Industrial equipment health monitoring and fault early warning system based on edge computing

Abstract

The invention discloses an industrial equipment health monitoring and fault early warning system based on edge computing, belongs to the technical field of edge computing, and aims to solve the technical problem of how to efficiently realize health monitoring and fault early warning of equipment in real time and identify different types of faults. Each edge computing node is used for taking local data as input, predicting a fault category and a probability value corresponding to the fault category through a local model, outputting the fault category and the probability value as a prediction result, encrypting the local data and the prediction result through the cloud edge collaboration module and then sending the local data and the prediction result to the central server layer, the fusion model comprises three fault diagnosis models, namely a first fault model constructed based on LSTM, a second fault analysis model constructed based on a rule expert model and a third fault analysis model constructed based on a physical model; and the central server layer is used for fusing prediction results based on a self-adaptive weight mechanism.

Description

Technical Field

[0001] The present invention relates to the field of edge computing technology, and specifically to an industrial equipment health monitoring and fault warning system based on edge computing. Background Art

[0002] Modern industrial production is highly dependent on the health status of equipment. However, traditional monitoring methods have problems such as slow response and poor accuracy, and cannot meet the needs of massive data and real-time warnings. Especially for key equipment, failures can lead to high downtime losses. Existing industrial equipment health monitoring systems mainly use single models, such as LSTM models based on time series analysis, rule-based expert models, or fault diagnosis models based on physical models. These models have the following shortcomings:

[0003] 1. LSTM model: It is difficult to effectively identify the complex changes in the equipment's operating status and different fault types, and the prediction accuracy is limited;

[0004] 2. Expert model: It relies on expert experience and domain knowledge, is difficult to adapt to the dynamic changes of equipment operation status, and has weak generalization ability;

[0005] 3. Physical model: It is necessary to establish an accurate physical model, and it is difficult to cover all types of faults. The cost of model establishment and maintenance is high.

[0006] How to efficiently and real-time implement equipment health monitoring and fault warning, and identify different types of faults, is a technical problem that needs to be solved. Summary of the invention

[0007] The technical task of the present invention is to address the above shortcomings and provide an industrial equipment health monitoring and fault warning system based on edge computing to solve the technical problem of how to efficiently and real-time realize equipment health monitoring and fault warning, and identify different types of faults.

[0008] In a first aspect, the present invention provides an industrial equipment health monitoring and fault warning system based on edge computing, including an edge device layer, an edge computing node layer, a cloud-edge collaboration module, and a central server layer;

[0009] The edge device layer includes multiple edge devices, which are used to collect monitoring data of industrial equipment in multiple dimensions through edge devices, and construct time series data based on the collected monitoring data and upload it to the edge computing node layer;

[0010] The edge computing node layer includes multiple edge computing nodes, each edge computing node receives time series data collected by the corresponding edge device, pre-processes the time series data, uses the pre-processed data as local data, and uses the fusion model issued by the central server layer as the local model. Each edge computing node is used to take local data as input, predict the fault category and the probability value corresponding to the fault category through the local model as the prediction result output, and encrypt the local data and the prediction result through the cloud-edge collaboration module and send them to the central server layer, wherein the fusion model includes three fault diagnosis models, namely, a first fault model constructed based on LSTM, a second fault analysis model constructed based on a rule-based expert model, and a third fault analysis model constructed based on a physical model;

[0011] For each edge computing node, the central server layer is used to fuse the prediction results output by the first fault diagnosis model, the second fault diagnosis model and the third fault diagnosis model based on an adaptive weight mechanism to obtain the fused prediction results, and perform model training on each fault diagnosis model based on the real results corresponding to the local data uploaded by the edge computing node and the fused prediction results to obtain the updated model parameters of each fault diagnosis model, and encrypt the updated model parameters of each fault diagnosis model through the cloud-edge collaboration module and send them to the edge computing node. Correspondingly, each edge computing node updates the local model based on the updated model parameters;

[0012] For various monitoring data and various fault categories, the central server layer is configured with thresholds corresponding to each health data and each fault category, and is also configured with risk levels and alarm mechanisms. The central server layer is used to divide risk levels and issue risk warnings based on local data uploaded by each edge node, prediction results after fusion, thresholds corresponding to various monitoring data, and thresholds corresponding to various fault categories, combined with risk levels and alarm mechanisms, and push risk warnings to corresponding operation and maintenance personnel.

[0013] Preferably, each edge device collects monitoring data periodically at a predetermined collection frequency, and constructs time series data based on the monitoring data within the period and uploads it to the corresponding edge computing node.

[0014] Preferably, each edge computing node is used to perform the following preprocessing of the time series data:

[0015] Identify and remove noise and outliers in time series data based on statistical methods and machine learning algorithms to obtain cleaned data;

[0016] Smoothing the cleaned data based on Fourier transform, sliding window averaging method or low-pass filter to obtain smoothed data;

[0017] The smoothed data is averaged based on the moving average method to remove random noise and obtain denoised data;

[0018] The denoised data is classified by cluster analysis method, the denoised data is clustered into different clusters, and the data far away from the cluster is removed as outliers to obtain the secondary cleaned data;

[0019] The feature extraction is performed on the data after secondary cleaning to obtain the time domain features, frequency domain features and statistical features, and the extracted features are processed in combination with the wavelet transform algorithm, and the wavelet transformed features are used as preprocessed data.

[0020] Preferably, for the fusion model, the first fault model includes an LSTM layer, a fully connected layer and an output layer, the LSTM layer includes a plurality of LSTM units, each LSTM unit includes an input gate, a forget gate and an output gate, and is used to capture long-term dependencies, the fully connected layer is used to map the output of the LSTM layer to a prediction result, and the output layer is used to output the prediction result, and the prediction result includes a fault category and a probability value of the fault category;

[0021] The expert model is used to judge the fault category based on the characteristics of a specific fault mode. The expert model includes a rule base, an inference engine and a knowledge update mechanism. The rule base stores expert-defined diagnostic rules. The inference engine is used to perform inference based on monitoring data and the rule base, and output the fault category and the probability value of the fault category as the fault diagnosis result. The knowledge update mechanism is used to update the rule base based on the expert's new experience, new monitoring data and the corresponding fault diagnosis result.

[0022] The physical model is used to make judgments based on monitoring data and the operating status of the target device through the constructed physical equations, and output fault categories and probability values ​​of the fault categories.

[0023] Preferably, the central server layer is used to fuse the prediction results output by the first fault diagnosis model, the second fault diagnosis model and the third fault diagnosis model based on an adaptive weight mechanism, including the following operations:

[0024] Assign a weight to each fault diagnosis model, and for the probability of each fault category, adjust the probability value of the fault category by weighted average method based on the weight of each fault diagnosis model and the probability value of the predicted output of each fault diagnosis model;

[0025] Based on the adjusted probability value of each fault category, the fault category with the highest probability is selected for each fault diagnosis model through a voting mechanism;

[0026] For each fault diagnosis model, based on the probability of the fault category corresponding to the fault diagnosis model and the prior probability value, the post-verification probability is updated through Bayesian fusion to obtain the posterior probability value;

[0027] The final probability value is obtained by performing statistical calculation on the posterior probability value corresponding to each fault diagnosis model, wherein the statistical calculation includes calculating the average value and calculating the maximum value.

[0028] Preferably, for each type of monitoring data, the central server layer is used to adjust the threshold value corresponding to the monitoring data based on a sliding window technology.

[0029] Preferably, the central server layer is used to perform trend analysis based on the local data and prediction results uploaded by each edge computing node and the fused prediction results generated by the central server layer itself, to regularly generate health check reports, and to display the health check reports, the local data and prediction results uploaded by each edge computing node, and the fused prediction results generated by the central server layer itself through a visual interface.

[0030] Preferably, the central server layer is further used to collect user feedback and device maintenance history records of the target device, and perform model training and update on the local model of the edge computing node based on the user feedback and device maintenance history records;

[0031] The user feedback and the equipment maintenance history records are both timestamped. The user feedback includes the user's rating of the alarm, and the equipment maintenance history records include the equipment operation data and alarm records of the target equipment.

[0032] Preferably, when training and updating the local model of the edge computing node based on user feedback and device maintenance history, the central server layer is used to perform the following:

[0033] Collect user feedback through a visual interface, including regular questionnaires;

[0034] The equipment maintenance history records of all equipment are stored in the maintenance log database, and the data in the equipment maintenance history records are standardized. The equipment maintenance history records are pushed to the maintenance log database by the edge device computing node layer through the message queue;

[0035] Align the timestamps of user feedback with the equipment maintenance history records for the corresponding time period;

[0036] Extract key features from equipment maintenance history, including maintenance frequency, mean time to repair, and fault type distribution;

[0037] Analyze the correlation between user feedback, alarm records and maintenance activities;

[0038] Based on the association relationship, the actual operation status of the equipment, and the last local data of the edge computing node, the local model of the edge computing node is trained and updated.

[0039] Preferably, the central server layer is also used to analyze the health status of edge devices based on local data collected by edge computing nodes and the actual operation of the devices, generate edge device maintenance and calibration reports based on the analysis results, and push the edge device maintenance and calibration reports to relevant operation and maintenance personnel.

[0040] The industrial equipment health monitoring and fault warning system based on edge computing of the present invention has the following advantages:

[0041] 1. Through edge devices, the monitoring data of industrial equipment is collected in multiple dimensions. Based on the monitoring data, the local model deployed on the edge computing node is used to analyze the faults of the industrial equipment, and the fault categories of the industrial equipment are predicted. Based on the monitoring data, the predicted fault categories, each health data, and the threshold corresponding to each fault category, the risk level and the alarm mechanism configured at the central server layer are used to classify the risk level and conduct risk warning, thereby achieving real-time monitoring, fault identification and warning of industrial equipment, avoiding major faults and improving equipment safety.

[0042] 2. Fault identification is performed through the local model deployed on the edge computing node, model training is performed on the local model of each edge computing node through the central server layer, and risk classification and alarm are performed through the central server layer, thus realizing cloud-edge collaborative computing, improving computing efficiency, and reducing computing burden;

[0043] 3. The local model is an adaptive fusion model that can effectively identify different types of faults and improve the model's adaptability to different equipment types and operating modes. It uses adaptive multi-model fusion technology to learn equipment operating modes, identify potential faults, and predict the time and type of faults, providing early warning for equipment maintenance and reducing maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] The present invention is further described below in conjunction with the accompanying drawings.

[0046] Figure 1 The present invention is a structural block diagram of an industrial equipment health monitoring and fault warning system based on edge computing. DETAILED DESCRIPTION

[0047] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it. However, the embodiments are not intended to limit the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments may be combined with each other.

[0048] The embodiment of the present invention provides an industrial equipment health monitoring and fault warning system based on edge computing, which is used to solve the technical problem of how to efficiently and real-time realize equipment health monitoring and fault warning, and identify different types of faults.

[0049] Example:

[0050] The present invention discloses an industrial equipment health monitoring and fault warning system based on edge computing, comprising an edge device layer, an edge computing node layer, a cloud-edge collaboration module and a central server layer.

[0051] The edge device layer includes multiple edge devices, which are used to collect monitoring data of industrial equipment in multiple dimensions through edge devices, and construct time series data based on the collected monitoring data and upload it to the edge computing node layer.

[0052] In this embodiment, each edge device periodically collects monitoring data according to a predetermined collection frequency, and constructs time series data based on the monitoring data within the period and uploads it to the corresponding edge computing node.

[0053] The edge devices in the edge device layer use a variety of sensors to form a sensor network. The key parts of industrial equipment are installed with temperature, vibration, current, pressure and other sensors to collect multi-dimensional data in real time and comprehensively monitor the equipment status. The edge device layer also has a data acquisition module that combines centralized and distributed architectures. The data acquisition module is connected to the sensor to achieve high-frequency data acquisition and processing, and supports multiple communication protocols.

[0054] Multidimensional data can provide more comprehensive equipment health information, reduce monitoring blind spots caused by single sensor failure, and improve prediction accuracy. Select appropriate sensors for installation according to equipment type and key components, set appropriate sampling frequency, obtain equipment data regularly and frequently, and ensure real-time and comprehensive collection.

[0055] The edge computing node layer includes multiple edge computing nodes. Each edge computing node receives the time series data collected by the corresponding edge device, pre-processes the time series data, uses the pre-processed data as local data, and uses the fusion model sent by the central server layer as the local model. Each edge computing node is used to take local data as input, predict the fault category and the probability value corresponding to the fault category through the local model as the output of the prediction result, and encrypt the local data and the prediction result through the cloud-edge collaborative module and send them to the central server layer. Among them, the fusion model includes three fault diagnosis models, namely the first fault model built based on LSTM, the second fault analysis model built based on the rule-based expert model, and the third fault analysis model built based on the physical model.

[0056] As a specific implementation of data preprocessing, each edge computing node is used to perform the following preprocessing of time series data:

[0057] (1) Identify and remove noise and outliers in time series data based on statistical methods and machine learning algorithms to obtain cleaned data;

[0058] (2) smoothing the cleaned data based on Fourier transform, sliding window averaging method or low-pass filter to obtain smoothed data;

[0059] (3) Averaging the smoothed data based on the moving average method to remove random noise and obtain denoised data;

[0060] (4) Classify the denoised data by clustering analysis, cluster the denoised data into different clusters, remove the data far away from the cluster as outliers, and obtain the secondary cleaned data;

[0061] (5) Extract features from the data after secondary cleaning to obtain time domain features, frequency domain features, and statistical features, and process the extracted features using a wavelet transform algorithm, using the wavelet transformed features as preprocessed data.

[0062] In the data preprocessing process described above in this embodiment, statistical methods and machine learning algorithms are used to identify and remove noise and outliers, and data smoothing is performed through Fourier transform, sliding window averaging or low-pass filtering techniques to improve data availability. The moving average method is used to smooth time series data, average the data within a period of time, and remove random noise. Cluster analysis clusters data points into different clusters, and data points far away from other clusters are regarded as outliers to remove outliers. Time domain features, frequency domain features and statistical features (such as mean, variance, kurtosis, etc.) are extracted, and combined with algorithms such as wavelet transform, reliable feature results are obtained to improve the sensitivity of the model to different fault modes.

[0063] As a specific implementation of model training and updating, training data is prepared before training: the adaptive multi-model fusion model is trained using historical data and actual fault data, and the model is updated according to the equipment operating status and fault type.

[0064] The model training process is as follows:

[0065] 1) Data preparation: preprocess the collected sensor data, including data cleaning and feature extraction, to generate input data suitable for model training;

[0066] 2) Model construction: Build an adaptive multi-model fusion framework and set model parameters, such as the number of LSTM layers, number of neurons, learning rate, etc.;

[0067] 3) Model training: Use historical data to train each model and use an adaptive weight mechanism to adjust the model weights;

[0068] 4) Model evaluation: Use the test data set to evaluate the prediction performance of the model, such as the prediction accuracy, recall rate and other indicators of the model;

[0069] 5) Model optimization: Adjust model parameters or use other optimization methods based on the evaluation results to further improve the prediction accuracy of the model.

[0070] Model update mechanism: According to the equipment operating status and fault type, new data is collected regularly and the adaptive multi-model fusion model is retrained to ensure that the model's prediction accuracy always remains at a high level.

[0071] As a specific implementation of the first fault model, the first fault model includes an LSTM layer, a fully connected layer and an output layer. The LSTM layer includes multiple LSTM units, each LSTM unit contains an input gate, a forget gate and an output gate, which are used to capture long-term dependencies. The fully connected layer is used to map the output of the LSTM layer to a prediction result, and the output layer is used to output the prediction result, which includes a fault category and a probability value of the fault category.

[0072] The formula of the first fault model is expressed as:

[0073] h_t=σ(W_hh*h_{t-1}+W_xh*x_t+b_h),

[0074] Among them, h_t: hidden state at time t, h_{t-1}: hidden state at time t-1

[0075] x_t: input at time t, W_hh: weight moment from hidden state to hidden state, W_xh: weight matrix from input to hidden state, b_h: bias vector of hidden state, σ: activation function.

[0076] The expert model is used to judge the fault category based on the characteristics of a specific fault mode. The expert model includes a rule base, an inference engine and a knowledge update mechanism. The rule base stores expert-defined diagnostic rules. The inference engine is used to perform inference based on monitoring data and the rule base, and output the fault category and the probability value of the fault category as the fault diagnosis result. The knowledge update mechanism is used to update the rule base based on the expert's new experience, new monitoring data and corresponding fault diagnosis results.

[0077] The physical model is used to make judgments based on the monitoring data and the operating status of the target equipment through the constructed physical equations, and output the fault category and the probability value of the fault category.

[0078] In this embodiment, a physical model of the device is used to establish a fault diagnosis model based on the physical model, for example, judging based on the physical parameters and operating status of the device. The implementation of the physical model includes:

[0079] (1) Physical equations: Establish mathematical equations that describe the operating mechanism of the equipment, such as heat conduction equations, stress-strain equations, etc.

[0080] (2) Parameter estimation: using real-time data to estimate key parameters in the physical model;

[0081] (3) State prediction: predicting the future state of the equipment based on physical equations and estimated parameters;

[0082] (4) Deviation analysis: Compare the predicted state with the actual state and detect abnormal situations.

[0083] For each edge computing node, the central server layer is used to fuse the prediction results output by the first fault diagnosis model, the second fault diagnosis model and the third fault diagnosis model based on an adaptive weight mechanism to obtain the fused prediction results, and perform model training on each fault diagnosis model based on the real results corresponding to the local data uploaded by the edge computing node and the fused prediction results to obtain the updated model parameters of each fault diagnosis model, and encrypt the updated model parameters of each fault diagnosis model through the cloud-edge collaborative module and send them to the edge computing node. Correspondingly, each edge computing node updates the local model based on the updated model parameters.

[0084] As a specific implementation, the central server layer is used to fuse the prediction results output by the first fault diagnosis model, the second fault diagnosis model, and the third fault diagnosis model based on the adaptive weight mechanism, including the following operations:

[0085] (1) Assigning a weight to each fault diagnosis model. For the probability of each fault category, the probability value of the fault category is adjusted by weighted average method based on the weight of each fault diagnosis model and the probability value of the predicted output of each fault diagnosis model;

[0086] (2) Based on the adjusted probability value of each fault category, the fault category with the highest probability is selected for each fault diagnosis model through a voting mechanism;

[0087] (3) For each fault diagnosis model, based on the probability of the fault category corresponding to the fault diagnosis model and the prior probability value, the posterior probability is updated through Bayesian fusion to obtain the posterior probability value;

[0088] (4) The final probability value is obtained by performing statistical calculation on the posterior probability value corresponding to each fault diagnosis model, wherein the statistical calculation includes calculating the average value and calculating the maximum value.

[0089] The specific example of fusing the prediction results of the three fault diagnosis models is as follows:

[0090] Assume there are three models, Model A, B, and C, each predicting the probability of a specific failure as:

[0091] Model A: overheating fault: 0.8, mechanical fault: 0.1, electrical fault: 0.05, insufficient lubrication: 0.05;

[0092] Model B: overheating fault: 0.7, mechanical fault: 0.2, electrical fault: 0.1, insufficient lubrication: 0.0;

[0093] Model C: Overheating failure: 0.9, Mechanical failure: 0.05, Electrical failure: 0.03, Insufficient lubrication: 0.02.

[0094] The weights assigned to each model are set as follows:

[0095] The weight of model A is 0.5, the weight of model B is 0.3, and the weight of model C is 0.2. The weighted average probability calculation formula of overheating failure is as follows:

[0096]

[0097] The weighted average probability of mechanical failure is calculated as follows:

[0098]

[0099] The weighted average probability of electrical failure is calculated as follows:

[0100]

[0101] The weighted average probability of insufficient lubrication is calculated as follows:

[0102]

[0103] The calculation results are overheating failure: 0.79 - mechanical failure: 0.12 - electrical failure: 0.061 - insufficient lubrication: 0.029.

[0104] Using the voting mechanism, each model will select the fault type with the highest probability. The judgment steps are as follows:

[0105] Model A: Overheat fault (0.8),

[0106] Model B: Overheat fault (0.7),

[0107] Model C: Overheat fault (0.9),

[0108] The voting results are as follows:

[0109] Overheating fault: 3 tickets,

[0110] Mechanical failure: 0 tickets,

[0111] Electrical failure: 0 tickets,

[0112] Insufficient lubrication: 0 votes,

[0113] Final result: Overheat failure (win).

[0114] Using Bayesian fusion, the probability of each model is combined with the prior probability to update the posterior probability.

[0115] Input values: fault type, probability, and prior probability, where the probability is the probability of each model for the fault type, and the prior probability is the previous statistics for each fault type without any evidence. The prior probability (based on historical data) is set as follows: overheating fault: 0.5, mechanical fault: 0.3, electrical fault: 0.1, insufficient lubrication: 0.1, for each fault type, use the Bayesian formula to update the posterior probability:

[0116]

[0117] Among them, P(Evidence|FaultTypei) is the model's predicted probability of the fault type (predicted evidence), corresponding to the output of the model, P(FaultTypei) is the prior probability of the fault type, and P(Evidence) is the sum of evidence under all fault types. The calculation results are: overheating fault: 0.901, mechanical fault: 0.067, electrical fault: 0.011, insufficient lubrication: 0.011.

[0118] An easy way to combine these results is to use their average or maximum value.

[0119] P final =w1·P vote +w2·Paug +w3·P bayesian

[0120] Among them, w1, w2, w3 are the weights of each strategy, reflecting the degree of trust in them. For example, w1 = 0.5, w2 = 0.3, w3 = 0.2. For example, the fault type is overheating fault:

[0121] P final (Overheating)=w1·P(Voting)+w2·P(Average)+w3·P(Bayesian)

[0122] The final overheat failure probability is 0.5+0.3*0.79+0.2*0.901=0.9172.

[0123] For various monitoring data and various fault categories, the central server layer is configured with thresholds corresponding to each health data and each fault category, and is also configured with risk levels and alarm mechanisms. The central server layer is used to divide risk levels and issue risk warnings based on local data uploaded by each edge node, prediction results after fusion, thresholds corresponding to various monitoring data, and thresholds corresponding to various fault categories, combined with risk levels and alarm mechanisms, and push risk warnings to corresponding operation and maintenance personnel.

[0124] In this embodiment, for each type of monitoring data, the central server layer is used to adjust the threshold corresponding to the monitoring data based on the sliding window technology.

[0125] This embodiment adopts an adaptive threshold and alarm mechanism in risk classification and early warning.

[0126] Intelligent threshold adjustment: Based on the characteristics of real-time data streams, the alarm threshold is automatically adjusted through algorithms to achieve higher flexibility. This mechanism is based on incremental learning, which enables the system to continuously learn and optimize threshold settings during operation. Specific implementations include:

[0127] Sliding window technique: Use a fixed-size time window of 30 days to calculate the statistical characteristics of recent data.

[0128] Dynamic threshold calculation: Based on the data distribution within the sliding window, the threshold is dynamically calculated using the 3σ principle.

[0129] Seasonal adjustment: Use the STL algorithm to decompose the time series, consider the seasonal changes in equipment operation, and adjust the threshold accordingly.

[0130] Incremental learning: Evaluate system performance weekly and use gradient descent to automatically adjust threshold calculation parameters to ensure that the threshold is consistent with the current operating status.

[0131] Multi-level alarm mechanism: Generate multi-level alarms for faults based on risk levels, and clarify the processing procedures for each level of alarms so that operation and maintenance personnel can make decisions quickly to ensure that faults can be quickly contained. Specific implementation includes:

[0132] Risk level classification: The failure risk is divided into four levels: low, medium, high and emergency.

[0133] Alarm trigger conditions: Set corresponding trigger conditions according to different risk levels, such as the degree and duration of data deviation from the mean.

[0134] Alarm escalation mechanism: If a low-level alarm is not handled in a timely manner, it will automatically escalate to a higher-level alarm.

[0135] Process definition: Establish clear process and response time requirements for each alarm level, ranging from 4 hours to 5 minutes.

[0136] With this approach, the system can dynamically adjust the alarm threshold according to the actual operating conditions of the equipment, while ensuring appropriate responses to abnormal situations of varying severity through a multi-level alarm mechanism, improving the accuracy and timeliness of fault detection.

[0137] As an improvement to the system, the central server layer is used to perform trend analysis based on the local data and prediction results uploaded by each edge computing node and the fused prediction results generated by the central server layer itself, regularly generate health check reports, and display the health check reports, the local data and prediction results uploaded by each edge computing node, and the fused prediction results generated by the central server layer itself through a visual interface.

[0138] Visual interface design: Design a user-friendly interface that displays monitoring data such as temperature and vibration, as well as health indexes, through a dynamic dashboard, providing an intuitive visual experience and allowing operators to obtain equipment status immediately.

[0139] Trend analysis and reporting: The system regularly generates health check reports, comprehensively analyzes historical data and current status, and assists decision makers in formulating maintenance strategies. It uses charts and data visualization methods to provide clear decision-making basis.

[0140] Regular health reports: Generate health check reports regularly, summarize health indexes, fault prediction results and historical data analysis, and provide scientific and effective maintenance recommendations for decision makers.

[0141] As an improvement to the system, the central server layer is also used to collect user feedback and equipment maintenance history records of the target equipment, and to train and update the local model of the edge computing node based on the user feedback and equipment maintenance history records; wherein, user feedback and equipment maintenance history records are both timestamped, user feedback includes the user's rating of the alarm, and equipment maintenance history records include the equipment operation data and alarm records of the target equipment.

[0142] As a specific implementation, when training and updating the local model of the edge computing node based on user feedback and device maintenance history, the central server layer is used to perform the following:

[0143] (1) Collect user feedback through a visual interface, including regular questionnaires;

[0144] (2) storing the equipment maintenance history records of all equipment through a maintenance log database, and performing data standardization processing on the data in the equipment maintenance history records, wherein the equipment maintenance history records are pushed to the maintenance log database by the edge device computing node layer through a message queue;

[0145] (3) aligning the timestamps of user feedback with the equipment maintenance history records for the corresponding time period;

[0146] (4) Extract key features from equipment maintenance history records, including maintenance frequency, mean repair time, and fault type distribution;

[0147] (5) Analyze the correlation between user feedback, alarm records, and maintenance activities;

[0148] (6) Based on the association relationship, the actual operation status of the equipment, and the last local data of the edge computing node, the local model of the edge computing node is trained and updated.

[0149] The above feedback mechanism integrates user operation feedback and equipment maintenance history as the basis for iterative optimization. This includes not only real-time feedback of the algorithm, but also user experience feedback during use. The implementation process of the feedback mechanism is as follows:

[0150] 1. User operation feedback collection:

[0151] Design an interactive feedback interface: embed quick feedback buttons in the system interface to allow users to rate the accuracy of the alarm (1-5 stars);

[0152] Real-time feedback capture: Use event listeners to record the user's operation sequence and timestamps when handling alarms;

[0153] Regular questionnaire surveys: A short questionnaire is pushed through the system every month to collect users' evaluation of the overall performance of the system.

[0154] 2. Equipment maintenance historical data integration:

[0155] Establish a unified maintenance log database: Use a relational database to store all maintenance activity records;

[0156] Data standardization: Establish unified maintenance activity coding standards to ensure that maintenance data from different sources can be consistently integrated;

[0157] Real-time synchronization mechanism: Use message queues to synchronize maintenance data to the central database in real time.

[0158] 3. Data association analysis

[0159] Time series alignment: align the timestamps of user feedback with the equipment operation data and alarm records of the corresponding time period;

[0160] Feature extraction: Extract key features from maintenance history, such as maintenance frequency, mean repair time, fault type distribution, etc.

[0161] Association rule mining: Use algorithms to analyze the associations between user feedback, alarm records, and maintenance activities;

[0162] Adaptive algorithm update: Regularly retrain and update the model based on new data collected and actual operation conditions.

[0163] As an improvement of this embodiment, the central server layer is also used to analyze the health status of the edge device based on the local data collected by the edge computing node and the actual operation of the device, and generate an edge device maintenance and calibration report based on the analysis results, and push the edge device maintenance and calibration report to relevant operation and maintenance personnel.

[0164] The above improvements enable sensor calibration and system auditing.

[0165] Sensor calibration: Calibrate sensors regularly to ensure the accuracy of collected data. Any deviation will affect subsequent analysis and judgment.

[0166] System review: Regularly review the monitoring system to ensure that all components are working properly and faults are detected and handled in a timely manner.

[0167] The present invention has been shown and described in detail above through the accompanying drawings and preferred embodiments. However, the present invention is not limited to these disclosed embodiments. Based on the above multiple embodiments, those skilled in the art can know that the means in the above different embodiments can be combined to obtain more embodiments of the present invention, and these embodiments are also within the protection scope of the present invention.

Claims

1. An industrial equipment health monitoring and fault warning system based on edge computing, characterized in that: It includes edge device layer, edge computing node layer, cloud-edge collaboration module and central server layer; The edge device layer includes multiple edge devices, which are used to collect monitoring data of industrial equipment in multiple dimensions through edge devices, and construct time series data based on the collected monitoring data and upload it to the edge computing node layer; The edge computing node layer includes multiple edge computing nodes, each edge computing node receives time series data collected by the corresponding edge device, pre-processes the time series data, uses the pre-processed data as local data, and uses the fusion model issued by the central server layer as the local model. Each edge computing node is used to take local data as input, predict the fault category and the probability value corresponding to the fault category through the local model as the prediction result output, and encrypt the local data and the prediction result through the cloud-edge collaboration module and send them to the central server layer, wherein the fusion model includes three fault diagnosis models, namely, a first fault model constructed based on LSTM, a second fault analysis model constructed based on a rule-based expert model, and a third fault analysis model constructed based on a physical model; For each edge computing node, the central server layer is used to fuse the prediction results output by the first fault diagnosis model, the second fault diagnosis model and the third fault diagnosis model based on an adaptive weight mechanism to obtain the fused prediction results, and perform model training on each fault diagnosis model based on the real results corresponding to the local data uploaded by the edge computing node and the fused prediction results to obtain the updated model parameters of each fault diagnosis model, and encrypt the updated model parameters of each fault diagnosis model through the cloud-edge collaboration module and send them to the edge computing node. Correspondingly, each edge computing node updates the local model based on the updated model parameters; For various monitoring data and various fault categories, the central server layer is configured with thresholds corresponding to each health data and each fault category, and is also configured with risk levels and alarm mechanisms. The central server layer is used to divide risk levels and issue risk warnings based on local data uploaded by each edge node, prediction results after fusion, thresholds corresponding to various monitoring data, and thresholds corresponding to various fault categories, combined with risk levels and alarm mechanisms, and push risk warnings to corresponding operation and maintenance personnel.

2. The industrial equipment health monitoring and fault warning system based on edge computing according to claim 1 is characterized in that: Each edge device collects monitoring data periodically according to a predetermined collection frequency, and builds time series data based on the monitoring data within the period and uploads it to the corresponding edge computing node.

3. The industrial equipment health monitoring and fault warning system based on edge computing according to claim 1 is characterized in that: Each edge computing node is used to perform the following preprocessing of time series data: Identify and remove noise and outliers in time series data based on statistical methods and machine learning algorithms to obtain cleaned data; Smoothing the cleaned data based on Fourier transform, sliding window averaging method or low-pass filter to obtain smoothed data; The smoothed data is averaged based on the moving average method to remove random noise and obtain denoised data; The denoised data is classified by cluster analysis method, the denoised data is clustered into different clusters, and the data far away from the cluster is removed as outliers to obtain the secondary cleaned data; The feature extraction is performed on the data after secondary cleaning to obtain the time domain features, frequency domain features and statistical features, and the extracted features are processed in combination with the wavelet transform algorithm, and the features after wavelet transform are used as preprocessed data.

4. The industrial equipment health monitoring and fault warning system based on edge computing according to claim 1 is characterized in that: For the fusion model, the first fault model includes an LSTM layer, a fully connected layer and an output layer, the LSTM layer includes multiple LSTM units, each LSTM unit includes an input gate, a forget gate and an output gate, which are used to capture long-term dependencies, the fully connected layer is used to map the output of the LSTM layer to a prediction result, and the output layer is used to output the prediction result, which includes a fault category and a probability value of the fault category; The expert model is used to judge the fault category based on the characteristics of a specific fault mode. The expert model includes a rule base, an inference engine and a knowledge update mechanism. The rule base stores expert-defined diagnostic rules. The inference engine is used to perform inference based on monitoring data and the rule base, and output the fault category and the probability value of the fault category as the fault diagnosis result. The knowledge update mechanism is used to update the rule base based on the expert's new experience, new monitoring data and the corresponding fault diagnosis result. The physical model is used to make judgments based on monitoring data and the operating status of the target device through the constructed physical equations, and output fault categories and probability values ​​of the fault categories.

5. The industrial equipment health monitoring and fault warning system based on edge computing according to claim 1 is characterized in that: The central server layer is used to fuse the prediction results output by the first fault diagnosis model, the second fault diagnosis model, and the third fault diagnosis model based on an adaptive weight mechanism, including the following operations: Assign a weight to each fault diagnosis model, and for the probability of each fault category, adjust the probability value of the fault category by weighted average method based on the weight of each fault diagnosis model and the probability value of the predicted output of each fault diagnosis model; Based on the adjusted probability value of each fault category, the fault category with the highest probability is selected for each fault diagnosis model through a voting mechanism; For each fault diagnosis model, based on the probability of the fault category corresponding to the fault diagnosis model and the prior probability value, the posterior probability is updated through Bayesian fusion to obtain the posterior probability value; The final probability value is obtained by performing statistical calculation on the posterior probability value corresponding to each fault diagnosis model, wherein the statistical calculation includes calculating the average value and calculating the maximum value.

6. The industrial equipment health monitoring and fault warning system based on edge computing according to claim 1 is characterized in that: For each type of monitoring data, the central server layer is used to adjust the threshold corresponding to the monitoring data based on the sliding window technology.

7. The industrial equipment health monitoring and fault warning system based on edge computing according to claim 1 is characterized in that: The central server layer is used to perform trend analysis based on the local data and prediction results uploaded by each edge computing node and the fused prediction results generated by the central server layer itself, regularly generate health check reports, and display the health check reports, the local data and prediction results uploaded by each edge computing node, and the fused prediction results generated by the central server layer itself through a visual interface.

8. The edge computing-based industrial equipment health monitoring and fault warning system according to claim 1 is characterized in that: The central server layer is also used to collect user feedback and device maintenance history records of target devices, and perform model training and update on the local model of the edge computing node based on the user feedback and device maintenance history records; The user feedback and the equipment maintenance history records are both timestamped. The user feedback includes the user's rating of the alarm, and the equipment maintenance history records include the equipment operation data and alarm records of the target equipment.

9. The edge computing-based industrial equipment health monitoring and fault warning system according to claim 8 is characterized in that: When training and updating the local models of edge computing nodes based on user feedback and device maintenance history, the central server layer is used to perform the following: Collect user feedback through a visual interface, including regular questionnaires; The equipment maintenance history records of all equipment are stored in the maintenance log database, and the data in the equipment maintenance history records are standardized. The equipment maintenance history records are pushed to the maintenance log database by the edge device computing node layer through the message queue; Align the timestamps of user feedback with the equipment maintenance history records for the corresponding time period; Extract key features from equipment maintenance history, including maintenance frequency, mean time to repair, and fault type distribution; Analyze the correlation between user feedback, alarm records and maintenance activities; Based on the association relationship, the actual operation status of the equipment, and the last local data of the edge computing node, the local model of the edge computing node is trained and updated.

10. The edge computing-based industrial equipment health monitoring and fault warning system according to claim 1 is characterized in that: The central server layer is also used to analyze the health status of edge devices based on local data collected by edge computing nodes and the actual operation of the devices, generate edge device maintenance and calibration reports based on the analysis results, and push the edge device maintenance and calibration reports to relevant operation and maintenance personnel.

Images