Safety equipment fault prediction and intelligent maintenance decision-making system based on AI large model

Abstract

The invention discloses a safety equipment fault prediction and intelligent maintenance decision-making system based on an AI large model, which belongs to the technical field of computers and comprises a data acquisition and integration unit, an AI large model construction and training unit, a fault prediction and root cause positioning unit, an intelligent maintenance decision-making generation unit and a dynamic optimization and self-learning unit. According to the safety equipment fault prediction and intelligent maintenance decision-making system based on the AI large model, through data integration, model optimization and decision-making closed loop, the equipment reliability is remarkably improved, the operation and maintenance cost is reduced, in the future, along with deepening of model interpretability, industry customization and ecological cooperation, the technology will become a core infrastructure, and the technology has a wide application prospect. The manufacturing industry is promoted to advance towards predictive maintenance and zero-fault production, a deep learning model is constructed by integrating multi-source data, and precise prediction of the equipment health state and intelligent optimization of a maintenance strategy are realized in combination with industry knowledge.

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and in particular relates to a safety equipment fault prediction and intelligent maintenance decision-making system based on an AI large model. Background Art

[0002] With the development of science and technology, the life cycle management of various equipment has become increasingly important. Effective management of the life cycle of equipment can ensure that the equipment can operate efficiently and stably throughout its entire period of use, thereby improving the use efficiency of the equipment, reducing the equipment's operation and maintenance costs, and extending the equipment's service life.

[0003] The prior art discloses some invention patents in the field of computer technology, including one with publication number CN117170998B, which discloses an intelligent equipment lifecycle management system. The system includes a prediction computing node and multiple state computing nodes, wherein the multiple state computing nodes are used to obtain equipment state data. The multiple state computing nodes also include an equipment failure state prediction module, which is used to predict the equipment failure state. When the equipment enters a failure state based on the output of the equipment failure state prediction module in the state computing node, it is determined whether to report to the prediction computing node. The prediction computing node predicts the expected remaining life of the equipment, determines the service life evaluation result of the equipment based on the predicted expected remaining life of the equipment, determines whether to maintain or replace the equipment, and provides corresponding prompts in the background of the equipment lifecycle management system. The system can improve the accuracy and efficiency of equipment management, optimize equipment use and maintenance decisions, and enhance the user experience of the equipment management system. However, this technical solution still has some shortcomings during its application. Most existing safety equipment monitoring systems rely on experienced maintenance personnel to regularly maintain and inspect safety equipment, and do not collect vibration data of safety equipment in real time. In addition, safety equipment is affected by the environment during operation, resulting in imbalance of the safety equipment and prone to safety problems.

[0004] Based on this, the present invention designs a safety equipment fault prediction and intelligent maintenance decision-making system based on AI big model to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to solve the problem that most of the existing safety equipment monitoring systems rely on experienced maintenance personnel to regularly maintain and inspect the safety equipment, and do not collect the vibration data of the safety equipment in real time. In addition, the safety equipment will be affected by the environment during operation, resulting in imbalance of the safety equipment and prone to safety problems. A safety equipment fault prediction and intelligent maintenance decision-making system based on AI large model is proposed.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] Safety equipment fault prediction and intelligent maintenance decision-making system based on AI big model, including data collection and integration unit, AI big model construction and training unit, fault prediction and root cause location unit, intelligent maintenance decision-making unit and dynamic optimization and self-learning unit;

[0008] The data acquisition and integration unit collects equipment operation data in real time through industrial Internet of Things sensors, cleans and normalizes the raw data, and extracts time series features and statistical features;

[0009] The AI ​​large model construction and training unit uses a deep learning model to process time series data, based on historical fault data and real-time monitoring data, and uses a cross-validation method to optimize hyperparameters;

[0010] The fault prediction and root cause location unit analyzes device data in real time through edge computing nodes and outputs fault characteristics;

[0011] The intelligent maintenance decision-making unit intelligently adjusts the maintenance strategy based on the equipment health status, maintenance cost and production plan, combined with the equipment usage frequency, working environment and historical maintenance records;

[0012] The dynamic optimization and self-learning unit automatically updates model parameters through real-time data and user feedback.

[0013] As a further description of the above technical solution:

[0014] The data acquisition and integration unit includes a multi-source data access module and a data preprocessing and feature engineering module;

[0015] The multi-source data access module collects equipment operation data in real time through industrial Internet of Things sensors, and integrates historical maintenance records, environmental data and expert experience knowledge to build a hybrid data set containing time series data and unstructured data;

[0016] The equipment operation data includes vibration, temperature, pressure and current, and the environmental data includes dust and humidity;

[0017] The data preprocessing and feature engineering module cleans and normalizes the raw data, extracts time series features and statistical features, and labels faults based on domain knowledge, providing high-quality data for model training.

[0018] The time series features include mean, variance and frequency domain features, and the statistical features include peak and skewness.

[0019] As a further description of the above technical solution:

[0020] The AI ​​large model construction and training unit includes a model selection and architecture design module and a model training and optimization module. The model selection and architecture design module uses a deep learning model to process time series data and combines knowledge graph technology to integrate equipment mechanisms and industry rules. The model training and optimization module is based on historical fault data and real-time monitoring data, accelerates model convergence through transfer learning technology, and uses cross-validation methods to optimize hyperparameters.

[0021] As a further description of the above technical solution:

[0022] The fault prediction and root cause location unit includes a real-time monitoring and anomaly detection module and a root cause analysis and fault diagnosis module. The real-time monitoring and anomaly detection module analyzes device data in real time through edge computing nodes, uses models to identify abnormal patterns, and triggers early warnings. The abnormal patterns include sudden vibration changes and temperature anomalies.

[0023] The root cause analysis and fault diagnosis module locates the fault source based on the fault characteristics output by the model and combines the knowledge graph to provide maintenance suggestions.

[0024] As a further description of the above technical solution:

[0025] The intelligent maintenance decision generation unit includes a maintenance strategy recommendation module and a resource optimization and scheduling module. The maintenance strategy recommendation module generates a dynamic maintenance plan based on the equipment health status, maintenance cost and production plan. The dynamic maintenance plan includes emergency maintenance, planned downtime and spare parts replacement.

[0026] The resource optimization and scheduling module intelligently adjusts maintenance strategies based on equipment usage frequency, working environment, and historical maintenance records to achieve optimal resource allocation. For equipment with stable operation, it appropriately extends maintenance cycles and reduces downtime.

[0027] As a further description of the above technical solution:

[0028] The dynamic optimization and self-learning unit includes a model continuous iteration module and a multi-system collaboration and standardization module. The model continuous iteration module automatically updates model parameters through real-time data and user feedback to improve prediction accuracy.

[0029] The multi-system collaboration and standardization module promotes the development of standardized interfaces between industrial Internet platforms and large AI models, and supports data sharing and model reuse among multiple enterprises and multiple devices.

[0030] As a further description of the above technical solution:

[0031] The root cause analysis and fault diagnosis module analyzes and predicts faults in security equipment operation and maintenance based on a discrete algorithm combined with a BP neural network:

[0032] Obtain the historical original data of safety equipment operation and establish an initial decision table for the historical original parameters. The historical original data is various historical fault data in the fault statistics table of each device;

[0033] Calculate the correlation between each row of data in the initial decision table, simplify the data in the horizontal dimension, and when establishing the initial decision table for the historical original data, remove the repeated parameters and add the necessary characteristic parameters to form a complete initial decision table;

[0034] Discretize the continuous variables after horizontal data simplification, and simplify the vertical attributes of the discretized data;

[0035] Obtain input data of the BP neural network model, obtain input data of the BP neural network model;

[0036] According to the input data of the BP neural network model, the BP neural network model is trained. After the BP neural network model is trained, the corresponding test samples are used to test it, and the prediction results of equipment failure are output and analyzed.

[0037] Determine the neural network model to be trained, and determine the BP neural network model to be trained;

[0038] According to the determined training BP neural network model, the mechanical and electrical equipment failure is predicted.

[0039] As a further description of the above technical solution:

[0040] The discretization of continuous variables after horizontal data simplification includes the following steps:

[0041] Obtain fault feature sequence L0 and fault feature parameter sequence L i ;

[0042] L0=(l0(1)), (l0(2)),…,(l0(n))

[0043] L i =(l i (1), l i (2),…,l i (n)), i = 1, 2, ..., m

[0044] Calculate the initial image of each sequence;

[0045] L′ i =L i / l i (1) = (l′ i (,), l′ i (2),…,l′ i(n)), i = 1, 2, ..., m

[0046] Calculate the fault sequence feature column L0 and the fault feature parameter sequence L i The absolute value sequence of the difference between the corresponding components of the initial image;

[0047] Δ i (r)=|l′0(r)-l′ i (r)|

[0048] Calculate the maximum and minimum values ​​of the absolute value sequence, respectively:

[0049]

[0050] Calculate the correlation coefficient:

[0051]

[0052] Where ξ is the resolution coefficient;

[0053] Calculate the grey correlation:

[0054]

[0055] The horizontal dimension data is simplified, including the following steps:

[0056] According to the decision table ZT = (W,, C∪Z, S, f), calculate the difference matrix M of the decision table ZT n×n (ZT), get M n×n (ZT)=(c ij ) n×n The lower triangular matrix of

[0057] Where i, j = 1, 2, 3, ..., n

[0058] Calculate the difference function Δ and set the disjunctive normal form:

[0059]

[0060] Perform conjunction operation on the disjunctive normal form to obtain the difference function:

[0061]

[0062] Transform the conjunctive normal form of the difference function into the disjunctive normal form, and we get Then Δ l Represents the result of l attribute reduction;

[0063] After obtaining the simplified processing of the fault decision table, the redundant column data is deleted to complete the simplification of the condition attributes of the vertical fault decision table.

[0064] As a further description of the above technical solution:

[0065] Obtaining input data of the BP neural network model includes the following steps:

[0066] Calculate the grey relational degree of m groups of sample data, eliminate p groups of irrelevant data, and obtain (mp) fault influencing factors;

[0067] Remove q redundant failure influencing factors and obtain (nq) failure influencing factors;

[0068] The input of the BP neural network model is a (mp)×(nq) fault data decision table.

[0069] As a further description of the above technical solution:

[0070] Determining the neural network model to be trained includes the following steps:

[0071] According to the analysis results, determine whether the BP neural network model meets the prediction requirements;

[0072] If the prediction requirements are met, the trained BP neural network model is determined; if the prediction requirements are not met, S2-S5 are repeated until a BP neural network model that meets the prediction requirements is determined.

[0073] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0074] 1. In this invention, the safety equipment fault prediction and intelligent maintenance decision-making system based on the AI ​​large model significantly improves equipment reliability and reduces operation and maintenance costs through data integration, model optimization and decision-making closed loop. In the future, with the deepening of model interpretability, industry customization and ecological collaboration, this technology will become the core infrastructure, driving the manufacturing industry towards predictive maintenance and zero-fault production. By integrating multi-source data, building deep learning models, and combining industry knowledge, it can achieve accurate prediction of equipment health status and intelligent optimization of maintenance strategies.

[0075] 2. In the present invention, by achieving the effects of saving manpower, excellent prediction accuracy and efficiency, and excellent safety through the safety equipment fault prediction method, the AI ​​large model has conducted research and analysis on the operation and maintenance of safety equipment, and constructed the operation and maintenance model based on past operation and maintenance data for different equipment. Based on AI, these models are integrated with the equipment's knowledge base to build a basic model for equipment operation and maintenance, enhance the model's professional knowledge and reasoning ability, allow data to run more, and provide customers with accurate operation and maintenance services. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 This is a module block diagram of the safety equipment fault prediction and intelligent maintenance decision-making system based on the AI ​​large model proposed by the present invention;

[0077] Figure 2 A schematic diagram of the fault analysis and prediction of the safety equipment fault prediction and intelligent maintenance decision-making system based on the AI ​​large model proposed by the present invention;

[0078] Figure 3 This is a schematic diagram of the input data of the BP neural network model of the safety equipment fault prediction and intelligent maintenance decision-making system based on the AI ​​large model proposed in the present invention;

[0079] Figure 4 This is a schematic diagram of the BP neural network model trained for the AI ​​large model-based safety equipment fault prediction and intelligent maintenance decision-making system proposed in the present invention. DETAILED DESCRIPTION

[0080] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0081] Example 1:

[0082] Safety equipment fault prediction and intelligent maintenance decision-making system based on AI big model, including data collection and integration unit, AI big model construction and training unit, fault prediction and root cause location unit, intelligent maintenance decision-making unit and dynamic optimization and self-learning unit;

[0083] The data acquisition and integration unit collects equipment operation data in real time through industrial Internet of Things sensors, cleans and normalizes the raw data, and extracts time series features and statistical features;

[0084] The AI ​​large model construction and training unit uses a deep learning model to process time series data, based on historical fault data and real-time monitoring data, and uses a cross-validation method to optimize hyperparameters;

[0085] The fault prediction and root cause location unit analyzes device data in real time through edge computing nodes and outputs fault characteristics;

[0086] The intelligent maintenance decision-making unit intelligently adjusts the maintenance strategy based on the equipment health status, maintenance cost and production plan, combined with the equipment usage frequency, working environment and historical maintenance records;

[0087] The dynamic optimization and self-learning unit automatically updates model parameters through real-time data and user feedback.

[0088] Specifically, the data acquisition and integration unit includes a multi-source data access module and a data preprocessing and feature engineering module;

[0089] The multi-source data access module collects equipment operation data in real time through industrial Internet of Things sensors, and integrates historical maintenance records, environmental data and expert experience knowledge to build a hybrid data set containing time series data and unstructured data;

[0090] The equipment operation data includes vibration, temperature, pressure and current, and the environmental data includes dust and humidity;

[0091] The data preprocessing and feature engineering module cleans and normalizes the raw data, extracts time series features and statistical features, and labels faults based on domain knowledge, providing high-quality data for model training.

[0092] The time series features include mean, variance and frequency domain features, and the statistical features include peak and skewness.

[0093] Specifically, the AI ​​large model construction and training unit includes a model selection and architecture design module and a model training and optimization module. The model selection and architecture design module uses a deep learning model to process time series data, and combines knowledge graph technology to integrate equipment mechanisms and industry rules. The model training and optimization module is based on historical fault data and real-time monitoring data, accelerates model convergence through transfer learning technology, and uses cross-validation methods to optimize hyperparameters.

[0094] Specifically, the fault prediction and root cause location unit includes a real-time monitoring and anomaly detection module and a root cause analysis and fault diagnosis module. The real-time monitoring and anomaly detection module analyzes device data in real time through edge computing nodes, uses models to identify abnormal patterns, and triggers early warnings. The abnormal patterns include sudden vibration changes and temperature anomalies.

[0095] The root cause analysis and fault diagnosis module locates the fault source based on the fault characteristics output by the model and combines the knowledge graph to provide maintenance suggestions.

[0096] Specifically, the intelligent maintenance decision generation unit includes a maintenance strategy recommendation module and a resource optimization and scheduling module. The maintenance strategy recommendation module generates a dynamic maintenance plan based on the equipment health status, maintenance cost and production plan. The dynamic maintenance plan includes emergency maintenance, planned downtime and spare parts replacement.

[0097] The resource optimization and scheduling module intelligently adjusts maintenance strategies based on equipment usage frequency, working environment, and historical maintenance records to achieve optimal resource allocation. For equipment with stable operation, it appropriately extends maintenance cycles and reduces downtime.

[0098] Specifically, the dynamic optimization and self-learning unit includes a model continuous iteration module and a multi-system collaboration and standardization module. The model continuous iteration module automatically updates model parameters through real-time data and user feedback to improve prediction accuracy.

[0099] The multi-system collaboration and standardization module promotes the development of standardized interfaces between industrial Internet platforms and large AI models, and supports data sharing and model reuse among multiple enterprises and multiple devices.

[0100] Specifically, the root cause analysis and fault diagnosis module analyzes and predicts faults in security equipment operation and maintenance based on a discrete algorithm combined with a BP neural network:

[0101] Obtain the historical original data of safety equipment operation and establish an initial decision table for the historical original parameters. The historical original data is various historical fault data in the fault statistics table of each device;

[0102] Calculate the correlation between each row of data in the initial decision table, simplify the data in the horizontal dimension, and when establishing the initial decision table for the historical original data, remove the repeated parameters and add the necessary characteristic parameters to form a complete initial decision table;

[0103] Discretize the continuous variables after horizontal data simplification, and simplify the vertical attributes of the discretized data;

[0104] Obtain input data of the BP neural network model, obtain input data of the BP neural network model;

[0105] According to the input data of the BP neural network model, the BP neural network model is trained. After the BP neural network model is trained, the corresponding test samples are used to test it, and the prediction results of equipment failure are output and analyzed.

[0106] Determine the neural network model to be trained, and determine the BP neural network model to be trained;

[0107] According to the determined training BP neural network model, the mechanical and electrical equipment failure is predicted.

[0108] Specifically, the discretization of continuous variables after horizontal data simplification includes the following steps:

[0109] Obtain fault feature sequence L0 and fault feature parameter sequence L i ;

[0110] L0=(l0(1)), (l0(2)),…,(l0(n))

[0111] L i =(l i (1), l i (2),…,l i (n)), i = 1, 2, ..., m

[0112] Calculate the initial image of each sequence;

[0113] L′ i =L i / l i (1) = (l′ i (,), l′ i (2),…,l′ i (n)), i = 1, 2, ..., m

[0114] Calculate the fault sequence feature column L0 and the fault feature parameter sequence L i The absolute value sequence of the difference between the corresponding components of the initial image;

[0115] Δ i (r)=|l′0(r)-l′ i (r)|

[0116] Calculate the maximum and minimum values ​​of the absolute value sequence, respectively:

[0117]

[0118] Calculate the correlation coefficient:

[0119]

[0120] Where ξ is the resolution coefficient;

[0121] Calculate the grey correlation:

[0122]

[0123] The horizontal dimension data is simplified, including the following steps:

[0124] According to the decision table ZT = (W, C∪Z, S, f), calculate the difference matrix M of the decision table ZT n×n (ZT), get M n×n (ZT)=(c ij ) n×n The lower triangular matrix of

[0125] Where i, j = 1, 2, 3, ..., n

[0126] Calculate the difference function Δ and set the disjunctive normal form:

[0127]

[0128] Perform conjunction operation on the disjunctive normal form to obtain the difference function:

[0129]

[0130] Transform the conjunctive normal form of the difference function into the disjunctive normal form, and we get Then Δ l Represents the result of l attribute reduction;

[0131] After obtaining the simplified processing of the fault decision table, the redundant column data is deleted to complete the simplification of the condition attributes of the vertical fault decision table.

[0132] Specifically, obtaining the input data of the BP neural network model includes the following steps:

[0133] Calculate the grey relational degree of m groups of sample data, eliminate p groups of irrelevant data, and obtain (mp) fault influencing factors;

[0134] Remove q redundant failure influencing factors and obtain (nq) failure influencing factors;

[0135] The input of the BP neural network model is a (mp)×(nq) fault data decision table.

[0136] Specifically, determining the trained neural network model includes the following steps:

[0137] According to the analysis results, determine whether the BP neural network model meets the prediction requirements;

[0138] If the prediction requirements are met, the trained BP neural network model is determined; if the prediction requirements are not met, S2-S5 are repeated until a BP neural network model that meets the prediction requirements is determined.

[0139] Example 2:

[0140] The multi-source data access module collects equipment operation data in real time through industrial IoT sensors, while integrating historical maintenance records, environmental data, and expert experience to build a hybrid data set containing time series data and unstructured data;

[0141] The equipment operation data includes vibration, temperature, pressure and current;

[0142] The environmental data include dust and humidity;

[0143] Example: In safety equipment maintenance, vibration and temperature sensors are deployed, combined with meteorological data and historical failure cases to form a multidimensional data model;

[0144] The data preprocessing and feature engineering module cleans and normalizes raw data, extracts time series and statistical features, and labels faults based on domain knowledge, providing high-quality data for model training.

[0145] The time series features include mean, variance and frequency domain features;

[0146] The statistical characteristics include peak and skewness;

[0147] Example: In safety equipment maintenance, transformer oil chromatogram data is analyzed to extract features to predict insulation faults;

[0148] The model selection and architecture design module uses deep learning models to process time series data and integrates equipment mechanisms and industry rules with knowledge graph technology;

[0149] Example: In the field of safety equipment manufacturing, deep learning models are used to analyze equipment vibration data, combine knowledge graphs to locate fault sources, and optimize maintenance strategies.

[0150] The model training and optimization module uses transfer learning technology to accelerate model convergence based on historical fault data and real-time monitoring data, and adopts cross-validation methods to optimize hyperparameters;

[0151] Example: In predictive maintenance of safety equipment, historical vibration data is used to train the model, and real-time data is used to verify the model accuracy, thus reducing the risk of overfitting.

[0152] Real-time monitoring and anomaly detection module, which uses edge computing nodes to analyze device data in real time, uses models to identify abnormal patterns, and triggers early warnings;

[0153] The abnormal modes include sudden vibration and abnormal temperature;

[0154] Example: During safety equipment maintenance, real-time monitoring of bearing vibration data is combined with a model to predict failure probability, issuing an early warning 30 days in advance.

[0155] The root cause analysis and fault diagnosis module locates the fault source based on the fault characteristics output by the model and combines it with the knowledge graph, and provides maintenance suggestions;

[0156] Example: In safety equipment maintenance, by analyzing vibration and temperature data and combining it with knowledge graphs, bearing wear or imbalance problems can be located and maintenance plans optimized.

[0157] The root cause analysis and fault diagnosis module analyzes and predicts faults in security equipment operation and maintenance based on discrete algorithms combined with BP neural networks;

[0158] Obtain the historical original data of safety equipment operation and establish an initial decision table for the historical original parameters. The historical original data is various historical fault data in the fault statistics table of each device;

[0159] Calculate the correlation of each row of data in the initial decision table, simplify the data in the horizontal dimension, and when establishing the initial decision table for the historical original data, remove the repeated parameters and add the necessary characteristic parameters to form a complete initial decision table. When discretizing the continuous variables after the horizontal data is simplified, the following steps are included:

[0160] Obtain fault feature sequence L0 and fault feature parameter sequence L i ;

[0161] L0=(l0(1)), (l0(2)),…,(l0(n))

[0162] L i =(l i (1), l i (2),…,l i (n)), i = 1, 2, ..., m

[0163] Calculate the initial image of each sequence;

[0164] L′ i =L i / l i (1) = (l′ i (1), l′ i (2),…,l′ i (n)), i = 1, 2, ..., m

[0165] Calculate the fault sequence feature column L0 and the fault feature parameter sequence L i The absolute value sequence of the difference between the corresponding components of the initial image;

[0166] Δ i (r)=|l′0(r)-l′ i (r)|

[0167] Calculate the maximum and minimum values ​​of the absolute value sequence, respectively:

[0168]

[0169] Calculate the correlation coefficient:

[0170]

[0171] Where ξ is the resolution coefficient;

[0172] Calculate the grey correlation:

[0173]

[0174] The horizontal dimension data is simplified, including the following steps:

[0175] According to the decision table ZT = (W, C∪Z, S, f), calculate the difference matrix M of the decision table ZT n×n (ZT), get M n×n (ZT)=(c ij ) n×n The lower triangular matrix of

[0176] Where i, j = 1, 2, 3, ..., n

[0177] Calculate the difference function Δ and set the disjunctive normal form:

[0178]

[0179] Perform conjunction operation on the disjunctive normal form to obtain the difference function:

[0180]

[0181] Transform the conjunctive normal form of the difference function into the disjunctive normal form, and we get Then Δ l Represents the result of l attribute reduction;

[0182] After obtaining the simplified fault decision table, delete the redundant column data and complete the simplification of the condition attributes of the vertical fault decision table;

[0183] Discretize the continuous variables after horizontal data simplification, and simplify the vertical attributes of the discretized data;

[0184] Obtaining input data of the BP neural network model includes the following steps:

[0185] Calculate the grey relational degree of m groups of sample data, eliminate p groups of irrelevant data, and obtain (mp) fault influencing factors;

[0186] Remove q redundant failure influencing factors and obtain (nq) failure influencing factors;

[0187] The input of the BP neural network model is a (mp)×(nq) fault data decision table;

[0188] According to the input data of the BP neural network model, the BP neural network model is trained. After the BP neural network model is trained, the corresponding test samples are used to test it, and the prediction results of equipment failure are output and analyzed.

[0189] Determining the neural network model to be trained includes the following steps:

[0190] According to the analysis results, determine whether the BP neural network model meets the prediction requirements;

[0191] If the prediction requirements are met, the trained BP neural network model is determined. If the prediction requirements cannot be met, the process is repeated until a BP neural network model that meets the prediction requirements is determined.

[0192] Predict mechanical and electrical equipment failures based on the determined training BP neural network model;

[0193] Maintenance strategy recommendation module, based on equipment health status, maintenance cost and production plan, the system generates dynamic maintenance plans;

[0194] The dynamic maintenance program includes emergency repairs, planned downtime and spare parts replacement;

[0195] Example: In safety equipment maintenance, by analyzing the equipment's remaining useful life (RUL), decision makers are provided with a maintenance time window to ensure maintenance is performed at the optimal time.

[0196] The resource optimization and scheduling module combines equipment usage frequency, working environment, and historical maintenance records to intelligently adjust maintenance strategies and optimize resource allocation. For equipment with stable operation, it appropriately extends maintenance cycles and reduces downtime.

[0197] Example: In safety equipment maintenance, more frequent preventive maintenance is used for critical equipment operating under high load;

[0198] The model continuous iteration module automatically updates model parameters and improves prediction accuracy through real-time data and user feedback;

[0199] Example: In safety equipment manufacturing plant maintenance, AI scheduling is used to optimize models based on actual maintenance results, reducing failure rates and lowering maintenance costs.

[0200] Multi-system collaboration and standardization modules promote the development of standardized interfaces between industrial Internet platforms and large AI models, supporting data sharing and model reuse across multiple enterprises and devices.

[0201] Example: In wind farm maintenance, unified data standards are used to achieve cross-device and cross-system fault prediction and maintenance collaboration.

[0202] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. Safety equipment fault prediction and intelligent maintenance decision-making system based on AI big model, characterized by: It includes data collection and integration unit, AI large model construction and training unit, fault prediction and root cause location unit, intelligent maintenance decision generation unit and dynamic optimization and self-learning unit; The data acquisition and integration unit collects equipment operation data in real time through industrial Internet of Things sensors, cleans and normalizes the raw data, and extracts time series features and statistical features; The AI ​​large model construction and training unit uses a deep learning model to process time series data, based on historical fault data and real-time monitoring data, and uses a cross-validation method to optimize hyperparameters; The fault prediction and root cause location unit analyzes device data in real time through edge computing nodes and outputs fault characteristics; The intelligent maintenance decision-making unit intelligently adjusts the maintenance strategy based on the equipment health status, maintenance cost and production plan, combined with the equipment usage frequency, working environment and historical maintenance records; The dynamic optimization and self-learning unit automatically updates model parameters through real-time data and user feedback.

2. The AI ​​large model-based safety equipment fault prediction and intelligent maintenance decision-making system according to claim 1 is characterized in that: The data acquisition and integration unit includes a multi-source data access module and a data preprocessing and feature engineering module; The multi-source data access module collects equipment operation data in real time through industrial Internet of Things sensors, and integrates historical maintenance records, environmental data and expert experience knowledge to build a hybrid data set containing time series data and unstructured data; The equipment operation data includes vibration, temperature, pressure and current, and the environmental data includes dust and humidity; The data preprocessing and feature engineering module cleans and normalizes the raw data, extracts time series features and statistical features, and labels faults based on domain knowledge, providing high-quality data for model training. The time series features include mean, variance and frequency domain features, and the statistical features include peak and skewness.

3. The AI ​​large model-based safety equipment fault prediction and intelligent maintenance decision-making system according to claim 2 is characterized in that: The AI ​​large model construction and training unit includes a model selection and architecture design module and a model training and optimization module. The model selection and architecture design module uses a deep learning model to process time series data and combines knowledge graph technology to integrate equipment mechanisms and industry rules. The model training and optimization module is based on historical fault data and real-time monitoring data, accelerates model convergence through transfer learning technology, and uses cross-validation methods to optimize hyperparameters.

4. The AI ​​large model-based safety equipment fault prediction and intelligent maintenance decision-making system according to claim 3 is characterized in that: The fault prediction and root cause location unit includes a real-time monitoring and anomaly detection module and a root cause analysis and fault diagnosis module. The real-time monitoring and anomaly detection module analyzes device data in real time through edge computing nodes, uses models to identify abnormal patterns, and triggers early warnings. The abnormal patterns include sudden vibration changes and temperature anomalies. The root cause analysis and fault diagnosis module locates the fault source based on the fault characteristics output by the model and combines the knowledge graph to provide maintenance suggestions.

5. The safety equipment fault prediction and intelligent maintenance decision-making system based on AI large model according to claim 4 is characterized in that: The intelligent maintenance decision generation unit includes a maintenance strategy recommendation module and a resource optimization and scheduling module. The maintenance strategy recommendation module generates a dynamic maintenance plan based on the equipment health status, maintenance cost and production plan. The dynamic maintenance plan includes emergency maintenance, planned downtime and spare parts replacement. The resource optimization and scheduling module intelligently adjusts maintenance strategies based on equipment usage frequency, working environment, and historical maintenance records to achieve optimal resource allocation. For equipment with stable operation, it appropriately extends maintenance cycles and reduces downtime.

6. The safety equipment fault prediction and intelligent maintenance decision-making system based on AI large model according to claim 5 is characterized in that: The dynamic optimization and self-learning unit includes a model continuous iteration module and a multi-system collaboration and standardization module. The model continuous iteration module automatically updates model parameters through real-time data and user feedback to improve prediction accuracy. The multi-system collaboration and standardization module promotes the development of standardized interfaces between industrial Internet platforms and large AI models, and supports data sharing and model reuse among multiple enterprises and multiple devices.

7. The safety equipment fault prediction and intelligent maintenance decision-making system based on AI big model according to claim 6 is characterized in that: The root cause analysis and fault diagnosis module analyzes and predicts faults in security equipment operation and maintenance based on a discrete algorithm combined with a BP neural network: Obtain the historical original data of safety equipment operation and establish an initial decision table for the historical original parameters. The historical original data is various historical fault data in the fault statistics table of each device; Calculate the correlation between each row of data in the initial decision table, simplify the data in the horizontal dimension, and when establishing the initial decision table for the historical original data, remove the repeated parameters and add the necessary characteristic parameters to form a complete initial decision table; Discretize the continuous variables after horizontal data simplification, and simplify the vertical attributes of the discretized data; Obtain input data of the BP neural network model, obtain input data of the BP neural network model; According to the input data of the BP neural network model, the BP neural network model is trained. After the BP neural network model is trained, the corresponding test samples are used to test it, and the prediction results of equipment failure are output and analyzed. Determine the neural network model to be trained, and determine the BP neural network model to be trained; According to the determined training BP neural network model, the mechanical and electrical equipment failure is predicted.

8. The safety equipment fault prediction and intelligent maintenance decision-making system based on AI large model according to claim 7 is characterized in that: The discretization of continuous variables after horizontal data simplification includes the following steps: Obtain fault feature sequence L0 and fault feature parameter sequence L i ; L0=(l0(1)), (l0(2)),…,(l0(n)) L i =(l i (1),l i (2),…,l i (n)),i=1,2,…,m Calculate the initial image of each sequence; L′ i =L i / l i (1)=(l′ i (1),l′ i (2),…,l′ i (n)),i=1,2,…,m Calculate the fault sequence feature column L0 and the fault feature parameter sequence L i The absolute value sequence of the difference between the corresponding components of the initial image; D i (r)=|l'0(r)-l' i (r)| Calculate the maximum and minimum values ​​of the absolute value sequence, respectively: Calculate the correlation coefficient: Where ξ is the resolution coefficient; Calculate the grey correlation: The horizontal dimension data is simplified, including the following steps: According to the decision table ZT = (W, C∪Z, S, f), calculate the difference matrix M of the decision table ZT n×n (ZT), get M n×n (ZT)=(c ij ) n×n The lower triangular matrix of Where i, j = 1, 2, 3, ..., n Calculate the difference function Δ and set the disjunctive normal form: Perform conjunction operation on the disjunctive normal form to obtain the difference function: Transform the conjunctive normal form of the difference function into the disjunctive normal form, and we get Then Δ l Represents the result of l attribute reduction; After obtaining the simplified processing of the fault decision table, the redundant column data is deleted to complete the simplification of the condition attributes of the vertical fault decision table.

9. The safety equipment fault prediction and intelligent maintenance decision-making system based on AI big model according to claim 8 is characterized in that: Obtaining input data of the BP neural network model includes the following steps: Calculate the grey relational degree of m groups of sample data, eliminate p groups of irrelevant data, and obtain (mp) fault influencing factors; Remove q redundant failure influencing factors and obtain (nq) failure influencing factors; The input of the BP neural network model is a (mp)×(nq) fault data decision table.

10. The safety equipment fault prediction and intelligent maintenance decision-making system based on AI large model according to claim 9 is characterized in that: Determining the neural network model to be trained includes the following steps: According to the analysis results, determine whether the BP neural network model meets the prediction requirements; If the prediction requirements are met, the trained BP neural network model is determined; if the prediction requirements are not met, S2-S5 are repeated until a BP neural network model that meets the prediction requirements is determined.

Images