Multi-working-condition abnormal mode recognition method for edge cloud cooperative equipment

Abstract

The invention provides a side cloud collaborative equipment multi-working-condition abnormal mode identification method. The method comprises the following steps: complex manufacturing system equipment sensor deployment and multi-modal processing data acquisition; equipment time sequence data preprocessing; cleaning, standardization and feature engineering are carried out on the multi-modal processing data collected from each sensor on the edge side; an edge side model is built and updated, on the edge side, an improved OPTICS algorithm is adopted, the preprocessed equipment time sequence data is clustered, other data, obtained through clustering, in the whole cluster is recognized through label data, and an edge side database is updated; uploading the equipment time sequence data subjected to edge side preprocessing in the step 2 to a cloud end, and deploying a transfer learning model in the cloud end; and precise identification of the abnormal mode of the complex manufacturing system is realized by utilizing edge cloud collaboration.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment monitoring and fault diagnosis of complex manufacturing systems, and in particular to a method for identifying abnormal patterns in multiple working conditions of equipment based on an edge-cloud collaborative architecture. By integrating an improved OPTICS algorithm (Ordering Points to identify the clustering structure) and a deep adversarial neural network (DANN), efficient and accurate identification of abnormal patterns under various working conditions during equipment operation is achieved, so as to ensure the stable operation of the complex manufacturing system and improve production efficiency. Background Art

[0002] With the continuous upgrading of the global manufacturing industry and the improvement of national industrialization levels, manufacturing has become an important pillar of economic development in various countries. In modern complex manufacturing systems, the stable operation of equipment has a vital impact on production efficiency, product quality and corporate economic benefits. With the continuous development of manufacturing technology, the structure and operating conditions of equipment are becoming increasingly complex, and traditional equipment abnormal pattern recognition methods face many challenges.

[0003] Traditional methods based on single sensor data or simple threshold judgments are difficult to fully and accurately capture abnormal information of equipment under multiple working conditions. Since the operating data of equipment in complex manufacturing systems is multimodal, time-series, and has complex data distribution, the analysis of a single data source often has limitations and cannot comprehensively consider the impact of multiple key factors on the equipment status, resulting in a low accuracy rate in abnormal pattern recognition. [1] .

[0004] In terms of data processing, traditional methods lack effective data preprocessing and feature extraction methods, making it difficult to extract valuable information for anomaly identification from massive raw data. In addition, when faced with data of different magnitudes and dimensions, effective normalization cannot be performed, affecting the accuracy of subsequent model training and analysis.

[0005] Existing clustering algorithms such as the OPTICS algorithm have problems such as difficulty in parameter selection, high sensitivity to data, and poor processing of density-varying data distribution when applied to equipment anomaly pattern recognition, resulting in unsatisfactory clustering results and inability to accurately classify equipment anomaly patterns. [2] In addition, traditional methods lack the ability to effectively transfer learning data from different working conditions. When faced with new working condition data, they need to collect a large amount of data and train the model again, which is computationally inefficient and difficult to meet the needs of intelligent manufacturing systems for real-time monitoring of equipment and rapid fault diagnosis. [3] .

[0006] At the same time, the data architecture of modern complex manufacturing systems is developing towards a cloud-edge-end model, forming a data uplink of "end-side collection, edge-side aggregation, and cloud-based processing" and a data downlink of "cloud-based training, edge-based reasoning, and end-side execution". [4] The change in data structure has also created new requirements for the deployment of algorithms, which requires giving full play to the advantages of cloud-edge-device architecture.

[0007] In summary, a new method for equipment multi-condition abnormal pattern recognition is needed to overcome the shortcomings of traditional methods, improve the accuracy and computational efficiency of abnormal pattern recognition of equipment in complex manufacturing systems, and ensure the stability and reliability of the production process.

[0008] Related Literature

[0009] [1]LONG M, CAO Y, WANG J, et al. Learning Transferable Features with DeepAdaptation Networks[A / OL].arXiv,2015[2024-11-25].http: / / arxiv.org / abs / 1502.02791.DOI:10.48550 / arXiv.1502.02791.

[0010] [2]RAJABI F,MCARTHUR J J.Applying OPTICS with and without PCA forfault detection of fan coil units using building automation system data[J / OL].Energy and Buildings,2024,317:114368.DOI:10.1016 / j.enbuild.2024.114368.

[0011] [3 Shan Fengwu, Wang Chaoren, Zeng Jianbang, et al. Analysis method of power battery monomer inconsistency fault degree based on OPTICS clustering and voltage anomaly index[J]. Science China: Technological Sciences, 2024, 54(8): 1611-1624.

[0012] [4] ECC released three white papers: "White Paper on Collaboration between Edge Computing and Cloud Computing 2.0", "White Paper on Edge Computing Network of Industrial Internet in the 5G Era", and "White Paper on Edge Computing Node of Industrial Internet 1.0" [J]. Automation Expo, 2021, 38(2): 3-3. Summary of the invention

[0013] The present invention proposes a method for identifying abnormal patterns of multiple working conditions of equipment in complex manufacturing systems. The method is based on the improved OPTICS algorithm and DANN using the edge-cloud collaborative data distribution architecture, aiming to achieve timely and efficient identification of abnormal patterns of equipment in discrete large-scale complex manufacturing systems. Compared with traditional methods, the present invention combines clustering and transfer learning techniques, and adopts an edge-cloud collaborative training architecture, thereby effectively improving the accuracy and computational efficiency of equipment abnormal pattern recognition. The main technical solutions are as follows:

[0014] A method for identifying abnormal patterns of multiple working conditions of edge-cloud collaborative equipment includes the following steps:

[0015] Step 1: Deployment of equipment sensors for complex manufacturing systems and collection of multi-modal processing data:

[0016] Step 2: Preprocessing of device time series data: Cleaning, standardization, and feature engineering of multimodal processing data collected from various sensors on the edge side;

[0017] Step 3: Building and updating edge-side models: On the edge side, an improved OPTICS algorithm is used to cluster the preprocessed device time series data, and the label data is used to identify other data in the entire cluster obtained by clustering, and the edge-side database is updated; the improved OPTICS algorithm is as follows: improving the local density calculation method, calculating the local density based on the inverse of the average distance of the k nearest neighbors of the data point; adaptively adjusting the density threshold, determining the density threshold based on the mean and standard deviation of the local density of the data point; determining the cluster center based on the density peak value, and taking the data point with a local density greater than the density threshold as the cluster center; cluster merging operation, calculating the distance matrix between cluster centers, and performing cluster merging operation when the cluster center distance is less than the preset distance threshold and the local density difference is within a certain range;

[0018] Step 4: upload the device time series data pre-processed on the edge side in step 2 to the cloud, and deploy the transfer learning model in the cloud;

[0019] Step five: Use edge-cloud collaboration to identify abnormal patterns in complex manufacturing systems.

[0020] Furthermore, the method of step 2 is:

[0021] 1) Data cleaning, including

[0022] Missing value processing: Newton interpolation method was used;

[0023] Outlier monitoring and processing: Use box plot technology and 3σ principle to identify outliers;

[0024] Data deduplication;

[0025] 2) Data standardization: Min-Max standardization is used;

[0026] 3) Feature Engineering: Using feature construction and multi-dimensional feature fusion methods, multiple time series signals containing equipment status features are fused through an automatic encoder to extract fault signal features.

[0027] Furthermore, in step 3, the density threshold ρ th The calculation formula is as follows:

[0028]

[0029] In the formula, is the mean and σ of the local density of the data points ρ is the standard deviation.

[0030] Furthermore, in step three, the distance threshold is 0.1 of the average neighbor distance; and the local density difference is within a certain range, which means that the local density difference is within a numerical range of 0.1 of the local density standard deviation of the data point.

[0031] Furthermore, in step three, the method for performing the cluster merging operation is: for two cluster centers, the data point densities of the two cluster centers are compared, the cluster center with smaller data point density is merged into another cluster center, and the cluster labels of the data points related thereto are modified to be the same label as that of the other cluster center.

[0032] Furthermore, in step 4, the pre-trained network RestNeXt is used to extract features from the pre-processed device timing data.

[0033] Furthermore, the method of step five is:

[0034] 1) Use historical operating data stored in the cloud for joint training to identify the status of operating status data;

[0035] 2) Use the DANN-based migration recognition architecture to identify the pattern of the operating status data: Use the ResNeXt pre-trained network to further extract features from the historical operating data and operating status data; initialize the DANN network, use the further extracted feature data, use the contrast loss joint training method to train the DANN network, and use the back-propagation method to update the parameters;

[0036] 4) Use the trained DANN network to perform pattern recognition on the operating status data and evaluate the recognition results;

[0037] 5) After identifying that there is enough label data, the label is sent to the edge side database for updating the edge side database.

[0038] This patent is based on the improved OPTICS algorithm and DANN algorithm, using the edge-cloud collaborative data distribution architecture, effectively realizing the abnormal pattern recognition of complex manufacturing system equipment in multiple working conditions with the advantages of real-time, high efficiency and accuracy, and providing an efficient solution for the abnormal pattern recognition of complex manufacturing system equipment. This method can not only improve production efficiency and reduce maintenance costs, but also enhance the safety and stability of the production line, injecting new vitality and impetus into the development of the intelligent manufacturing industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Transfer learning technology route

[0040] Figure 2 Clustering algorithm route

[0041] Figure 3 DANN network structure

[0042] Figure 4 ResNeXt network structure data DETAILED DESCRIPTION

[0043] The present invention is implemented by the following technical solutions, and the transfer learning process is as follows: Figure 1 As shown, the specific implementation steps are:

[0044] Step 1: Complex manufacturing system equipment sensor deployment and multi-modal processing data collection

[0045] 1) Select and arrange various types of sensors according to the specific structure of complex manufacturing system equipment and the operating characteristics of key parts. Deploy the following types of sensors (including but not limited to the following categories) at each key node:

[0046] Vibration sensors are installed on mechanical moving parts. They can detect tiny vibrations and irregular movements during equipment operation, and are important for early warning of mechanical failures such as bearing damage and gear wear.

[0047] Temperature sensors are fixed near heat-generating components such as motors and transformers. These sensors can monitor temperature changes during equipment operation;

[0048] Current sensors are installed on the main power lines to monitor the current status of cable production equipment. Abnormal current fluctuations are often a precursor to electrical faults, such as short circuits or aging of electrical components;

[0049] Acoustic sensors, deployed around noisy equipment, identify mechanical faults by analyzing changes in the sound spectrum.

[0050] 2) Data collection and integration

[0051] After the sensor deployment is completed, all sensors are connected to a centralized data collection system, which is responsible for collecting data from each sensor and forwarding the data to the main data processing center on the edge side.

[0052] 3) Communication technology

[0053] Industrial Ethernet or wireless communication technology is used to ensure real-time data transmission. High data transmission rate and strong anti-interference ability are used to ensure the reliability and real-time performance of data transmission.

[0054] 4) Data visualization

[0055] Implement advanced data visualization technology and use real-time data dashboards to provide operators with instant production line status displays. These visualization tools not only help operators quickly identify problem areas, but also monitor the production process in real time and optimize operational decisions.

[0056] Step 2: Preprocessing of device timing data

[0057] Data preprocessing is a crucial step in realizing abnormal pattern recognition of equipment in complex manufacturing systems. In order to ensure the accuracy and effectiveness of subsequent model training, the raw data collected from each sensor is thoroughly cleaned, standardized, and feature engineered on the edge side. The specific steps are as follows:

[0058] 1) Data cleaning. The main purpose of data cleaning is to eliminate invalid, erroneous or incomplete data that may interfere with model training and prediction accuracy. This process mainly includes: missing value processing. Since this patent collects time series data, Newton interpolation method is used to estimate missing values. The steps of using Newton interpolation method for calculation are as follows: First, find the known n point pairs (x1, y1), (x2, y2)…(x n ,y n ) all order difference quotient formulas:

[0059]

[0060] Combining the above difference quotient formulas, we can establish the following interpolation polynomial f(x).

[0061] f(x)=f(x1) + (x-x1)f[x2,x1] + (x-x1)(x-x2)f[x3,x2,x1]+(x-x1)(x-x2)(x-x3)f[x4,x3,x2,x1]+…+(x-x1)(x-x2)…(xx n-1 )f[x n ,x n-1 ,…x2,x1]+(x-x1)(x-x2)…(xn )f[x n ,x n-1 ,…x1,x]=P(x)+R(x)

[0062] in:

[0063] P(x)=f(x1) + (x-x1)f[x2,x1] + (x-x1)(x-x2)f[x3,x2,x1]+(x-x1)(x-x2)(x-x3)f[x4,x3,x2,x1]+…+(x-x1)(x-x2)…(x-xn -1 )f[x n ,x n-1 ,…,x2,x1]

[0064] R(x)=(x-x1)(x-x2)…(x n )f[x n ,x n-1 ,…x1,x]

[0065] P(x) is the Newton interpolation approximation function, and R(x) is the error function. Substituting the point x corresponding to the missing value into the interpolation polynomial can get the approximate value f(x) of the missing value.

[0066] Outlier monitoring and processing: Sensor failure, operational errors, or external environmental interference may cause particularly obvious outliers. Box plot (IQR, interquartile range) technology and the 3σ principle (data values ​​outside the mean ± 3 times the standard deviation range are considered abnormal) are used to identify outliers. Once an outlier is detected, it is treated as a missing value and supplemented by interpolation.

[0067] 2) Data standardization. Data standardization is to eliminate the impact of different magnitudes and dimensions, so that the data is at the same magnitude, which is convenient for comprehensive analysis. This patent uses Min-Max standardization to scale the data to a specified maximum and minimum value (usually 1 and 0). The formula is:

[0068]

[0069] 3) Feature engineering. This patent uses feature construction and multi-dimensional feature fusion methods to extract the most useful information for model prediction from the original data. This patent performs STFT transformation on the data obtained by the vibration sensor, constructs frequency domain features, effectively reveals the operating status of the mechanical equipment, and then uses an automatic encoder to fuse multiple time series signals containing equipment status features, such as multiple vibration features, to extract more comprehensive fault signal features.

[0070] Through the above detailed data preprocessing steps, the quality of data and the performance of subsequent models can be greatly improved, laying a solid data foundation for the accurate identification of equipment operating status. These processing processes not only accurately process the raw data collected from complex production environments, but also extract information that is crucial to abnormality identification through scientific methods, providing reliable support for the subsequent identification of abnormal equipment patterns. At the same time, when training the subsequent cloud model, uploading the preprocessed data to the cloud can reduce the amount of data transmitted and save computing resources.

[0071] Step 3: Building the edge model. On the edge, this patent uses an improved OPTICS algorithm to identify unknown data using labeled data. OPTICS clusters the preprocessed data, and different types of data are assigned to the same cluster. If there is a certain labeled data in the same cluster, then other unknown data in the cluster should also be of the same label type. In this way, the recognition of unknown patterns is achieved. The improvements to the OPTICS algorithm are mainly in the following four aspects:

[0072] 1) Improvement of local density calculation method

[0073] In the OPTICS algorithm, the density calculation of data points is often based on a fixed domain radius and a fixed number of neighbors, and the density is measured by counting the number of data points in the neighborhood. For example, when the density calculation method based on the domain radius is used, for a data point x i , and its density is estimated as:

[0074]

[0075] Among them, N ∈ (x i ) represents x i The area with the center and radius ∈, χ(x j ) is an indicator function, when x j The value is 1 if it is in the domain, otherwise it is 0.

[0076] The traditional local density calculation method is difficult to capture complex distribution data, which will lead to inaccurate division of different density areas. Therefore, this patent uses the inverse of the average distance of the k nearest neighbors of the data point to calculate the local density. Suppose the data point set is X = {x1, x2, ..., x n}, for data point x i , his k nearest neighbor distance set is D k (x i )={d i1 ,d i2 ,…,d ik}, then the local density calculation formula is as follows:

[0077]

[0078] Compared with the traditional method based on a fixed neighborhood radius, the improved calculation method of this patent can better reflect the closeness of the data point with the surrounding data points in its local area. For data with uneven density or complex distribution, it can more flexibly capture the actual density of the data point. At the same time, because this method dynamically measures the density according to the nearest neighbor distance of the data itself, it avoids the problem of inaccurate division of different density areas that may be caused by pre-setting a fixed neighborhood radius.

[0079] 2) Adaptive adjustment of density threshold

[0080] The traditional OPTICS algorithm usually requires humans to set a fixed value based on experience or preliminary observation of the data when determining the density threshold. It is difficult to grasp the appropriate threshold when faced with complex and diverse data. At the same time, the manual setting method does not meet the requirements of intelligent manufacturing for automated identification.

[0081] Based on the above reasons, this patent introduces a density threshold adaptive adjustment mechanism. First, the mean of the density of all data points is calculated. and standard deviation σ ρ , density threshold ρ th The formula is as follows:

[0082]

[0083] Through this adaptive method based on the statistical characteristics of the data's own density, the algorithm can dynamically determine a reasonable threshold based on the actual density distribution of the input data to distinguish different potential clusters. This allows the clustering algorithm to better adapt to density changes in different data sets without much human intervention and more accurately identify clusters with different density characteristics.

[0084] 3) Determination of cluster center based on density peak

[0085] The OPTICS algorithm does not explicitly determine the cluster center in a specific density peak-based manner during the clustering process. Instead, it gradually reveals the cluster structure by constructing an ordered queue and subsequently analyzing reachable relationships. The cluster center is relatively implicit in the structure formed by the entire clustering process.

[0086] Therefore, this patent explicitly defines the local density as being greater than the determined density threshold ρ th The data points are taken as cluster centers. Assuming the data point index set is I = {1, 2, ..., n}, the cluster center index set C can be determined as follows:

[0087] C={i∈I|ρ(x i )>ρth}

[0088] This method of determining cluster centers based on density peaks allows the algorithm to focus more specifically on those points in the data distribution that are relatively dense and at the core of the cluster. These cluster centers can provide a reasonable reference for subsequent cluster merging and other operations, and help to more clearly divide and integrate different clusters, especially when there are multiple cluster structures with different densities but are interrelated in the data, which can better sort out the core areas of each cluster.

[0089] 4) Cluster merging operation

[0090] After clustering, the traditional OPTICS algorithm often generates clustering results according to its rules based on density reachability and ordered queue construction. However, for some data distributions with blurred boundaries and density gradients, it may lead to the generation of too many relatively fragmented clusters. These clusters should be merged into a whole in the actual data sense, but the algorithm itself lacks an effective automatic merging mechanism.

[0091] Based on the above problems, the cluster merging operation is introduced. First, the distance matrix D between cluster centers is calculated. cc , for cluster centers and (where c i ,c j ∈C), and its distance is recorded as The distance matrix elements are expressed as The Euclidean distance can be obtained by calculating the 'cdist' function in Python. Then for any two cluster centers and (i≠j), cluster merging is performed if the following two conditions are met:

[0092]

[0093] Among them, the first condition indicates that the distance between the two cluster centers is relatively small, which is less than a certain proportion of the average neighbor distance (here 0.1), which means that they are relatively close in spatial position; the second condition indicates that the density difference between the two cluster centers is within a certain range (measured by a certain multiple of the density standard deviation), indicating that they are relatively similar in local density characteristics.

[0094] When the above conditions are met, clusters are merged according to the center density of the two clusters. Then the cluster center The cluster labels of related data points (determined by the neighbor relationship index in the 'indices' record) are modified to be related to the cluster center The same label and vice versa. Through such cluster merging operation, the clustering results can be reasonably integrated to avoid over-segmentation, so that the final clustering results are more in line with the actual internal structure and distribution pattern of the data.

[0095] Through the above improvements, the adaptability of the algorithm in the task of abnormal pattern recognition of equipment in intelligent manufacturing system is enhanced, and the degree of automation and accuracy of the algorithm are improved. Finally, the overall technical route of using the improved OPTICS algorithm to identify abnormal patterns of equipment is as follows: Figure 2 shown.

[0096] Step 4: Deploy the transfer learning model in the cloud. The specific steps are as follows:

[0097] 1) Use the pre-trained network RestNeXt to extract features from the pre-processed data. The network structure information used is as follows Figure 4 shown.

[0098] 2) Initialize the DANN network. Figure 3 As shown in the figure, it mainly consists of three parts: Feature Extractor, Label Predictor, and Domain Discriminator.

[0099] The feature extractor has the characteristic of being shared between domains. Let the input data be x, and the feature extractor uses the function G f Indicates that the extracted feature is f = G f (x). Usually the feature extractor is a neural network composed of multiple convolutional layers or fully connected layers. For example, for a simple full neural network as a feature extractor, assuming that the input x is a d-dimensional vector, the first fully connected layer has n1 neurons, the weight matrix is ​​W1, and the bias vector is b1, then the output of this layer is h1 = σ(W1x+b1), where σ is the activation function (for example, the ReLU function σ(z) = max(0,z)). After multiple layers of such calculations, the final feature f is obtained.

[0100] The label predictor is used to predict the label of the data based on the extracted features. Let the label predictor G y , the predicted label probability distribution is y = G y (f). For the abnormal pattern recognition task of complex manufacturing system equipment, we can regard it as a multi-classification problem and use the softmax function to predict the label probability. Assume that the dimension of feature f is m, the number of classification categories is C, and the weight matrix of the last layer of the label predictor is W y , the bias vector is b y ,but:

[0101] y=softmax(W y f+b y )

[0102]

[0103] z=W y f+b y

[0104] where each element y i It represents the probability that the data belongs to the i-th category.

[0105] The role of the domain discriminator is to distinguish whether the feature comes from the source domain or the target domain. Let the domain discriminator be G d , which takes the extracted feature f as input and outputs a scalar, indicating the probability d = G that the input feature belongs to the source domain d (f). For example, for a simple fully connected layer as a domain discriminator, assuming that the dimension of the input feature f is n, the weight matrix is ​​Wd, and the bias vector is bd, the sigmoid function can be used to output the probability:

[0106] d=σ s (W d f+b d )

[0107] in:

[0108] The objective functions of DANN during training include label prediction loss (Label Prediction Loss), domain discrimination loss (Domain Discrimination Loss) and overall loss function.

[0109] For the source domain data x s , whose label is known to be y s , cross entropy loss is used to measure the accuracy of label prediction.

[0110] Assume the label prediction loss is Ly, then

[0111]

[0112] where y s,i is the true label y s The i-th element of G y (G f (x s )) i ) is the predicted label probability distribution G y (G f (x s ))'s i-th element.

[0113] The goal of domain discrimination loss is to make the domain discriminator distinguish the source domain and the target domain as much as possible, while making the features learned by the feature extractor confuse the domain discriminator. Let the source domain data be represented as x s , the target domain data is represented as x t , the probability of the domain discriminator outputting the source domain data is:

[0114] d s =G d (G f (x s ))

[0115] The probability of the target domain data output is

[0116] d t =G d (G f (x t ))

[0117] The binary cross entropy loss is used to measure the performance of the domain discriminator. The domain discrimination loss is

[0118]

[0119] where d x When x=x s When d s , when x=x t When d t The final overall training goal is the total loss function

[0120] L=L y -λL d

[0121] Where λ is a balancing parameter used to balance the importance of label prediction loss and domain discrimination loss. During the training process, the gradient reversal layer (GRL) is used to achieve this training process to minimize the overall loss L.

[0122] 3) Use contrast loss joint training to train the model. Suppose two samples are represented as x1 and x2 in the feature space, then the distance between them is measured using the Euclidean distance (where n is the feature dimension)

[0123]

[0124] Define the contrast loss L. For a pair of samples (x1, x2) and their labels y (y = 1 means the samples are similar, y = 0 means the samples are not similar), the contrast loss formula is:

[0125]

[0126] During calculation, the contrast loss of source domain data and target domain data is averaged to obtain a stable loss to optimize the model.

[0127] 4) Use back propagation to update model parameters.

[0128] Step 5: Use edge-cloud collaboration to accurately identify abnormal patterns in complex manufacturing systems, such as Figure 1 , the specific steps are as follows:

[0129] 1) Edge-side sensors collect and pre-process data.

[0130] 2) Upload the pre-processed data on the edge side to the cloud, and use the historical operating condition data stored in the cloud for joint training to identify the status of the operating status data.

[0131] 3) Use the DANN-based migration recognition architecture to identify the pattern of the operating status data. First, use the ResNeXt pre-trained network to further extract features from the historical operating data and operating status data. Then initialize the DANN network, use the extracted feature data, and use the contrast loss joint training method to train the network, and use the back propagation method to update the parameters.

[0132] 4) Use the trained DANN network to perform pattern recognition on the operating status data and evaluate the recognition results.

[0133] 5) After identifying that there is enough labeled data, the labels can be sent to the edge database and synchronously identified using the improved OPTICS clustering algorithm mentioned in step 3. Using the label data to identify other data in the entire cluster obtained by clustering can improve recognition efficiency and accuracy and make efficient use of computing resources.

Claims

1. A method for identifying abnormal patterns of multiple working conditions of edge-cloud collaborative equipment, comprising the following steps: Step 1: Deployment of equipment sensors for complex manufacturing systems and collection of multi-modal processing data: Step 2: Preprocessing of device time series data: Cleaning, standardization, and feature engineering of multimodal processing data collected from various sensors on the edge side; Step 3: Building and updating edge-side models: On the edge side, an improved OPTICS algorithm is used to cluster the preprocessed device time series data, and label data is used to identify other data in the entire cluster obtained by clustering, and the edge-side database is updated; the improved OPTICS algorithm is as follows: improving the local density calculation method, calculating the local density based on the inverse of the average distance of the k nearest neighbors of the data point; adaptively adjusting the density threshold, determining the density threshold based on the mean and standard deviation of the local density of the data point; determining the cluster center based on the density peak value, and taking the data point with a local density greater than the density threshold as the cluster center; cluster merging operation, calculating the distance matrix between cluster centers, and performing cluster merging operation when the cluster center distance is less than the preset distance threshold and the local density difference is within a certain range; Step 4: upload the device time series data pre-processed on the edge side in step 2 to the cloud, and deploy the transfer learning model in the cloud; Step five: Use edge-cloud collaboration to identify abnormal patterns in complex manufacturing systems.

2. The method for identifying abnormal patterns of multiple working conditions of edge-cloud collaborative devices according to claim 1 is characterized in that: The method for step one is as follows: 1) Arrange various types of sensors according to the equipment structure and operating characteristics of key parts, including vibration sensors, temperature sensors, current sensors and sound sensors; Vibration sensors, mounted on mechanical moving parts; Temperature sensors, which are fixed near heat-generating components including motors and transformers; Current sensors, installed on the main power lines, are used to monitor the current status of cable production equipment; Acoustic sensors are used to identify mechanical faults by analyzing changes in the sound spectrum. 2) Sensor data collection and integration: Collect data from each sensor, obtain multimodal processing data, and forward the multimodal processing data to the main data processing center on the edge side.

3. The method for recognizing abnormal patterns of multiple working conditions of edge-cloud collaborative devices according to claim 1 is characterized in that: The method of step 2 is: 1) Data cleaning, including Missing value processing: Newton interpolation method was used; Outlier monitoring and processing: Use box plot technology and 3σ principle to identify outliers; Data deduplication; 2) Data standardization: Min-Max standardization is used; 3) Feature Engineering: Using feature construction and multi-dimensional feature fusion methods, multiple time series signals containing equipment status features are fused through an automatic encoder to extract fault signal features.

4. The method for identifying abnormal patterns of multiple working conditions of edge-cloud collaborative devices according to claim 1 is characterized in that: In step 3, the density threshold ρ th The calculation formula is as follows: In the formula, is the mean and σ of the local density of the data points ρ is the standard deviation.

5. The method for recognizing abnormal patterns of multiple working conditions of edge-cloud collaborative devices according to claim 1 is characterized in that: In step three, the distance threshold is 0.1 of the average neighbor distance; the local density difference is within a certain range, which means that the local density difference is within the numerical range of 0.1 of the local density standard deviation of the data point.

6. The method for recognizing abnormal patterns of multiple working conditions of edge-cloud collaborative devices according to claim 1 is characterized in that: In step three, the method for performing the cluster merging operation is as follows: for two cluster centers, the data point densities of the two cluster centers are compared, the cluster center with smaller data point density is merged into another cluster center, and the cluster labels of the data points related to it are modified to be the same label as that of the other cluster center.

7. The method for recognizing abnormal patterns of multiple working conditions of edge-cloud collaborative devices according to claim 1 is characterized in that: In step 4, the pre-trained network RestNeXt is used to extract features from the pre-processed device timing data.

8. The method for recognizing abnormal patterns of multiple working conditions of edge-cloud collaborative devices according to claim 1 is characterized in that: The method of step five is: 1) Use historical operating data stored in the cloud for joint training to identify the status of operating status data; 2) Use the DANN-based migration recognition architecture to identify the pattern of the operating status data: Use the ResNeXt pre-trained network to further extract features from the historical operating data and operating status data; initialize the DANN network, use the further extracted feature data, use the contrast loss joint training method to train the DANN network, and use the back-propagation method to update the parameters; 4) Use the trained DANN network to perform pattern recognition on the operating status data and evaluate the recognition results; 5) After identifying that there is sufficient label data, the label is sent to the edge side database for updating the edge side database.

Images