Equipment health examination method and system based on multi-agent cooperation

Abstract

The invention provides an equipment health examination method and system based on multi-agent collaboration. The method comprises the following steps: a data acquisition step: acquiring target data to generate a multi-dimensional feature vector; the target data comprises frequency data, process data and control data; an equipment monitoring step: monitoring the equipment based on the generated multi-dimensional feature vector, and outputting a monitoring result; and a deployment processing step: deploying and implementing a corresponding processing scheme according to a monitoring result. The invention provides an equipment early warning and diagnosis method and system based on multi-agent cooperation, and the method comprises the steps: combining the advantages of a professional mechanism model, an AI model and a large language model through the multi-agent cooperation of feature extraction, threshold adjustment, fault diagnosis, decision generation, summary review and closed-loop feedback; the problems that a traditional method is single in model, high in false alarm rate, lagged in decision and the like are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment health management. Specifically, it relates to a method and system for equipment health inspection based on multi-agent collaboration, and in particular, a method and system for equipment early warning and diagnosis based on multi-agent collaboration. Background Art

[0002] Equipment Prognostics and Health Management (PHM) is a key technology for improving equipment reliability and reducing operation and maintenance costs in the industrial field. The following problems exist in traditional equipment early warning and diagnosis methods:

[0003] 1. Single model: Existing technologies mostly rely on a single mechanism model or data-driven model. Although the mechanism model has strong interpretability, it is difficult to handle complex non-linear faults; the data-driven model relies on a large amount of labeled data and has insufficient generalization ability.

[0004] 2. Poor real-time performance: Most systems adopt an offline or lagged analysis mode and cannot dynamically adjust thresholds and rules, resulting in lagged anomaly detection.

[0005] 3. High false alarm rate: Traditional threshold rules are fixed and do not combine with the dynamic working conditions of the equipment, making it easy to generate false alarms.

[0006] 4. Coarse maintenance strategy: Fault diagnosis and maintenance decision-making are disjointed, relying on manual experience and lacking intelligent closed-loop management.

[0007] In recent years, multi-agent technology has provided new ideas for equipment PHM. However, the collaborative efficiency of agents in existing solutions is low, and the advantages of mechanism models, AI models, and large models (LLMs) have not been effectively integrated. Therefore, there is an urgent need for a high-precision, real-time, and closed-loop management early warning and diagnosis method.

[0008] Aiming at the deficiencies of the existing technology, the present invention aims to propose a method and system for equipment early warning and diagnosis based on multi-agent collaboration. Through the division of labor and cooperation of multi-agents, combining small models (mechanism model + AI model) with large models (LLM model), it realizes the full-process intelligence of equipment status monitoring, fault early warning, fault diagnosis, and maintenance decision-making.

[0009] In the patent document with the publication number CN112016828A, an industrial equipment health management cloud platform architecture based on streaming big data is disclosed. This platform architecture includes: an access layer for data acquisition to obtain multi-source heterogeneous data; a test layer for testing to obtain quality test results; a primary processing layer for standardization processing to obtain standardized data; a metadata storage layer for optimization processing and storage; a secondary processing layer for correlation analysis and feature analysis to obtain temporal correlation feature data between data; a feature data storage layer for storing the temporal correlation feature data; a data set fusion layer for data integration and fusion; a calculation model layer for constructing a calculation model that conforms to the usage scenario; a diagnosis and prediction layer for selecting a corresponding calculation model for calculation to obtain a calculation result; and a service layer for providing the data and results to the corresponding platform service objects. However, this patent document aims to achieve distributed data acquisition and phased data processing methods to ultimately reduce the development cycle, while this application aims to solve problems such as a single traditional method model, a high false alarm rate, and a lag in decision-making. The technical problems solved by the two are different. Summary of the Invention

[0010] Aiming at the defects in the prior art, the purpose of the present invention is to provide a device health inspection method and system based on multi-agent collaboration.

[0011] According to a device health inspection method based on multi-agent collaboration provided by the present invention, it includes:

[0012] Data acquisition step: Acquire target data to generate a multi-dimensional feature vector;

[0013] The target data includes frequency data, process data, and control data;

[0014] Device monitoring step: Monitor the device based on the generated multi-dimensional feature vector and output a monitoring result;

[0015] Deployment and processing step: Deploy and implement a corresponding processing plan according to the monitoring result.

[0016] Preferably, it further includes:

[0017] Closed-loop feedback step: After the maintenance is implemented, enable the review and summary agent to analyze the maintenance effect, optimize the model parameters and threshold rules, and form a closed-loop iteration.

[0018] Preferably, the data acquisition step includes:

[0019] Collect target data through multi-source sensors and use a feature extraction agent for real-time processing to generate a multi-dimensional feature vector.

[0020] Preferably, the device monitoring step includes:

[0021] Anomaly detection steps: Dynamically adjust the threshold using a threshold / rule agent, and combine machine learning and artificial intelligence models for preliminary anomaly judgment; if the threshold is exceeded, trigger the anomaly warning agent.

[0022] Fault diagnosis steps: Use a diagnosis agent to call a small model to obtain data on fault location, fault type, and fault severity, and verify the diagnosis result through a large model to output a comprehensive diagnosis conclusion; the small model includes a mechanism model and an AI model; the large model includes an LLM model.

[0023] Preferably, the deployment and processing steps include:

[0024] Enable the decision-making agent to generate a maintenance strategy, spare parts deployment plan, and management measures based on the diagnosis result, combined with historical operation and maintenance data, and push them to the execution terminal.

[0025] According to an equipment health inspection system based on multi-agent collaboration provided by the present invention, it includes:

[0026] The sensing layer is used to collect target data and transmit it to the edge computing layer;

[0027] The target data includes frequency data, process data, and control data;

[0028] The edge computing layer is used to receive the target data and generate a multi-dimensional feature vector, and transmit it together with the detected anomaly information to the cloud analysis layer;

[0029] The cloud analysis layer receives the multi-dimensional feature vector and anomaly information, and further analyzes the anomaly situation using a mechanism + AI alarm agent and generates a maintenance instruction;

[0030] The execution layer, based on the maintenance instruction, implements specific maintenance operations through a maintenance execution terminal and feeds back the maintenance implementation result to the feedback layer;

[0031] The feedback layer is used to receive the feedback result for model optimization and knowledge base update.

[0032] Preferably, the sensing layer collects the target data under the operating state of the equipment through multi-source sensors and transmits the collected data to the edge computing layer.

[0033] Preferably, the edge computing layer uses a feature extraction agent to perform real-time processing on the target data to generate a multi-dimensional feature vector, and at the same time, through a threshold / rule alarm agent, combines the dynamically adjusted threshold and an artificial intelligence model for preliminary anomaly detection. If an anomaly is detected, trigger an alarm signal and transmit the multi-dimensional feature vector and anomaly information to the cloud analysis layer.

[0034] Preferably, the cloud analysis layer uses a diagnostic agent to call small models to obtain data on fault location, fault type, and fault severity, and verifies the diagnostic results through a large model to output a comprehensive diagnostic conclusion; the small models include a mechanism model and an AI model; the large model includes an LLM model; enabling the decision-making agent to generate a maintenance strategy, spare part allocation plan, and management measures based on the diagnostic conclusion and historical operation and maintenance data, and sending a maintenance instruction to the execution layer.

[0035] Preferably, the feedback layer analyzes the maintenance effect through a review and summary agent, optimizes the model parameters, threshold rules, and knowledge base, and feeds back the optimized parameters and rules to the edge computing layer and the cloud analysis layer to form a closed-loop iteration.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. The present invention provides a device warning and diagnosis method and system based on multi-agent collaboration. Through the collaboration of multi-agents such as feature extraction, threshold adjustment, fault diagnosis, decision generation, summary review, and closed-loop feedback, and by combining the advantages of professional mechanism models, AI models, and large language models, it solves the problems of single model, high false alarm rate, and decision lag in traditional methods.

[0038] 2. The present invention has characteristics such as dynamic adaptive thresholds, multi-model fusion diagnosis, and intelligent closed-loop optimization, which can significantly improve the accuracy of device fault identification and maintenance efficiency. It can be applied to industrial scenarios with high reliability requirements such as steel, energy, and rail transit, and has good practicality.

[0039] 3. The small models (mechanism model + AI model) adopted by the present invention ensure interpretability and physical consistency, and the large model (LLM model) improves the complex fault identification ability and operation and maintenance decision-making ability. The two cooperate to improve the reliability of diagnosis and decision-making.

[0040] 4. The threshold / rule agent used in the present invention can be dynamically adjusted according to the working conditions and negative samples, reducing the false alarm rate, and the entire process from data collection to maintenance decision-making is automated. Compared with traditional methods, the response speed has increased by more than 90%.

[0041] 5. The decision-making agent used in the present invention combines the device health status, production plan, and cost constraints to generate an optimal maintenance strategy. At the same time, the review and summary agent realizes the autonomous evolution of model parameters and rules through closed-loop feedback.

[0042] Other beneficial effects of the present invention will be elaborated through the introduction of specific technical features and technical solutions in the specific implementation manner. Those skilled in the art should be able to understand the beneficial technical effects brought by the said technical features and technical solutions through these introductions. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Other features, objectives, and advantages of the present invention will become more apparent by reading the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0044] Figure 1 This is the overall system architecture diagram of the present invention.

[0045] Figure 2 This is the multi-agent collaborative flow chart of the present invention.

[0046] Figure 3 This is the schematic diagram of the fusion diagnosis of the small model and the large model of the present invention. Detailed implementation manners

[0047] The present invention will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several changes and improvements can be made. These all belong to the protection scope of the present invention.

[0048] Referring to Figure 2 and Figure 3 shown, a device health check method based on multi-agent collaboration includes:

[0049] Step S1: Data feature acquisition: High-frequency, low-frequency, process, and control data are collected by multi-source sensors and are processed in real time by a feature extraction agent to generate a multi-dimensional feature vector.

[0050] Step S2: Anomaly detection: The threshold / rule agent dynamically adjusts the threshold, and combines machine learning and artificial intelligence models to make a preliminary anomaly judgment; if the threshold is exceeded, the anomaly warning agent is triggered.

[0051] The specific process is as follows:

[0052] 1. Obtain real-time device-related data through data collection, perform data alignment, data cleaning, data imputation, remove outliers and noise data, and perform necessary normalization processing to improve the stability of the algorithm;

[0053] 2. Calculate feature values (mean, effective value, standard deviation, peak value, skewness, etc.) using historical data and current data, set a static initial threshold, or set an initial threshold according to device characteristics and performance limits, industry standards, and international standard specifications;

[0054] 3. Based on the strategy of dynamically adjusting the threshold, one is to adjust or predict the threshold through algorithm-driven (moving average method, machine learning model), and the other is to set conditional logic through a rule engine in combination with scenario requirements and expert knowledge and experience.

[0055] Step S3: Fault Diagnosis: The diagnostic agent invokes small models (mechanism model + AI model) to locate the fault, determine the fault type and severity, and verifies the diagnostic results through a large model (LLM model), and outputs a comprehensive diagnostic conclusion.

[0056] In complex industrial scenarios, the fault diagnosis agent forms an efficient and accurate diagnostic closed-loop through the collaborative location of the mechanism model + AI small model and the verification and optimization of the large language model (LLM). The specific process is as follows:

[0057] 1. Data collection and feature extraction

[0058] Extract vibration signal features (such as root mean square value, kurtosis, spectral peak value) through time-domain / frequency-domain analysis, and construct a basic feature set in combination with the physical characteristics of the equipment (such as bearing model, structural size and rotational speed). Use models such as Transformer or SVM to perform deep feature fusion on multi-source data (vibration, temperature, current, liquid level, pressure and flow, etc.) to capture non-linear relationships.

[0059] 2. Fault location and classification

[0060] Through the initial screening of the mechanism model, quickly locate the position of the abnormal equipment based on the mechanism threshold rule, and narrow the scope of investigation. Use the trained classification model (such as SVM or CNN) to judge the fault type and severity (light, medium, severe).

[0061] 3. LLM verification and inference optimization

[0062] Combine the output of the small model (fault type code, fault location, fault severity and fault confidence) with the knowledge base such as equipment historical data, maintenance manuals and expert knowledge and experience, generate a natural language description through the large model (LLM), and verify the logical consistency. The LLM invokes the RAG (retrieval-augmented generation) technology to extract supplementary information from unstructured data (maintenance records, summary reports, communication and reporting materials, expert knowledge and experience, etc.) and correct misjudgments.

[0063] 4. Output of comprehensive conclusion

[0064] Perform credibility weighted fusion on the output of the small model and the LLM verification score, and output the final diagnostic conclusion. The LLM recommends maintenance strategies according to the fault type, and generates standard operating procedures (SOP) and risk warnings.

[0065] Step S4: Maintenance decision-making: The decision-making agent generates maintenance strategies, spare parts allocation plans and management measures based on the diagnostic results and in combination with historical operation and maintenance data, and pushes them to the execution terminal.

[0066] Step S5: Closed-loop feedback: After the maintenance implementation, the review and summary agent analyzes the maintenance effect, optimizes the model parameters and threshold rules, and forms a closed-loop iteration.

[0067] The present invention uses a small model (mechanism model + AI model) to ensure interpretability and physical consistency, and a large model (LLM model) to improve the complex fault recognition ability and operation and maintenance decision-making ability. The two cooperate to improve the reliability of diagnosis and decision-making. The threshold / rule agent dynamically adjusts according to the working conditions and negative samples, reduces the false alarm rate, and automates the entire process from data collection to maintenance decision-making, with the response speed increased by more than 90%.

[0068] The decision agent combines the equipment health, production plan, and cost constraints to generate the optimal maintenance strategy, and the review and summary agent realizes the autonomous evolution of the model parameters and rules through closed-loop feedback.

[0069] The above is the basic embodiment of the present invention. The technical solution of the present invention will be further described below through one or several preferred embodiments.

[0070] Embodiment 1

[0071] Deploy multi-source vibration, temperature, current and other sensing and real-time data collection devices in the transmission chain (motor and gearbox) of the high-speed wire rod rolling mill in steel.

[0072] The edge computing node runs the feature extraction agent to extract time-domain and frequency-domain features, and dynamically sets the alarm threshold through the threshold rule agent (such as the vibration RMS amplitude threshold is adaptively adjusted according to the load rate).

[0073] After receiving the abnormal data, the cloud diagnosis agent first locates the fault preliminarily through a small model (vibration mechanism-based or expert knowledge and experience model), then further diagnoses the fault mode through a machine learning or artificial intelligence model, and finally calls a large model (LLM large language model) for a diagnosis summary.

[0074] The decision agent generates operation and maintenance strategies such as "immediately stop the machine to replace the bearing" or "handle it during the maintenance window period this week" according to the fault diagnosis conclusion (fault location, mode and level), and synchronously pushes the spare part requirements to the warehousing system.

[0075] After the maintenance, the review agent compares the predicted and actual fault causes, and optimizes the fault mode weights and their thresholds of the small model, as well as the training data set of the large model.

[0076] Refer to Figure 1 As shown, a device health inspection system based on multi-agent collaboration includes:

[0077] Perception layer: High-frequency, low-frequency, process, and control data under the operating state of the device are collected through multi-source sensors (such as vibration, noise, temperature, current, liquid level, pressure, flow, etc.), and the collected raw data is transmitted to the edge computing layer;

[0078] Edge computing layer: Receives the raw data transmitted by the perception layer, uses the feature extraction agent to process the data in real time to generate multi-dimensional feature vectors, and at the same time conducts preliminary anomaly detection through the threshold / rule alarm agent combined with dynamically adjusted thresholds and artificial intelligence models; if an anomaly is detected, an alarm signal is triggered and the multi-dimensional feature vector and anomaly information are transmitted to the cloud analysis layer;

[0079] Cloud analysis layer: Receives the multi-dimensional feature vector and anomaly information transmitted by the edge computing layer, further analyzes the anomaly situation using the mechanism + AI alarm agent, calls the small models (mechanism model + AI model) integrated in the fault diagnosis agent for fault location, type judgment, and severity assessment, and verifies the diagnosis result through the large model (LLM model) to output a comprehensive diagnosis conclusion; subsequently, the decision-making agent generates a maintenance strategy, spare parts allocation plan, and management measures based on the diagnosis conclusion and historical operation and maintenance data, and sends the maintenance instructions to the execution layer;

[0080] Execution layer: Receives the maintenance instructions generated by the cloud analysis layer, implements specific maintenance operations through the maintenance execution terminal, and feeds back the maintenance implementation result to the feedback layer;

[0081] Feedback layer: Receives the maintenance implementation result of the execution layer, analyzes the maintenance effect through the review and summary agent, optimizes the model parameters, threshold rules, and knowledge base, and feeds back the optimized parameters and rules to the edge computing layer and the cloud analysis layer to form a closed-loop iteration.

[0082] Those skilled in the art know that in addition to implementing the system and its various devices, modules, and units provided by the present invention in the form of pure computer-readable program code, the method steps can be logically programmed to enable the system and its various devices, modules, and units provided by the present invention to be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers to achieve the same functions. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be regarded as the structure within the hardware component; the devices, modules, and units for implementing various functions can also be regarded as both software modules for implementing the method and the structure within the hardware component.

[0083] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. Without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A device health check method based on multi-agent collaboration, characterized in that, Including: Data acquisition step: Acquire target data to generate a multi-dimensional feature vector; The target data includes frequency data, process data, and control data; Equipment monitoring step: Monitor the equipment based on the generated multi-dimensional feature vector and output the monitoring result; Deployment and processing step: Deploy and implement the corresponding processing scheme according to the monitoring result.

2. The device health check method based on multi-agent collaboration according to claim 1, wherein It also includes: Closed-loop feedback step: After the maintenance is implemented, enable the review and summary agent to analyze the maintenance effect, optimize the model parameters and threshold rules, and form a closed-loop iteration.

3. The device health check method based on multi-agent collaboration according to claim 1, characterized in that The data acquisition step includes: Collect target data through multi-source sensors and use the feature extraction agent for real-time processing to generate a multi-dimensional feature vector.

4. The device health check method based on multi-agent collaboration according to claim 1, characterized in that The equipment monitoring step includes: Abnormality detection step: Use the threshold / rule agent to dynamically adjust the threshold, and combine machine learning and artificial intelligence models for preliminary abnormality judgment; if the threshold is exceeded, trigger the abnormality alarm agent; Fault diagnosis step: Use the diagnosis agent to call the small model to obtain the fault location, fault type, and fault severity data, and verify the diagnosis result through the large model, and output the comprehensive diagnosis conclusion; the small model includes a mechanism model and an AI model; the large model includes an LLM model.

5. The device health check method based on multi-agent collaboration according to claim 1, wherein, The deployment and processing step includes: Enable the decision-making agent to generate a maintenance strategy, spare part deployment plan, and management measures based on the diagnosis result, combined with historical operation and maintenance data, and push them to the execution terminal.

6. A device health check system based on multi-agent collaboration, which adopts the device health check method based on multi-agent collaboration according to any one of claims 1-5, characterized in that, Including: Perception layer, used to collect target data and transmit it to the edge computing layer; The target data includes frequency data, process data, and control data; Edge computing layer, used to receive target data and generate a multi-dimensional feature vector, and transmit it and the detected abnormal information to the cloud analysis layer; Cloud analysis layer, receive the multi-dimensional feature vector and abnormal information, use the mechanism + AI alarm agent to further analyze the abnormal situation and generate a maintenance instruction; Execution layer, based on the maintenance instruction, implement specific maintenance operations through the maintenance execution terminal, and feedback the maintenance implementation result to the feedback layer; Feedback layer, used to receive the feedback result for model optimization and knowledge base update.

7. The device health check system based on multi-agent collaboration according to claim 6, characterized in that, The perception layer collects target data under the operating state of the equipment through multi-source sensors and transmits the collected data to the edge computing layer.

8. The device health check system based on multi-agent collaboration according to claim 7, characterized in that, The edge computing layer uses the feature extraction agent to perform real-time processing on the target data to generate a multi-dimensional feature vector, and at the same time performs preliminary abnormality detection through the threshold / rule alarm agent in combination with the dynamically adjusted threshold and artificial intelligence model. If an abnormality is detected, trigger an alarm signal and transmit the multi-dimensional feature vector and abnormal information to the cloud analysis layer.

9. The device health check system based on multi-agent collaboration according to claim 8, wherein The cloud analysis layer uses the diagnosis agent to call the small model to obtain the fault location, fault type, and fault severity data, and verify the diagnosis result through the large model, and output the comprehensive diagnosis conclusion; the small model includes a mechanism model and an AI model; the large model includes an LLM model; enable the decision-making agent to generate a maintenance strategy, spare part deployment plan, and management measures based on the diagnosis conclusion and historical operation and maintenance data, and send the maintenance instruction to the execution layer.

10. The device health inspection system based on multi-agent collaboration according to claim 8, characterized in that, The feedback layer summarizes the maintenance effect through the review by the intelligent agent, optimizes the model parameters, threshold rules, and knowledge base, and feeds back the optimized parameters and rules to the edge computing layer and the cloud analysis layer to form a closed-loop iteration.

Images