Equipment maintenance man-hour prediction system and method based on association item anomaly analysis

Abstract

The invention discloses an equipment maintenance man-hour prediction system and method based on association item anomaly analysis. The prediction system comprises a data acquisition layer, a man-hour influence analysis layer, a maintenance execution anomaly analysis layer and a maintenance man-hour prediction layer. The data acquisition layer comprises a maintenance core element acquisition module and an association relationship data acquisition module; the man-hour influence analysis layer comprises an influence factor screening module and an incidence relation analysis module; the maintenance execution abnormity analysis layer comprises a real-time abnormity monitoring module and an abnormity reason positioning module; the maintenance man-hour prediction layer comprises a prediction model construction module, a real-time prediction execution module and a prediction precision evaluation module. According to the method, the personnel participation degree is low, the analysis granularity is fine, the analysis accuracy is high, emergency situations in the maintenance process can be dynamically adapted, continuous optimization of the maintenance man-hour prediction model is achieved, and the accuracy of maintenance man-hour prediction is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of maintenance time prediction technology, and in particular to a system and method for predicting equipment maintenance time based on correlation item anomaly analysis. Background Technology

[0002] Current methods for predicting equipment maintenance man-hours mainly include the stopwatch method, photographic method, work sampling method, and PTS method. The stopwatch method is suitable for operations with stable cycles; the photographic method can be used for analyzing complex short-cycle movements; the work sampling method is suitable for long-cycle or discontinuous operations; and the PTS method is often used for short-cycle and highly standardized operations. However, these methods still have the following shortcomings: (1) High level of human involvement and strong subjectivity: It relies on manual observation and judgment, such as the observer's recording of operation time and setting of evaluation coefficients in the stopwatch method, which are easily affected by subjective factors, leading to data deviation; (2) Coarse analysis granularity: The maintenance process is not broken down in detail. For example, the work sampling method only counts the percentage of effective working hours, and cannot deeply analyze the specific situation of each process, and cannot accurately identify the key links of working hours consumption. (3) Poor accuracy of analysis: The lack of comprehensive consideration of various complex factors, such as different types of equipment failures and differences in the skills of maintenance personnel, leads to a large error between the predicted results and the actual working hours. It is not helpful for maintenance improvement and cannot provide effective support for optimizing maintenance processes and reducing maintenance costs. (4) Subjective bias caused by high personnel involvement: Relying on subjective operations such as manual observation and recording and experience to set compensation coefficients, different maintenance personnel have different judgments on the same maintenance scenario, and a large human team is required for maintenance and management, resulting in distorted working hour data and low efficiency. Summary of the Invention

[0003] In view of the above-mentioned defects or deficiencies in the existing technology, it is desirable to provide a system and method for predicting equipment maintenance time based on correlation anomaly analysis, which has low personnel participation, fine analysis granularity, high analysis accuracy, can dynamically adapt to unexpected situations in the maintenance process, realize continuous optimization of maintenance time prediction model, and greatly improve the accuracy of maintenance time prediction.

[0004] This invention provides a system for predicting equipment maintenance time based on correlation item anomaly analysis, comprising a data acquisition layer, a time impact analysis layer, a maintenance execution anomaly analysis layer, and a maintenance time prediction layer; The data acquisition layer includes: The core maintenance element acquisition module acquires maintenance element data during the maintenance process; The correlation data acquisition module obtains maintenance correlation data between maintenance elements; The time impact analysis layer includes: The influencing factor screening module filters the main influencing factors of maintenance man-hours. The correlation analysis module uncovers the correlation patterns between maintenance elements from different dimensions; The maintenance execution anomaly analysis layer includes: The real-time anomaly monitoring module collects abnormal status information during the maintenance process; The anomaly cause localization module analyzes and determines the core root cause of the abnormal state; The maintenance man-hour prediction layer includes: The prediction model building module builds and optimizes the maintenance man-hour prediction model. The real-time predictive execution module predicts the maintenance man-hours for the current maintenance task. The prediction accuracy evaluation module evaluates the prediction results and optimizes the maintenance man-hour prediction model.

[0005] Furthermore, the maintenance core element acquisition module includes: The equipment information collection submodule collects equipment information including equipment model, service life, historical fault records, and historical maintenance records. The maintenance resource acquisition submodule collects maintenance resource data, including parts and maintenance tools. The personnel information collection submodule collects information on maintenance personnel, including their maintenance experience, skills, and tool proficiency ratings. The real-time execution status acquisition submodule dynamically acquires execution status data during the maintenance process, including real-time data on maintenance progress, operation time, and resource consumption. The association data acquisition module includes: The personnel-tool association data collection submodule collects association data between maintenance personnel and maintenance tools; The Problem-Parts Association Data Collection submodule collects correlation data between faults and parts. The execution status-impact factor correlation data collection submodule collects the linkage data between execution status and maintenance impact factors.

[0006] Furthermore, the impact factor screening module includes: The historical data retrieval submodule extracts historical maintenance man-hours, corresponding maintenance element data, and maintenance-related data from the database. The feature engineering submodule extracts the feature characteristics of maintenance element data and transforms maintenance-related data into quantified relational features. The factor importance assessment submodule uses the XGBoost algorithm to calculate the impact weight of each maintenance influencing factor on maintenance man-hours based on the characteristics of the maintenance element data and the relationship characteristics of the maintenance related data. The top 50% of maintenance influencing factors with the largest and smallest impact weights are selected as the main influencing factors. The abnormal factor library construction submodule is executed to include abnormal factors related to abnormal working hours in historical maintenance records into the abnormal factor library and to dynamically accumulate the abnormal factor library.

[0007] Furthermore, the relationship analysis module includes: The element association analysis submodule uses the Apriori algorithm to discover strong association combinations between repair parts, fault problems, and spare parts. The relational impact quantification submodule uses a logistic regression model to quantify the degree of influence of strongly associated combinations on maintenance man-hours. The anomaly association rule mining submodule uses the FP-Growth algorithm to mine the association rules between maintenance elements in anomaly scenarios, providing a basis for identifying anomaly factors.

[0008] Furthermore, the real-time anomaly monitoring module includes: The time deviation monitoring submodule continuously compares the real-time maintenance time with the predicted maintenance time during the maintenance process. When the deviation between the two exceeds the set threshold, it is marked as an abnormal time deviation. The abnormal operation behavior monitoring submodule monitors the operational procedures of maintenance personnel and abnormal use of maintenance tools, and identifies operational anomalies. The resource anomaly consumption monitoring submodule tracks the wear and tear of parts and the usage status of maintenance tools during the maintenance process, and identifies resource consumption anomalies.

[0009] Furthermore, the anomaly cause localization module includes: The multi-dimensional factor correlation analysis submodule combines real-time monitored abnormal data with historically accumulated abnormal factors to initially identify the abnormal factors that may have caused this anomaly. The root cause reasoning submodule uses Bayesian networks to perform probabilistic reasoning verification on the initially identified anomaly factors, and finally determines the core root cause of this anomaly. The root cause visualization submodule displays the analysis results of the core root causes.

[0010] Furthermore, the prediction model building module includes: The LSTM neural network submodule is used to build a maintenance time prediction model. The model training submodule trains the maintenance time prediction model to enable it to predict maintenance time. The model optimization submodule further improves the prediction accuracy of the maintenance man-hour prediction model by enhancing training with abnormal samples.

[0011] Furthermore, the real-time prediction execution module includes: The current maintenance task data input submodule receives maintenance element data and maintenance-related data of the current maintenance task transmitted from the data acquisition layer; The impact factor matching submodule matches the corresponding main impact factors based on the current maintenance task; The prediction calculation submodule calls the maintenance time prediction model trained by the prediction model construction module, inputs the current maintenance task and the main influencing factors into the maintenance time prediction model, and outputs the maintenance time prediction value. The anomaly warning submodule verifies the reasonableness of the predicted maintenance man-hours and triggers an alarm if there is an abnormal deviation.

[0012] Furthermore, the prediction accuracy evaluation module includes: The results comparison submodule calculates the prediction deviation, MAE, and RMSE between the predicted maintenance man-hours and the actual maintenance man-hours after the maintenance task is completed. The prediction accuracy is measured by MAE and RMSE. The deviation analysis submodule distinguishes between normal deviations and abnormal deviations caused by anomalous factors in the prediction deviations. The model iteration submodule, when the prediction deviation exceeds the set threshold, supplements the model training set with relevant data and abnormal data of the current maintenance task, retrains the maintenance time prediction model, and continuously optimizes the maintenance time prediction model.

[0013] In addition, the present invention also provides a method for predicting equipment maintenance time using the above-mentioned equipment maintenance time prediction system based on correlation item anomaly analysis.

[0014] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention adopts a multi-dimensional relational data system to break through the limitations of single elements. It adopts a dual-dimensional collection framework of core elements + relational relationships, breaking through the traditional technology mode of only collecting isolated data such as equipment and personnel, covering maintenance elements such as equipment, resources, personnel, and real-time execution status. At the same time, it accurately collects implicit relational data such as personnel-maintenance tool familiarity, fault problem-parts matching degree, and execution status-maintenance influencing factors, and constructs a full-scale basic data covering entities-relationships. It transforms unstructured relational information such as personnel's proficiency with tools and fault problems and parts matching frequency into quantifiable features, realizes the manifestation of implicit influences, solves the pain point of traditional technology that relational influences cannot be included in the analysis, and provides data support for accurate modeling.

[0015] (2) This invention breaks down data silos through multi-level association analysis technology. It employs a combined strategy of using the Apriori algorithm to mine strong associations of maintenance elements and the FP-Growth algorithm to mine abnormal associations. This strategy not only identifies the conventional association patterns of maintenance parts, faults, and parts, but also locates special association rules under abnormal scenarios, significantly improving the depth of mining the impact patterns of maintenance time. By using a dynamic construction and quantitative evaluation method for anomaly factor databases, historical anomaly data is accumulated. Combined with the maintenance impact factor importance evaluation capability of the XGBoost algorithm, the impact weight of abnormal maintenance impact factors is accurately calculated, solving the problem that traditional technologies lack quantitative analysis of the impact of anomalies and providing a quantitative basis for prediction and correction.

[0016] (3) This invention uses an anomaly-aware LSTM neural network to improve the upper limit of accuracy. An LSTM neural network is selected and combined with anomaly sample enhancement training to adapt to the temporal and abnormal fluctuation characteristics of data in maintenance scenarios. Compared with traditional regression models, it has a stronger fitting ability for dynamically changing maintenance processes. Simultaneously, the risk of invalid predictions is reduced through pre-verification of predicted values. A real-time monitoring-root cause localization-dynamic correction closed-loop response mechanism is adopted. Anomalies are captured in real time through three dimensions: time deviation, operational behavior, and resource consumption. Then, a Bayesian network is used to complete the accurate reasoning of the root cause of the anomaly. Finally, the predicted value is dynamically corrected based on the root cause, completely changing the passive mode of traditional static prediction + post-event comparison, enabling the prediction results to dynamically adapt to unexpected situations in the maintenance process.

[0017] (4) This invention adopts a data-driven, full-process self-evolution mechanism to achieve hierarchical attribution of deviations and iterative optimization of the model. By distinguishing between normal fluctuation deviations and abnormal factor deviations, abnormal-related data is added to the training set for model iteration, avoiding the inefficiency of traditional full-data retraining. This achieves lightweight and accurate evolution of the maintenance time prediction model, and realizes the value transformation of analysis results into maintenance improvement. The prediction accuracy assessment results, abnormal root cause analysis conclusions, and maintenance process optimization are directly linked to generate targeted suggestions for personnel training, tool maintenance, etc. At the same time, the abnormal factor library and association rule library are updated, upgrading the system from a maintenance time prediction tool to a maintenance efficiency optimization engine, solving the limitation of traditional technology that only predicts but does not empower.

[0018] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to restrict the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0019] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1A framework diagram of an equipment maintenance man-hour prediction system; Figure 2 This is a flowchart for data acquisition and time impact analysis; Figure 3 A flowchart for performing anomaly analysis and predicting maintenance man-hours.

[0020] The diagram is labeled as follows: 1. Data Acquisition Layer; 2. Working Hour Impact Analysis Layer; 3. Maintenance Execution Anomaly Analysis Layer; 4. Maintenance Working Hour Prediction Layer. 11. Core Maintenance Element Acquisition Module; 12. Relationship Data Acquisition Module; 21. Impact Factor Screening Module; 22. Correlation Analysis Module; 31. Real-time anomaly monitoring module; 32. Anomaly cause localization module; 41. Prediction model building module; 42. Real-time prediction execution module; 43. Prediction accuracy evaluation module; 111. Equipment Information Acquisition Submodule; 112. Maintenance Resource Acquisition Submodule; 113. Personnel Information Acquisition Submodule; 114. Real-time Execution Status Acquisition Submodule; 121. Personnel-Tool Association Data Collection Submodule; 122. Problem-Parts Association Data Collection Submodule; 123. Execution Status-Influence Factor Association Data Collection Submodule; 211. Historical Data Retrieval Submodule; 212. Feature Engineering Submodule; 213. Factor Importance Assessment Submodule; 214. Execution Anomaly Factor Library Construction Submodule; 221. Element Association Analysis Submodule; 222. Relationship Influence Quantification Submodule; 223. Anomaly Association Rule Mining Submodule; 311. Time Deviation Monitoring Submodule; 312. Operational Behavior Anomaly Monitoring Submodule; 313. Resource Anomaly Consumption Monitoring Submodule; 321. Multidimensional Factor Association Analysis Submodule; 322. Root Cause Inference Submodule; 323. Root Cause Visualization Submodule; 411. LSTM Neural Network Submodule; 412. Model Training Submodule; 413. Model Optimization Submodule; 421. Current Maintenance Task Data Input Submodule; 422. Influence Factor Matching Submodule; 423. Prediction Calculation Submodule; 424. Anomaly Early Warning Submodule; 431. Result Comparison Submodule; 432. Deviation Analysis Submodule; 433. Model Iteration Submodule. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] Please refer to Figure 1 The present invention provides an equipment maintenance time prediction system based on correlation item anomaly analysis, including a data acquisition layer 1, a time impact analysis layer 2, a maintenance execution anomaly analysis layer 3, and a maintenance time prediction layer 4; Data acquisition layer 1 includes: The maintenance core element acquisition module 11 acquires maintenance element data during the maintenance process; The correlation data acquisition module 12 acquires the maintenance correlation data between maintenance elements; The 11 core elements acquisition modules for maintenance include: The equipment information collection submodule 111 collects equipment information including equipment model, service life, historical fault records, and historical maintenance records. The maintenance resource acquisition submodule 112 collects maintenance resource data, including parts (type and quantity) and maintenance tools (type). The personnel information collection submodule 113 collects information on maintenance personnel, including maintenance experience, skills, and tool proficiency scores. The real-time execution status acquisition submodule 114 dynamically acquires execution status data during the maintenance process, including maintenance progress (currently completed maintenance steps), operation time (actual time spent on each maintenance step), and real-time data on resource consumption (number of parts used and usage of maintenance tools). The relationship data acquisition module 12 includes: The personnel-tool association data collection submodule 121 collects the association data between maintenance personnel and maintenance tools (such as the historical success rate of maintenance personnel using a certain maintenance tool, average operation time, etc.). Problem-part association data collection submodule 122 collects association data between fault problems and parts (such as the compatibility frequency and replacement probability of a certain fault problem and its corresponding parts). The execution status-influence factor association collection submodule 123 collects the linkage data between execution status and maintenance influencing factors (such as the influence of environmental factors such as workshop temperature and humidity on maintenance operation time). Among them, maintenance influencing factors are factors that can affect the execution status data during the maintenance process, such as environmental temperature and humidity, the skill level and physical condition of maintenance personnel, the type and quantity of maintenance tools, the condition of maintenance tools, the condition of water, electricity and gas for maintenance, and the condition of spare parts. Layer 2 of the time impact analysis includes: Module 21 for screening influencing factors of maintenance man-hours; Module 22, which analyzes relationships between maintenance elements, explores the patterns of relationships between them from different dimensions. Impact factor screening module 21 includes: The historical data retrieval submodule 211 extracts historical maintenance man-hours, corresponding maintenance element data, and maintenance-related data from the database. Feature engineering submodule 212 extracts the feature characteristics of maintenance element data and transforms maintenance-related data into quantitative relational features; The factor importance assessment submodule 213 uses the XGBoost algorithm to calculate the influence weight of each maintenance influencing factor on maintenance man-hours based on the characteristics of the maintenance element data and the relationship characteristics of the maintenance related data, and selects the top 50% of maintenance influencing factors with the largest influence weight as the main influencing factors. The abnormal factor library construction submodule 214 is executed to include abnormal factors related to abnormal working hours in historical maintenance records into the abnormal factor library and to dynamically accumulate the abnormal factor library. The relationship analysis module 22 includes: The element association analysis submodule 221 uses the Apriori algorithm to mine strong association combinations between repair parts, fault problems, and parts (such as the type and probability of parts that need to be replaced when a certain fault problem occurs in a certain repair part). The relational impact quantification submodule 222 uses a logistic regression model to quantify the degree of influence of strong association combinations on maintenance time (e.g., the average increase or decrease in maintenance time when a certain strong association combination occurs). The anomaly association rule mining submodule 223 uses the FP-Growth algorithm to mine the association rules between maintenance elements in anomaly scenarios (such as unqualified parts quality, non-standard operation, etc.), providing a basis for identifying anomaly factors; Maintenance execution anomaly analysis layer 3 includes: The real-time anomaly monitoring module 31 collects abnormal status information during the maintenance process; The abnormal cause localization module 32 analyzes and determines the core root cause of the abnormal state; The real-time anomaly monitoring module 31 includes: The time deviation monitoring submodule 311 continuously compares the real-time maintenance time with the predicted maintenance time during the maintenance process. When the deviation between the two exceeds the set threshold, it is marked as an abnormal time deviation. The abnormal operation behavior monitoring submodule 312 monitors the operational standardization of maintenance personnel (such as whether they operate maintenance tools according to standard procedures) and abnormal use of maintenance tools (such as incorrect use of maintenance tools) through relevant monitoring methods (such as cameras, sensors, etc.) to identify operational abnormalities. The resource anomaly consumption monitoring submodule 313 tracks the wear and tear of parts during the maintenance process (such as whether the number of parts worn out exceeds the normal range) and the usage status of maintenance tools (such as whether the maintenance tools are malfunctioning and cannot be used normally), and identifies resource consumption anomalies. The anomaly cause localization module 32 includes: The multi-dimensional factor correlation analysis submodule 321 combines real-time monitored abnormal data with historically accumulated abnormal factors to initially identify the abnormal factors that may have caused this anomaly. The root cause reasoning submodule uses Bayesian networks to perform probabilistic reasoning verification on the initially identified anomaly factors, and finally determines the core root cause of this anomaly. The root cause visualization submodule uses intuitive formats such as charts to display the analysis results of core root causes (such as root cause relationship chains), making it easier for staff to understand and process them. Maintenance time prediction layer 4 includes: Prediction model building module 41, builds and optimizes the maintenance man-hour prediction model; The real-time prediction execution module 42 predicts the maintenance man-hours for the current maintenance task. The prediction accuracy evaluation module 43 evaluates the prediction results and optimizes the maintenance man-hour prediction model. Prediction model building module 41 includes: The LSTM neural network submodule 411 constructs a maintenance time prediction model, which can handle the characteristics of time series data and adapt to the time series characteristics and abnormal fluctuation characteristics of data in maintenance scenarios. The model training submodule 412 trains the maintenance time prediction model to enable it to predict maintenance time. The model optimization submodule 413 further improves the prediction accuracy of the maintenance time prediction model by using abnormal sample augmentation training (i.e., training the maintenance time prediction model with sample data under abnormal scenarios). The real-time prediction execution module 42 includes: The current maintenance task data input submodule 421 receives maintenance element data and maintenance-related data of the current maintenance task transmitted from the data acquisition layer; The impact factor matching submodule 422 matches the corresponding main impact factors based on the current maintenance task; The prediction calculation submodule 423 calls the maintenance time prediction model trained by the prediction model construction module 41, inputs the current maintenance task and the main influencing factors into the maintenance time prediction model, and outputs the maintenance time prediction value. The anomaly warning submodule 424 performs a reasonableness check on the predicted maintenance man-hours (e.g., by comparing with the average man-hours of similar tasks in history). If there is an abnormal deviation, an alarm is triggered. Prediction accuracy evaluation module 43 includes: Result comparison submodule 431 calculates the prediction deviation, MAE (mean absolute error), and RMSE (root mean square error) between the predicted maintenance time and the actual maintenance time after the maintenance task is completed. The prediction accuracy is measured by MAE and RMSE. The deviation analysis submodule 432 distinguishes between normal deviations (such as deviations caused by normal environmental fluctuations) and abnormal deviations caused by abnormal factors (such as deviations caused by abnormal parts or abnormal operations) in the predicted deviations. The model iteration submodule 433, when the prediction deviation exceeds the set threshold, supplements the relevant data and abnormal data of the current maintenance task into the model training set, retrains the maintenance time prediction model, and continuously optimizes the maintenance time prediction model.

[0024] Also, please refer to Figures 2-3 The embodiments of the present invention also provide a method for predicting equipment maintenance time using the above-described equipment maintenance time prediction system based on correlation item anomaly analysis.

[0025] In this embodiment, the maintenance element data and maintenance-related data collected by the data acquisition layer are cleaned and standardized to remove invalid data. After unifying the data format, the data is classified and stored in the corresponding database to ensure the accuracy of maintenance time prediction.

[0026] In the description of this specification, the terms "one embodiment," "some embodiments," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0027] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A system for predicting equipment maintenance man-hours based on correlation anomaly analysis, characterized in that, It includes a data acquisition layer (1), a time impact analysis layer (2), a maintenance execution anomaly analysis layer (3), and a maintenance time prediction layer (4); The data acquisition layer (1) includes: The maintenance core element acquisition module (11) acquires maintenance element data during the maintenance process; The correlation data acquisition module (12) acquires the maintenance correlation data between maintenance elements; The time impact analysis layer (2) includes: The impact factor screening module (21) screens the main impact factors of maintenance man-hours; The correlation analysis module (22) explores the correlation patterns between maintenance elements from different dimensions; The maintenance execution anomaly analysis layer (3) includes: The real-time anomaly monitoring module (31) collects abnormal status information during the maintenance process; The abnormal cause localization module (32) analyzes and determines the core root cause of the abnormal state; The maintenance time prediction layer (4) includes: Prediction model building module (41) builds and optimizes the maintenance man-hour prediction model; The real-time prediction execution module (42) predicts the maintenance time for the current maintenance task; The prediction accuracy evaluation module (43) evaluates the prediction results and optimizes the maintenance time prediction model.

2. The equipment maintenance time prediction system based on correlation item anomaly analysis according to claim 1, characterized in that, The maintenance core element acquisition module (11) includes: The equipment information collection submodule (111) collects equipment information including equipment model, service life, historical fault records, and historical maintenance records; The maintenance resource acquisition submodule (112) collects maintenance resource data, including parts and maintenance tools; The personnel information collection submodule (113) collects information on maintenance personnel, including maintenance experience, skills, and tool proficiency scores. The real-time execution status acquisition submodule (114) dynamically acquires execution status data, including real-time data on maintenance progress, operation time, and resource consumption during the maintenance process; The association data acquisition module (12) includes: The personnel-tool association data collection submodule (121) collects the association data between maintenance personnel and maintenance tools; The Problem-Parts Association Data Collection Submodule (122) collects association data between fault problems and parts; The execution status-influence factor association collection submodule (123) collects the linkage data between execution status and maintenance influence factors.

3. The equipment maintenance time prediction system based on correlation item anomaly analysis according to claim 2, characterized in that, The impact factor screening module (21) includes: The historical data retrieval submodule (211) extracts historical maintenance man-hours and corresponding maintenance element data and maintenance-related data from the database; The feature engineering submodule (212) extracts the feature features of the maintenance element data and transforms the maintenance-related data into quantitative relational features; The factor importance assessment submodule (213) uses the XGBoost algorithm to calculate the influence weight of each maintenance influence factor on maintenance man-hours based on the characteristics of the maintenance element data and the relationship characteristics of the maintenance related data, and selects the maintenance influence factors with the largest and smallest influence weights as the main influence factors. The abnormal factor library construction submodule (214) is executed to include abnormal factors related to abnormal working hours in historical maintenance records into the abnormal factor library and to dynamically accumulate the abnormal factor library.

4. The equipment maintenance time prediction system based on correlation item anomaly analysis according to claim 3, characterized in that, The association analysis module (22) includes: The element association analysis submodule (221) uses the Apriori algorithm to mine strong association combinations between maintenance parts, fault problems, and parts; The relational impact quantification submodule (222) quantifies the impact of strong associations on maintenance man-hours using a logistic regression model; The abnormal association rule mining submodule (223) uses the FP-Growth algorithm to mine the association rules between maintenance elements in abnormal scenarios, providing a basis for identifying abnormal factors.

5. The equipment maintenance time prediction system based on correlation item anomaly analysis according to claim 4, characterized in that, The real-time anomaly monitoring module (31) includes: The time deviation monitoring submodule (311) continuously compares the real-time maintenance time with the predicted maintenance time during the maintenance process. When the deviation between the two exceeds the set threshold, it is marked as an abnormal time deviation. The abnormal operation behavior monitoring submodule (312) monitors the operation standardization of maintenance personnel and abnormal use of maintenance tools, and identifies operation-related abnormalities; The resource anomaly consumption monitoring submodule (313) tracks the wear and tear of parts and the usage status of maintenance tools during the maintenance process, and identifies resource consumption anomalies.

6. The equipment maintenance time prediction system based on correlation item anomaly analysis according to claim 5, characterized in that, The anomaly cause localization module (32) includes: The multi-dimensional factor correlation analysis submodule (321) combines real-time monitored abnormal data with historically accumulated abnormal factors to initially identify the abnormal factors that may cause this abnormality. The root cause reasoning submodule (322) uses a Bayesian network to perform probabilistic reasoning verification on the initially identified abnormal factors, and finally determines the core root cause of this anomaly. The root cause visualization submodule (323) displays the analysis results of the core root causes.

7. The equipment maintenance time prediction system based on correlation item anomaly analysis according to claim 6, characterized in that, The prediction model construction module (41) includes: The LSTM neural network submodule (411) is used to construct a maintenance time prediction model; The model training submodule (412) trains the maintenance time prediction model to enable it to predict maintenance time. The model optimization submodule (413) further improves the prediction accuracy of the maintenance time prediction model by using abnormal sample enhancement training.

8. The equipment maintenance time prediction system based on correlation item anomaly analysis according to claim 7, characterized in that, The real-time prediction execution module (42) includes: The current maintenance task data input submodule (421) receives maintenance element data and maintenance-related data of the current maintenance task transmitted from the data acquisition layer; The impact factor matching submodule (422) matches the corresponding main impact factors based on the current maintenance task; The prediction calculation submodule (423) calls the maintenance time prediction model trained by the prediction model construction module, inputs the current maintenance task and the main influencing factors into the maintenance time prediction model, and outputs the maintenance time prediction value. The abnormal warning submodule (424) performs a reasonableness check on the predicted maintenance time value. If there is an abnormal deviation, an early warning is triggered.

9. The equipment maintenance time prediction system based on correlation item anomaly analysis according to claim 8, characterized in that, The prediction accuracy evaluation module (43) includes: The results comparison submodule (431) calculates the prediction deviation, MAE and RMSE between the predicted maintenance time and the actual maintenance time after the maintenance task is completed, and measures the prediction accuracy by using MAE and RMSE. The deviation analysis submodule (432) distinguishes between normal deviations and abnormal deviations caused by abnormal factors in the prediction deviations; The model iteration submodule (433) supplements the model training set with relevant data and abnormal data of the current maintenance task when the prediction deviation exceeds the set threshold, retrains the maintenance time prediction model, and continuously optimizes the maintenance time prediction model.

10. A method for predicting equipment maintenance time using the equipment maintenance time prediction system based on correlation item anomaly analysis as described in any one of claims 1-9.

Images