectrum to obtain a new amplitude spectrum. Specifically, the random algorithm of the Las Vegas algorithm is used to randomly shuffle the obtained amplitude spectrum to make it as different from the original signal as possible to obtain a series of new amplitude spectra with different lengths.

[0087] In some embodiments, the edge computing server cluster is used to randomly shuffle the amplitude spectrum to obtain a new amplitude spectrum.

[0088] Step 1213: performing element multiplication operation on the new amplitude spectrum and the phase spectrum to obtain a frequency domain signal;

[0089] In the embodiment, matrix multiplication operation is performed on the new amplitude spectrum and the phase spectrum to obtain a new frequency domain signal.

[0090] In some embodiments, the edge computing server cluster is used to perform element multiplication operation on the new amplitude spectrum and the phase spectrum to obtain a frequency domain signal.

[0091] Step 1214: performing time domain conversion on the frequency domain signal to obtain a time series signal.

[0092] In the embodiment, inverse Fourier transform is performed on the frequency domain signal to obtain a time series signal. The time series signal is a new time series signal, and compared with the signal data before preprocessing, deeper signal features are extracted.

[0093] In some embodiments, the edge computing server cluster is used to perform inverse Fourier transform on the frequency domain signal to obtain a time series signal. Specifically, the second edge computing server in the edge computing server cluster is used to perform inverse Fourier transform on the frequency domain signal to obtain a time series signal.

[0094] It should be noted that the signal preprocessing process of the signal data can be performed on the application terminal of the manufacturing process anomaly detection method. In addition, in order to reduce the waste of computing resources of the application terminal, the signal preprocessing process of the signal data can be performed on the edge computing server cluster to process the signal data on site, that is, the signal data collected by the sensor cluster is directly processed by the edge computing server cluster.

[0095] In an embodiment, the above step 121 includes:

[0096] Step 1215: Receive timing signals sent by the edge computing server cluster. The timing signals are obtained by the edge computing server cluster performing signal preprocessing on the signal data.

[0097] In this embodiment, an edge computing server cluster is used to process the raw signal data, generating a series of data suitable for machine processing. That is, edge computing provides the system with data preprocessing capabilities, thereby reducing the waste of system computing resources. The specific signal preprocessing process can be referred to the signal processing process described above, and will not be repeated here.

[0098] An edge computing server cluster can include one or more edge computing servers, which can be configured according to actual needs.

[0099] In one embodiment, such as Figure 3 As shown, Figure 3 This is a schematic diagram of signal preprocessing for an edge computing server cluster according to an embodiment of the present invention. The edge computing server cluster includes two edge computing servers. Specifically, edge computing server 1 uses a signal encryption algorithm to perform frequency domain conversion on the signal data to obtain the amplitude spectrum and phase spectrum corresponding to the signal data; the amplitude spectrum is randomly shuffled to obtain a new amplitude spectrum; the new amplitude spectrum and the phase spectrum are element-wise multiplied to obtain a frequency domain signal; edge computing server 2 uses a signal decryption algorithm to perform time domain conversion on the frequency domain signal to obtain a time-series signal.

[0100] In this embodiment, by setting up an edge computing server cluster, signal preprocessing is performed on-site, reducing the waste of computing resources of the application terminal of the manufacturing process anomaly detection method of the present invention. At the same time, deeper signal features are extracted to obtain data suitable for machine processing.

[0101] Step 122: Based on the time-series signal, perform image synthesis processing to obtain a synthesized image.

[0102] In some embodiments, the manufacturing process anomaly detection method is applied to a central server cluster, receiving a time-series signal processed by an edge computing server cluster, and performing image synthesis processing based on the time-series signal to obtain a synthesized image.

[0103] In one embodiment, image synthesis processing is performed directly based on a time-series signal to obtain a synthesized image. This synthesized image may include one or more images.

[0104] In another embodiment, step 122 above includes:

[0105] Step 1221: Perform random segmentation on the time-series signal to obtain multiple time-series segments;

[0106] In the embodiment, the time sequence signal is randomly segmented by using a random algorithm to obtain a plurality of time sequence segments. The plurality of time sequence segments are a series of time sequence segments with different lengths.

[0107] In an embodiment, the time sequence signal is randomly segmented by using a Monte Carlo algorithm to obtain a plurality of time sequence segments. Specifically, the time sequence signal is randomly segmented by using a random algorithm of generating random numbers by using the Monte Carlo algorithm to obtain a series of time sequence segments with different lengths.

[0108] In some embodiments, the time sequence signal is randomly segmented by using a central processing server cluster to obtain a plurality of time sequence segments. Specifically, the time sequence signal is randomly segmented by using a first central server in the central server cluster to obtain a plurality of time sequence segments.

[0109] In step 1222, the plurality of time sequence segments are subjected to correlation discrimination processing to obtain correlation data.

[0110] In the embodiment, the correlation of the plurality of time sequence segments is measured by using the maximum mean difference to obtain the correlation data (correlation ranking data).

[0111] In some embodiments, the plurality of time sequence segments are subjected to correlation discrimination processing by using a central processing server cluster to obtain correlation data. Specifically, the plurality of time sequence segments are subjected to correlation discrimination processing by using a second central server in the central server cluster to obtain the correlation data.

[0112] In step 1223, the plurality of time sequence segments are integrated according to the correlation based on the correlation data to obtain a synthesized sequence.

[0113] In the embodiment, the plurality of time sequence segments are integrated according to the correlation based on the correlation data to obtain a synthesized sequence with high to low correlation.

[0114] In some embodiments, the plurality of time sequence segments are integrated according to the correlation based on the correlation data by using a central processing server cluster to obtain a synthesized sequence with high to low correlation. Specifically, the plurality of time sequence segments are integrated according to the correlation based on the correlation data by using a second central server in the central server cluster to obtain a synthesized sequence with high to low correlation.

[0115] In step 1224, an image synthesis processing is performed based on the synthesized sequence to obtain a synthesized image.

[0116] In the embodiment, the synthesized image includes a plurality of images, i.e., a plurality of synthesized images are obtained based on the synthesized sequence.

[0117] In some embodiments, a composite image is obtained by performing image compositing processing based on the composite sequence through a central processing server cluster. Specifically, the composite image is obtained by performing image compositing processing based on the composite sequence through a third central server in the central server cluster.

[0118] In this embodiment, after the time-series signal is randomly segmented, it is then processed into a composite image. Since the same feature information is separated, each composite image input to the anomaly detection model can include deeper feature information, so that the anomaly detection model can obtain more feature information, thereby improving the anomaly detection performance of the anomaly detection model, and further improving the anomaly detection performance of the manufacturing process.

[0119] In a specific embodiment, such as Figure 4 As shown, Figure 4 This is a schematic diagram of signal preprocessing of a central server cluster according to an embodiment of the present invention. The central server cluster includes three central servers. Specifically, central server 1 uses a signal segmentation program to randomly segment the processed time-series signal to obtain multiple segmented time-series segments; central server 2 uses a correlation discrimination program to perform correlation discrimination processing on the multiple time-series segments to obtain correlation data, and based on the correlation data, integrates the multiple time-series segments according to their correlation to obtain a synthesized new sequence; central server 3 uses an image synthesis program to perform image synthesis processing on the synthesized new sequence to obtain a synthesized image.

[0120] In this embodiment, by performing signal preprocessing on the signal data generated during the manufacturing process, a timing signal suitable for machine processing is obtained to better generate a synthetic image, thereby further improving the anomaly detection performance of the manufacturing process.

[0121] Furthermore, based on the first and / or second embodiments described above, a third embodiment of the manufacturing process anomaly detection method of the present invention is proposed. Figure 5 The third flowchart of the manufacturing process anomaly detection method provided by the present invention is as follows: Figure 5 As shown, in this embodiment, the manufacturing process anomaly detection method further includes:

[0122] Step 510: Obtain signal data training samples for new abnormal situations;

[0123] In this embodiment, as the manufacturing process continues to run, new abnormal situations may arise. At this time, the signal data corresponding to the new abnormal situation and its corresponding label are used as signal data training samples to retrain the anomaly detection model, thereby enabling it to detect and classify the abnormal situation corresponding to the new abnormal situation.

[0124] The signal data training sample includes signal data and an anomaly detection label thereof. The specific content of the signal data is basically the same as that of the signal data in step 110, and will not be repeated here.

[0125] In step 520, image synthesis processing is performed based on the signal data training sample to obtain a new abnormal image.

[0126] In this embodiment, image synthesis processing is performed based on the signal data in the signal data training sample to obtain a new abnormal image. The specific execution process is basically the same as that of step 120, and will not be repeated here.

[0127] In step 530, the anomaly detection model is trained based on the new abnormal image to obtain an updated anomaly detection model.

[0128] In this embodiment, the anomaly detection model is deeply trained based on the new abnormal image to obtain an anomaly detection model with enhanced anomaly detection discrimination ability.

[0129] In another embodiment, step 530 includes:

[0130] In step 531, a training sample set of a preset proportion is selected from the original signal data training sample set.

[0131] The original signal data training sample set includes a plurality of original signal data training samples, and each original signal data training sample includes training data corresponding to signal data and an anomaly detection label thereof.

[0132] The preset proportion can be set according to actual needs, for example, the preset proportion is 10%, and the training sample set includes 10% of the training samples in the original signal data training sample set.

[0133] In step 532, an image sample set corresponding to the training sample set is obtained.

[0134] The image sample set includes a plurality of image samples, that is, each original signal data training sample corresponds to an image sample.

[0135] In step 533, the anomaly detection model is trained based on the image sample set and the new abnormal image to obtain an updated anomaly detection model.

[0136] Specifically, the image sample set and the new abnormal image are input into the anomaly detection model for synchronous training to update the anomaly detection model and obtain an updated anomaly detection model.

[0137] In some embodiments, a preset proportion of training sample sets is selected from the original signal data training sample set by the central server cluster, an image sample set corresponding to the training sample set is obtained, the abnormality detection model is trained based on the image sample set and the new abnormal image, and an updated abnormality detection model is obtained.

[0138] The central server cluster comprises a continuous learning program, and the continuous learning program updates the performance of the abnormality detection model to increase the abnormality detection and recognition capability of the abnormality detection model for new abnormalities while not losing the abnormality detection and recognition capability of the abnormality detection model for old abnormalities.

[0139] Specifically, a preset proportion of training sample sets is selected from the original signal data training sample set by a fourth central server in the central server cluster, an image sample set corresponding to the training sample set is obtained, the abnormality detection model is trained based on the image sample set and the new abnormal image, and an updated abnormality detection model is obtained.

[0140] In this embodiment, the abnormality detection model has the capability of continuous learning, and when a new abnormality occurs, the abnormality detection model is trained based on a new abnormal image corresponding to the new abnormality to obtain an updated abnormality detection model, so that the detection and recognition capability of the abnormality detection model for new abnormalities can be increased without losing the detection and recognition capability of the abnormality detection model for old abnormalities, the abnormality detection performance of the abnormality detection model is improved, and finally the abnormality detection performance of the manufacturing process is improved.

[0141] The manufacturing process abnormality detection system provided by the present application is described below, and the manufacturing process abnormality detection system described below can be referred to in conjunction with the manufacturing process abnormality detection method described above.

[0142] In this embodiment, the manufacturing process abnormality detection system comprises a sensor cluster, an edge computing server cluster and a central server cluster.

[0143] The sensor cluster is configured to obtain signal data generated in a manufacturing process.

[0144] The edge computing server cluster is configured to perform signal preprocessing on the signal data to obtain preprocessed time series signals.

[0145] The central server cluster is configured to perform image synthesis processing based on the time series signals to obtain a synthesis image, input the synthesis image into an abnormality detection model capable of continuous learning, perform abnormality detection and abnormality classification, and obtain an abnormality and an abnormality type output by the abnormality detection model.

[0146] The edge computing server cluster is further configured to perform frequency domain conversion on the signal data to obtain an amplitude spectrum and a phase spectrum corresponding to the signal data, perform random shuffling on the amplitude spectrum to obtain a new amplitude spectrum, perform element multiplication on the new amplitude spectrum and the phase spectrum to obtain a frequency domain signal, and perform time domain conversion on the frequency domain signal to obtain a time sequence signal.

[0147] The center server cluster is further configured to perform random segmentation on the time sequence signal to obtain a plurality of time sequence segments, perform correlation discrimination on the plurality of time sequence segments to obtain correlation data, integrate the plurality of time sequence segments according to the correlation data to obtain a synthetic sequence, and perform image synthesis on the synthetic sequence to obtain a synthetic image.

[0148] The center server cluster is further configured to obtain signal data training samples of a new abnormal situation, perform image synthesis on the signal data training samples to obtain a new abnormal image, and train the abnormal detection model based on the new abnormal image to obtain an updated abnormal detection model.

[0149] The center server cluster is further configured to select a training sample set of a preset proportion from an original signal data training sample set, obtain an image sample set corresponding to the training sample set, and train the abnormal detection model based on the image sample set and the new abnormal image to obtain an updated abnormal detection model.

[0150] The manufacturing process abnormality detection device provided by the present application is described below, and the manufacturing process abnormality detection device described below can be referred to in correspondence with the manufacturing process abnormality detection method described above.

[0151] In this embodiment, the manufacturing process abnormality detection device comprises:

[0152] The acquisition module is configured to acquire signal data generated by a manufacturing process.

[0153] The synthesis module is configured to perform image synthesis on the signal data to obtain a synthetic image.

[0154] The detection module is configured to input the synthetic image into a continuously learnable abnormality detection model to perform abnormality detection and abnormality classification, and obtain an abnormal situation and an abnormal type output by the abnormality detection model.

[0155] The synthesis module is further configured to perform signal preprocessing on the signal data to obtain a preprocessed time sequence signal, and perform image synthesis on the time sequence signal to obtain a synthetic image.

[0156] The synthesis module is further configured to perform frequency domain conversion on the signal data to obtain an amplitude spectrum and a phase spectrum corresponding to the signal data; perform random shuffling on the amplitude spectrum to obtain a new amplitude spectrum; perform element multiplication on the new amplitude spectrum and the phase spectrum to obtain a frequency domain signal; and perform time domain conversion on the frequency domain signal to obtain a time sequence signal.

[0157] The synthesis module is further configured to receive a time sequence signal sent by an edge computing server cluster, the time sequence signal being obtained by performing signal preprocessing on the signal data by the edge computing server cluster.

[0158] The synthesis module is further configured to perform random segmentation processing on the time sequence signal to obtain a plurality of time sequence segments; perform correlation discrimination processing on the plurality of time sequence segments to obtain correlation data; integrate the plurality of time sequence segments according to the correlation based on the correlation data to obtain a synthesis sequence; and perform image synthesis processing based on the synthesis sequence to obtain a synthesis image.

[0159] The manufacturing process anomaly detection apparatus further comprises:

[0160] The sample acquisition module is configured to acquire signal data training samples of a new abnormal situation;

[0161] The image synthesis module is configured to perform image synthesis processing based on the signal data training samples to obtain a new abnormal image.

[0162] The anomaly detection module is configured to train the anomaly detection model based on the new abnormal image to obtain an updated anomaly detection model.

[0163] The anomaly detection module is further configured to select a training sample set of a preset proportion from an original signal data training sample set; acquire an image sample set corresponding to the training sample set; and train the anomaly detection model based on the image sample set and the new abnormal image to obtain an updated anomaly detection model.

[0164] Figure 6 An example of an entity structure diagram of an electronic device is shown in FIG. 1. Figure 6As shown, the electronic device can include a processor 610, a communications interface 620, a memory 630, and a communications bus 640, wherein the processor 610, the communications interface 620, and the memory 630 complete mutual communication through the communications bus 640. The processor 610 can invoke a logical instruction in the memory 630 to execute a manufacturing process anomaly detection method, which includes: acquiring signal data generated by a manufacturing process; performing image synthesis processing based on the signal data to obtain a synthesized image; inputting the synthesized image into a continuously learnable anomaly detection model to perform anomaly condition detection and anomaly classification, and obtaining an anomaly condition and an anomaly type output by the anomaly detection model.

[0165] In addition, the logical instruction in the memory 630 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer-readable storage medium. Based on such understanding, the technical solutions of the present application or parts of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0166] On the other hand, the present application also provides a computer program product, which includes a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor, so that the computer can execute the manufacturing process anomaly detection method provided by the above-mentioned methods, which includes: acquiring signal data generated by a manufacturing process; performing image synthesis processing based on the signal data to obtain a synthesized image; inputting the synthesized image into a continuously learnable anomaly detection model to perform anomaly condition detection and anomaly classification, and obtaining an anomaly condition and an anomaly type output by the anomaly detection model.

[0167] In yet another aspect, the present application also provides a non-transitory computer readable storage medium having stored thereon a computer program, which, when executed by a processor, implements a manufacturing process anomaly detection method provided by each of the above methods, the method comprising: acquiring signal data generated by a manufacturing process; performing image synthesis processing based on the signal data to obtain a synthesized image; inputting the synthesized image into a continuously learnable anomaly detection model to perform anomaly condition detection and anomaly classification, and obtaining an anomaly condition and an anomaly type output by the anomaly detection model.

[0168] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0169] From the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and necessary general hardware platforms, and of course can also be realized by hardware. Based on such understanding, the above technical solutions, essentially or in the form of software products, can be embodied in a computer software product, which can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0170] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A manufacturing process anomaly detection method characterized by, The method comprises the following steps: acquiring signal data generated by a manufacturing process; performing signal preprocessing on the signal data to obtain preprocessed time series signals; performing image synthesis processing based on the time series signals to obtain a synthesized image; inputting the synthesized image into a continuously learnable anomaly detection model to perform anomaly detection and anomaly classification, and obtaining an anomaly situation and an anomaly type output by the anomaly detection model; the signal preprocessing on the signal data to obtain preprocessed time series signals comprises the following steps: performing frequency domain conversion on the signal data to obtain an amplitude spectrum and a phase spectrum corresponding to the signal data; performing random shuffling processing on the amplitude spectrum to obtain a new amplitude spectrum; performing element multiplication operation on the new amplitude spectrum and the phase spectrum to obtain a frequency domain signal; performing time domain conversion on the frequency domain signal to obtain a time series signal; the image synthesis processing based on the time series signals to obtain a synthesized image comprises the following steps: performing random segmentation processing on the time series signals to obtain a plurality of time series segments; performing correlation discrimination processing on the plurality of time series segments to obtain correlation data; integrating the plurality of time series segments according to the correlation data to obtain a synthesized sequence; performing image synthesis processing based on the synthesized sequence to obtain a synthesized image; the method further comprises the following steps: when a new anomaly situation occurs, acquiring signal data training samples of the new anomaly situation; performing image synthesis processing based on the signal data training samples to obtain a new anomaly image; training the anomaly detection model based on the new anomaly image to obtain an updated anomaly detection model; the training of the anomaly detection model based on the new anomaly image to obtain an updated anomaly detection model comprises the following steps: selecting a training sample set with a preset proportion from an original signal data training sample set; acquiring an image sample set corresponding to the training sample set; training the anomaly detection model based on the image sample set and the new anomaly image to obtain an updated anomaly detection model.

2. The manufacturing process abnormality detection method according to claim 1, characterized by, the signal preprocessing on the signal data to obtain preprocessed time series signals comprises the following steps: receiving time series signals sent by an edge computing server cluster, wherein the time series signals are obtained by performing signal preprocessing on the signal data by the edge computing server cluster.

3. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps of the manufacturing process anomaly detection method according to any one of claims 1 to 2.

4. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the manufacturing process anomaly detection method according to any one of claims 1 to 2.

5. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the manufacturing process anomaly detection method according to any one of claims 1 to 2.

Images