CN-121037196-B - AI large model driven Internet of things fault diagnosis system

Abstract

The invention discloses an AI large-model-driven fault diagnosis system of an Internet of things, which comprises a data acquisition module, a dynamic detection module, a positioning analysis module and a fault diagnosis module, wherein the data acquisition module is used for acquiring original data of terminal equipment and classifying features to obtain initial data of the Internet of things, the dynamic detection module is used for establishing a multi-mode fault alarm device according to the initial data of the Internet of things, the positioning analysis module is used for modeling by utilizing three-dimensional influence and analyzing the influence range of downstream equipment by adopting a reverse push false method, and the fault diagnosis module is used for calling a large-model driving interface and diagnosing the root cause of the equipment according to fault source positioning to realize complete closed loop from original signal to feature extraction to multi-mode alarm to influence analysis and root cause diagnosis.

Inventors

• WANG QIANG
• WANG YI

Assignees

• 中科网谷(武汉)科技有限公司

Dates

Publication Date

20260508

Application Date

20250710

Claims (4)

1. The system is characterized by comprising a data acquisition module, a dynamic detection module, a positioning analysis module and a fault diagnosis module; the data acquisition module is used for acquiring the original data of the terminal equipment and classifying the characteristics to obtain the initial data of the Internet of things; The dynamic detection module is used for establishing a multi-mode fault alarm device according to the initial data of the Internet of things; the positioning analysis module is used for modeling by utilizing three-dimensional influence and analyzing the influence range of downstream equipment by adopting a reverse push false seeking method; the fault diagnosis module is used for calling the large model driving interface and diagnosing the root cause of the equipment fault according to the fault source positioning; the dynamic detection module is used for establishing a multi-mode fault alarm device, and the multi-mode fault alarm device comprises a pulse alarm unit, a progressive dyeing unit, a reverse tracing unit and a topology abnormality unit; The pulse alarm unit comprises a peak value storage device which is established according to the state signal of first fault data, detects the state signal in the current sampling period when each clock rises, calculates the standard deviation of the state signal in the current sampling period and the average value of the state signal in the current sampling period, and calculates a dynamic threshold value in a set unit threshold value clock period, wherein the calculation expression of the dynamic threshold value is as follows: dynamic threshold = state signal mean +5x state signal standard deviation; triggering a first alarm and starting a timing device when the status signal is greater than the dynamic threshold; The progressive dyeing unit comprises: Performing independent materialization simulation on equipment for acquiring data corresponding to second fault data, setting corresponding data interfaces of each equipment, setting three-dimensional equipment model coloring according to slope fluctuation amplitude of the second fault data, setting initial model coloring to be blue, namely equipment state health, rendering hue from high to low according to the slope fluctuation amplitude, setting red as equipment state fault, triggering a second alarm when the equipment state is fault, and starting a long-term positioning timing device; the reverse traceback unit includes: Calculating cross-correlation function mean values of vibration signals, current signals and temperature signals under normal working conditions of the original data in unit time, automatically updating the cross-correlation function mean values every seven days by adopting a self-adaptive adjustment mechanism to obtain a dynamic cross-correlation function mean value, triggering a third alarm when any cross-correlation value is larger than the dynamic cross-correlation function mean value and the duration exceeds the correlation time mean value of third fault data, and starting a reverse tracing positioning device; the topology anomaly unit includes: A lightweight topological feature extraction module is deployed, the number of connected components and the number of annular structures are calculated in real time, the average value of the number of connected components and the number of annular structures in unit time is calculated, a preliminary threshold value is set according to the average value, the fluctuation range of the number of connected components and the number of annular structures is obtained, the fluctuation range is used for obtaining the unit time span period of the number of connected components and the number of annular structures, and when the number of connected components and the number of annular structures are larger than the preliminary threshold value and the fluctuation range exceeds three span periods, a fourth alarm is triggered, and a terminal feedback device is started; the positioning analysis module comprises: Automatically generating N possible fault sources by using equipment based on a random algorithm, constructing a three-dimensional influence model by using N possible fault source equipment, constructing a three-dimensional influence model by using mechanical vibration, heat conduction and current disturbance, calculating a vibration signal by using an exponential decay model, calculating current signal disturbance distribution by using a non-homogeneous heat equation by using a current disturbance, calculating current signal disturbance distribution by using an impedance network model, inputting initial data of the Internet of things into the three-dimensional influence model based on a digital twin technology, displaying an influence field by using a color temperature gradual curved surface, setting the equipment as a red core area when the vibration signal, heat conduction and current signal disturbance distribution exceeds a model 1.5 times threshold of the three-dimensional influence model, setting the equipment as a yellow transition area when the vibration signal, heat conduction and current signal disturbance distribution is lower than a 1.5 times threshold and exceeds a 1.2 times threshold, and setting the equipment as a green safety area when the vibration signal, heat conduction and current signal disturbance distribution is lower than a 1.2 times threshold; the fault diagnosis module includes: Extracting fault data of red core area equipment, carrying out multi-mode feature fusion on the fault data, wherein the multi-mode feature fusion is used for sequencing sensor data and maintenance log text of the fault data according to a time sequence, inputting the fault data into a large model for training, fusing historical fault data and an online networking mechanism, obtaining a knowledge enhancement large model, calling a knowledge enhancement large model interface, and obtaining a diagnosis result and a decision suggestion.
2. 2. The AI large model driven Internet of things fault diagnosis system of claim 1, wherein the specific method of the data acquisition module comprises: Acquiring original data of each terminal and each sensor in real time by utilizing edge calculation, uploading the original data to an Internet of things cloud platform, and extracting characteristics of the original data according to operation characteristics of the original data to obtain initial data of the Internet of things, wherein the initial data of the Internet of things comprises first fault data, second fault data, third fault data and fourth fault data; The original data comprises equipment state logs, running time, state signal data, physical monitoring data, environment data and positioning beacon data, wherein the physical monitoring data comprises vibration signals and current signals, and the environment data comprises temperature signals, air pressure signals and air humidity; the first fault data is used for storing transient burst faults; the second fault data is used for storing long-time progressive faults; the third fault data is used for storing multi-factor composite faults; and the fourth fault data is used for storing unknown novel faults.
3. 3. The AI large model driven fault diagnosis system of Internet of things of claim 2, wherein the specific method for extracting the characteristics of the original data according to the operation characteristics of the original data is as follows: extracting original data in a sliding window in unit time by utilizing the sliding window on the cloud platform of the Internet of things, calculating a state signal peak value and an RMS value of the original data, carrying out wavelet packet decomposition on the state signal data, calculating the energy ratio of the extracted signal which is larger than unit frequency by using a Welch method, and storing the original data as first fault data when the ratio of the state signal peak value to the RMS value is larger than a set threshold value and the energy ratio is larger than the average energy ratio of the partial signal; According to the running time of the original data, the running time is taken as a unit period for 30 days, a regular fluctuation diagram of the original data is drawn, a regular fluctuation trend of every seven days is obtained, the slope of the regular fluctuation trend is calculated, and when the slope continuously increases in the unit period, the original data is stored as second fault data; respectively calculating the average value of the vibration signal, the current signal and the temperature signal of the original data, deleting the original data of which the vibration signal, the current signal and the temperature signal are equal to the corresponding average values, calculating the cross-correlation function value of the vibration signal, the current signal and the temperature signal by adopting a cross-correlation function, and storing the original data as third fault data when the cross-correlation function value is larger than a correlation threshold value; And constructing Vietoris-Rips complex topological structure of the original data according to a single sample data point, obtaining the annular structure number and the connected component number of the sample data point, recording the life cycle of the topological feature, calculating the average value of the annular structure number and the connected component number, and storing the sample data as fourth fault data when the connected component number and the annular structure number are larger than the average value.
4. 4. The AI large model driven Internet of things fault diagnosis system of claim 3 wherein said first alarm is for timing a countdown alarm; The second alarm is used for activating a sensor continuous recording mode and starting a positioning device to obtain an abnormal source azimuth; The third alarm is used for reversely tracking the rendering propagation path and automatically locking the three-level association equipment at the upstream and downstream; and the fourth alarm is used for comparing the virtual model with the actual fault time in real time and adjusting the fault instant stop time.

Description

AI large model driven Internet of things fault diagnosis system Technical Field The invention relates to the technical field of anomaly detection and fault classification, in particular to an AI large-model-driven fault diagnosis system of the Internet of things. Background Along with the development of modern technology, the requirements of enterprises on equipment health association are remarkably increased, the traditional fault diagnosis mode cannot meet the high-efficiency operation and maintenance requirements of the modern industry, and the intelligent fault diagnosis system based on the Internet of things can realize real-time monitoring, fault early warning and predictive diagnosis, so that the production efficiency is improved. The invention of China with the application number of 202410624493.X discloses an Internet of things fault diagnosis method and system based on an intelligent optimization algorithm, wherein the method comprises the steps of collecting operation data of equipment in the Internet of things of a mobile phone, and preprocessing the operation data to eliminate the influence of noise and abnormal values; the method comprises the steps of preprocessing operation data, processing the preprocessed operation data, identifying abnormal data and potential fault equipment, judging the operation state of the equipment and positioning the fault equipment, analyzing the fault equipment in detail based on the output result of an intelligent optimization algorithm, determining the fault cause of the fault equipment, mining the potential cause of the fault, and comprehensively analyzing the fault by combining the actual operation condition and historical data. The method fails to perform cross-dimensional data fusion based on nonlinear data, a model is not set, a reverse-push assumption is utilized to locate fault sources, a topological relation between devices is not established, and an alarm system is set. Disclosure of Invention The method solves the technical problems that cross-dimension data fusion cannot be performed based on nonlinear data, a model is not set, a reverse-push assumption is utilized to position a fault source, the establishment of topological relation between equipment is lacked, and an alarm system is set. In order to solve the technical problems, the invention provides the following technical scheme: The system for diagnosing the faults of the Internet of things driven by the AI large model comprises a data acquisition module, a dynamic detection module, a positioning analysis module and a fault diagnosis module; the data acquisition module is used for acquiring the original data of the terminal equipment and classifying the characteristics to obtain the initial data of the Internet of things; The dynamic detection module is used for establishing a multi-mode fault alarm device according to the initial data of the Internet of things; the positioning analysis module is used for modeling by utilizing three-dimensional influence and analyzing the influence range of downstream equipment by adopting a reverse push false seeking method; The fault diagnosis module is used for calling the large model driving interface and diagnosing the root cause of the equipment fault according to the fault source positioning. Preferably, the specific method of the data acquisition module comprises the following steps: Acquiring original data of each terminal and each sensor in real time by utilizing edge calculation, uploading the original data to an Internet of things cloud platform, and extracting characteristics of the original data according to operation characteristics of the original data to obtain initial data of the Internet of things, wherein the initial data of the Internet of things comprises first fault data, second fault data, third fault data and fourth fault data; The original data comprises equipment state logs, running time, state signal data, physical monitoring data, environment data and positioning beacon data, wherein the physical monitoring data comprises vibration signals and current signals, and the environment data comprises temperature signals, air pressure signals and air humidity; the first fault data is used for storing transient burst faults; the second fault data is used for storing long-time progressive faults; the third fault data is used for storing multi-factor composite faults; and the fourth fault data is used for storing unknown novel faults. Preferably, according to the operation characteristics of the original data, the specific method for extracting the characteristics of the original data is as follows: Extracting original data in a sliding window in unit time by utilizing the sliding window on the Internet of things platform, calculating a state signal peak value and an RMS value of the original data, carrying out wavelet packet decomposition on the state signal data, calculating the energy ratio extracted to be larger than unit frequency by usin