Shield tunneling machine maintenance management system

Abstract

The invention discloses a shield tunneling machine maintenance management system, relates to the technical field of shield tunneling machine maintenance, and solves the technical problem that a maintenance scheme and personnel configuration cannot be determined by automatically identifying fault conditions according to shield tunneling machine fault information acquired on site in the prior art. The system comprises a shield fault database, a maintenance guidance database, a maintenance personnel database, a fault identification module, a scheme generation module, a scheme selection module, a scheduling module, a maintenance recording module, a model updating module and a user side. According to the invention, fault identification, maintenance flow recommendation, maintenance scheme determination and specific maintenance personnel notification can be realized, automation of maintenance management is realized, manual statistics, adjustment and arrangement are not needed, and the accuracy of fault identification, the accuracy of the pushed maintenance scheme and the rationality of personnel arrangement are ensured.

Description

Technical Field

[0001] The invention relates to the technical field of shield machine maintenance, and in particular to a shield machine maintenance management system. Background Art

[0002] With the rapid development of my country's shield industry, the number of shield machines has increased rapidly. Long-term use of shield machines and rational maintenance of shield machines have become a trend. Effectively integrating equipment maintenance, moving equipment from front-line operations to factory maintenance and then to front-line operations, making service a line, efficiently using shield machines, and scientifically managing shield machines are the development trends of enterprises.

[0003] Due to the complex structure of the shield machine, the types of faults involved are also relatively numerous. Manual confirmation of the fault type and treatment plan takes a long time, affecting the working efficiency of the shield machine. How to promptly confirm the fault type, formulate a treatment plan and arrange corresponding maintenance personnel to perform maintenance when a shield machine fails, so as to optimize the shield machine maintenance process, speed up the shield machine fault handling speed, and ensure that the shield machine can resume normal operation as soon as possible, is a problem that needs to be solved at present. Summary of the invention

[0004] In order to solve the problems existing in the above-mentioned prior art, the present invention provides a shield machine maintenance management system to solve the technical problem that the prior art cannot automatically identify fault conditions based on shield machine fault information collected on-site, and clarify maintenance plans and personnel allocation.

[0005] A shield machine maintenance management system includes a shield machine fault database, a maintenance instruction database, a maintenance personnel database, a fault identification module, a solution generation module, a solution selection module, a scheduling module, a maintenance record module, a model update module and a user terminal.

[0006] The shield machine fault database is used to store the collected historical fault data of the shield machine, and the fault data includes the fault area, main fault phenomenon, fault parameters and fault cause;

[0007] The maintenance instruction database is used to store recommended maintenance procedures for different fault conditions;

[0008] The maintenance personnel database is used to store maintenance personnel information, which includes length of service, working status, and experience value for each fault area;

[0009] The fault identification module is used to generate predicted fault conditions based on the on-site fault data of the shield machine acquired in real time;

[0010] The solution generation module is used to generate multiple preliminary maintenance solutions according to the predicted fault conditions in combination with the maintenance guidance database and the maintenance personnel database;

[0011] The scheme selection module is used to calculate the estimated maintenance time of each preliminary maintenance scheme and determine the final maintenance scheme according to the estimated maintenance time;

[0012] The scheduling module notifies the corresponding maintenance personnel through the user terminal according to the final maintenance plan, and pushes the predicted fault situation and the final maintenance plan to the maintenance personnel;

[0013] The maintenance record module is used to assist maintenance personnel in feeding back actual fault conditions and actual maintenance processes, and feeding back actual fault conditions and actual maintenance processes to the shield fault database and maintenance guidance database respectively;

[0014] The model updating module is used to update the fault identification module based on the actual fault situation related data accumulated in the shield fault database;

[0015] The user end is used for maintenance personnel, maintenance personnel to interact with other modules.

[0016] The shield fault database includes a historical shield data sub-database, a predicted fault condition sub-database and a training database. The historical shield data sub-database includes pre-processed historical fault data of the shield machine. The predicted fault condition sub-database includes all predicted fault conditions output by the fault identification module and the corresponding on-site fault data of the shield machine. The training database includes original training data and real-time training data. The original training data is produced based on historical shield fault data, and the real-time training data is produced based on actual fault conditions and the corresponding predicted fault conditions in the predicted fault condition sub-database.

[0017] Preferably, the fault identification module includes a fault inference sub-model and a semantic extraction sub-model, and the fault identification module has the following working process:

[0018] The on-site monitoring system collects the working conditions and working parameters of the shield machine in real time. When a warning or alarm message is found in the shield machine, the fault-related information is automatically uploaded to the fault identification module. Then the semantic extraction sub-model identifies and extracts key information based on the information uploaded by the on-site monitoring system, including the main fault phenomena and fault parameters, and transmits it to the fault inference sub-model. The fault inference sub-model infers the fault area and fault cause based on the main fault phenomena and fault parameters, and then packages the fault area, main fault phenomena and fault causes and transmits them to the semantic extraction sub-model. The semantic extraction sub-model converts them into a unified array form as a predicted fault condition and sends it to the solution generation module.

[0019] Preferably, after the solution generation module collects the predicted fault conditions, it extracts relevant recommended maintenance processes from the maintenance guidance database according to the predicted fault conditions, clarifies the requirements of maintenance personnel and maintenance tools based on the recommended maintenance processes, and automatically generates multiple preliminary maintenance plans based on the current maintenance personnel conditions recorded in the maintenance personnel database. The preliminary maintenance plans include maintenance personnel information, the number of maintenance personnel, and the number of tools; the maintenance personnel information includes the maintenance personnel's age and the maintenance areas that the maintenance personnel are good at;

[0020] Preferably, after the scheme selection module calculates the estimated maintenance duration of all preliminary maintenance schemes, the preliminary maintenance schemes with estimated maintenance durations greater than a maintenance duration threshold are deleted, and the remaining preliminary maintenance schemes are used as alternative schemes. The duration threshold is set in advance by maintenance personnel based on expert experience.

[0021] Then read the maintenance personnel information in all alternative plans, send information to the corresponding maintenance personnel user end to confirm whether they have other work arrangements next, so as to determine the real-time available maintenance personnel. If a maintenance personnel in an alternative plan has other work arrangements next, delete the maintenance personnel from the alternative plan and recalculate the estimated maintenance time. Finally, select the plan with the shortest estimated maintenance time among the qualified alternative plans as the final maintenance plan and send it to the allocation module.

[0022] The deployment module sends the maintenance task details to the corresponding maintenance personnel user terminal according to the maintenance personnel information in the final maintenance plan. The maintenance task details include the fault area, main fault phenomenon, fault parameters, fault cause, recommended maintenance process, maintenance personnel configuration, tool configuration and estimated maintenance time, and adjusts the working status of the corresponding maintenance personnel in the maintenance personnel database to busy.

[0023] Preferably, after the maintenance personnel confirms the completion of the maintenance task through the user terminal, the maintenance record module automatically pops up a maintenance feedback interface, which includes a fault registration sub-interface, a maintenance record sub-interface and a personnel configuration sub-interface in sequence. The fault registration sub-interface involves the fault prediction content in the final maintenance plan, including the fault area, the main fault phenomenon, the fault parameters, and the fault cause. The maintenance personnel judge whether the actual fault situation is consistent with the fault prediction content. If it is consistent, click Recommendation. If it is wrong or insufficient, click Feedback to make adjustments based on the content of the final maintenance plan, and transmit the actual fault situation to the training data sub-library of the shield fault database through the maintenance record module as the actual fault situation for storage;

[0024] Preferably, the maintenance record sub-interface involves a recommended maintenance process, and the recommended maintenance process is updated according to the data input by the maintenance personnel in the fault registration sub-interface. The specific updating process is: if the maintenance personnel confirms in the fault registration sub-interface that the actual fault situation is consistent with the fault prediction content, the maintenance record sub-interface will display the recommended maintenance process in the final maintenance plan; if the maintenance personnel confirms that it is inconsistent in the fault registration sub-interface, the maintenance record module will automatically retrieve the recommended maintenance process corresponding to the fault situation from the maintenance guidance database based on the actual fault situation registered by the maintenance personnel in the fault registration sub-interface, and display it in the maintenance record sub-interface.

[0025] The maintenance personnel determine whether the updated recommended maintenance process is consistent with the actual workflow. If so, they click to confirm. If not, they can choose to adjust the maintenance process based on the recommended maintenance process, or re-record the maintenance process as the actual maintenance workflow; the maintenance record module feeds back the corresponding data collected by the maintenance record sub-interface to the maintenance guidance database.

[0026] Preferably, the maintenance guidance database compares the actual maintenance process fed back by the maintenance personnel with the recommended maintenance process, calculates the similarity, divides the similarity range into three levels: high, medium and low, and stores the similarity and the corresponding actual maintenance process in different databases according to the similarity level, specifically including three databases of high, medium and low; wherein the similarity of the input medium and low levels needs to be additionally judged whether the predicted fault condition corresponding to the actual maintenance process corresponding to the similarity is consistent with the actual fault condition, and if they are consistent, the similarity and the corresponding actual maintenance process are placed in the corresponding database; after the amount of data in the three databases reaches a basic amount, the average value of all similarities is calculated regularly, and if the average value of all similarities is lower than the basic value, the data in the high, medium and low databases are statistically processed;

[0027] Statistics include the proportion of data in each database a ,t b ,t c If t a >α and t c >t b , the similarity of the three databases, the similarity in the low-level database and the corresponding actual maintenance process are transmitted to the maintenance personnel to determine whether the operation of the relevant maintenance personnel is wrong. If the maintenance personnel determine that the operation of the relevant maintenance personnel is wrong, the prompt information is fed back to the user end; if t a <α and t c >β, the similarity of the three databases, the similarity in the lower-level database and the corresponding actual maintenance process are fed back to the maintenance personnel to determine whether the recommended maintenance process needs to be modified; if t a <α and t b>β, the similarities of the three databases, the similarities in the middle-level database and the corresponding actual maintenance process are transmitted to the maintenance personnel to determine whether the recommended maintenance process needs to be modified, where α is the first threshold and β is the second threshold.

[0028] Preferably, the statistics also include counting the average similarity value y in each database a ,y b ,y c After the maintenance personnel receive the feedback information from the maintenance guidance database, if they think that the feedback information does not match the actual work, the maintenance guidance database will calculate the average similarity value y of each database. a ,y b ,y c Feedback is provided to maintenance personnel and maintenance personnel are prompted to adjust the similarity ranges at different levels.

[0029] The personnel configuration sub-interface involves the ratio of the expected maintenance time to the actual maintenance time. If the ratio is not within the qualified range, the current values ​​of the calculation parameters for calculating the expected maintenance time will pop up, including the recommended length of service for fault repair, the recommended number of maintenance personnel, length of service impact parameters, area impact parameters and quantity impact parameters. The maintenance personnel input the expected value with reference to the current value, and the maintenance record module collects the corresponding expected value and transmits it to the solution generation module for storage.

[0030] Preferably, the model updating module includes a historical model library, which is used to count the content fed back to the training data sub-library by the maintenance record module in real time, and judge whether the similarity between the content in the training data sub-library and the corresponding content in the fault prediction sub-library meets the standard. If the standard is met, no operation is performed. If the standard is not met, it is judged whether the substandard data in the training data sub-library meets the retraining requirements. If the standard is met, the substandard data is extracted from the shield fault database to train and optimize the fault inference sub-model and the semantic extraction sub-model in the fault identification module.

[0031] The historical model library stores the fault inference sub-model and semantic extraction sub-model obtained from historical training. When performing corresponding training, the fault inference sub-model and semantic extraction sub-model versions currently used by the fault identification module are extracted from the historical model library for model optimization. After the optimization is completed, they are deployed to the fault identification module and corresponding backup is performed in the historical model library.

[0032] Preferably, the fault inference sub-model and the semantic extraction sub-model are trained, and the specific training process is as follows:

[0033] The fault inference sub-model is built based on CNN, and the CNN is inferred and trained using a fault-cause sample set. The fault cause sample set includes a variety of main fault phenomena, fault parameters and fault causes involved in each fault area, and a logical relationship between the main fault phenomena, fault parameters and the fault area and fault causes is established, so that the trained CNN can infer the fault area and fault cause based on the main fault phenomena and fault parameters.

[0034] The semantic extraction sub-model is built based on Transformer, and Transformer is trained in stages. First, the Transformer is semantically trained using a preprocessed sample set, which includes original fault data and annotated data. The annotated data clarifies the fault area, main fault phenomenon, and fault cause based on the original fault data. The semantically trained Transformer can give accurate corresponding contents of the fault area, main fault phenomenon, fault parameters, and fault cause based on the original fault data.

[0035] The semantically trained Transformer is then inferred and trained using a standard sample set, where the standard sample set includes annotated content data and array data. The annotated content data includes a brief fault area, main fault phenomenon, fault parameters and fault cause for each fault, and the array data includes the annotated content data of each fault and the corresponding array form data. The inference-trained Transformer can process the original fault data into annotated data, and then convert the annotated data into a unified array form.

[0036] The beneficial effects of the present invention include:

[0037] 1. This application designs a shield machine maintenance management system that can identify faults, recommend maintenance processes, determine maintenance plans and notify specific maintenance personnel through on-site fault-related data collected by the on-site monitoring system, thereby realizing the automation of maintenance management without the need for manual statistical adjustment and arrangement, ensuring the accuracy of fault identification, the accuracy of the pushed maintenance plan and the rationality of personnel arrangements.

[0038] 2. The maintenance record module collects maintenance personnel’s feedback on the fault conditions automatically identified by the system, as well as the clear maintenance plans and maintenance personnel configurations, so as to update the fault identification module, fault guidance database and solution selection module in real time to ensure the accuracy and applicability of the results pushed by the system.

[0039] 3. The model update module is used to update and optimize the model in the fault identification module in real time based on the data fed back by the maintenance record module, fully ensuring the accuracy of fault identification, making subsequent solution generation and personnel allocation more accurate and reasonable. At the same time, the model update module stores historically trained models. If the processing effect of the fault identification module degrades after model training, the model can be backtracked to fully ensure the accuracy of fault identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 The present invention is an operation flow chart of a shield machine maintenance management system involved in an embodiment. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0042] A shield machine maintenance management system includes a shield machine fault database, a maintenance instruction database, a maintenance personnel database, a fault identification module, a solution generation module, a solution selection module, a scheduling module, a maintenance record module, a model update module and a user terminal.

[0043] The shield machine fault database is used to store the collected historical fault data of the shield machine, and the fault data includes the fault area, main fault phenomenon, fault parameters and fault cause;

[0044] The maintenance instruction database is used to store recommended maintenance procedures for different fault conditions;

[0045] The maintenance personnel database is used to store maintenance personnel information, which includes length of service, working status, and experience value for each fault area;

[0046] The fault identification module is used to generate predicted fault conditions based on the on-site fault data of the shield machine acquired in real time;

[0047] The solution generation module is used to generate multiple preliminary maintenance solutions according to the predicted fault conditions in combination with the maintenance guidance database and the maintenance personnel database;

[0048] The scheme selection module is used to calculate the estimated maintenance time of each preliminary maintenance scheme and determine the final maintenance scheme according to the estimated maintenance time;

[0049] The scheduling module notifies the corresponding maintenance personnel through the user terminal according to the final maintenance plan, and pushes the predicted fault situation and the final maintenance plan to the maintenance personnel;

[0050] The maintenance record module is used to assist maintenance personnel in feeding back actual fault conditions and actual maintenance processes, and feeding back actual fault conditions and actual maintenance processes to the shield fault database and maintenance guidance database respectively;

[0051] The model updating module is used to update the fault identification module based on the actual fault situation related data accumulated in the shield fault database;

[0052] The user end is used for maintenance personnel, maintenance personnel to interact with other modules.

[0053] In another embodiment, the shield fault database includes a historical shield data sub-library, a predicted fault condition sub-library and a training database. The historical shield data sub-library is obtained by corresponding preprocessing of historical shield machine fault-related data extracted from the shield machine big data cloud platform. The predicted fault condition sub-library is the predicted fault condition output by the fault identification module. The training database includes original training data and real-time training data. The original training data is produced based on historical shield fault data, and the real-time training data is produced based on the predicted fault condition and the actual fault condition collected by the maintenance record sub-module.

[0054] Each fault data in the historical shield data sub-database includes fault area, main fault phenomenon, fault parameters and fault cause, etc. The fault area includes cutter head, main drive system, foam system, propulsion system, shield tail, screw conveyor, hydraulic system, segment installation system, belt conveyor system, grease pump station, etc. The data is preprocessed, including standardization, normalization and encoding, and the fault data is organized into array form for storage. For example, a certain cutter head fault data is organized into sample data and represented as [0, 1, 0, 1, 0, 0, 1, 0, 1, 1, 0, 1], where the first 4 bits represent the fault area, the first 8 bits represent the main fault phenomenon and all 12 bits represent the fault cause.

[0055] Maintenance instruction database; the maintenance instruction database includes recommended maintenance procedures for various types of faults and corresponding procedure accuracy, and the recommended maintenance procedure index is an array including fault area, main fault phenomenon and fault cause.

[0056] Maintenance personnel database; the maintenance personnel database includes information of all maintenance personnel on duty, each maintenance personnel information includes age, length of service, working status, and experience value for each fault area, among which the working status includes busy, reporting and waiting to be arranged, among which the highest experience value is 1, indicating that it is often handled, and the lowest is 0, indicating that it has never been handled.

[0057] The specific workflow of the trained fault identification module is as follows:

[0058] The on-site monitoring system collects the working conditions and working parameters of the shield machine in real time. When a warning or alarm information is found in the shield machine, it is automatically uploaded to the fault identification module. Then the semantic extraction sub-model identifies and extracts the key main fault phenomena and fault parameters based on the information uploaded by the on-site monitoring system and transmits them to the fault inference sub-model. The fault inference sub-model infers the fault area and fault cause based on the main fault phenomena and fault parameters, and then packages the fault area, main fault phenomena and fault causes and transmits them to the semantic extraction sub-model. The semantic extraction sub-model converts the collected content into a unified array form as a predicted fault condition and sends it to the solution generation module.

[0059] In another embodiment, after the solution generation module collects the predicted fault conditions, it extracts relevant recommended maintenance processes from the maintenance instruction database according to the predicted fault conditions, clarifies the requirements for maintenance personnel and maintenance tools based on the recommended maintenance processes, and automatically combines and generates multiple preliminary maintenance solutions based on the current maintenance personnel conditions recorded in the maintenance personnel database.

[0060] The preliminary maintenance plan includes maintenance personnel information, the number of maintenance personnel, and the number of tools; maintenance personnel information includes maintenance personnel age and maintenance areas that maintenance personnel are good at;

[0061] In another embodiment, the solution selection module calculates the estimated maintenance time according to the information of the preliminary maintenance solution: the estimated maintenance time is related to the fault guidance maintenance time, the maintenance personnel's age and the number of maintenance personnel involved.

[0062] After the scheme selection module calculates the estimated maintenance time of all scheme selection modules, the preliminary maintenance schemes with estimated maintenance time longer than the maintenance time threshold are deleted, and the remaining preliminary maintenance schemes are used as alternative schemes. The time threshold is set in advance by the maintenance personnel based on expert experience.

[0063] The plan selection module reads the maintenance personnel information in all alternative plans, and sends information to the corresponding maintenance personnel user end to confirm whether they have other work arrangements next, so as to determine the real-time available maintenance personnel. If a maintenance personnel in an alternative plan has other work arrangements next, the maintenance personnel will be deleted from the alternative plan and the estimated maintenance time will be recalculated. Finally, the plan with the shortest estimated maintenance time among the qualified alternative plans will be selected as the final maintenance plan and sent to the allocation module.

[0064] In another embodiment, the deployment module sends maintenance task details to the corresponding maintenance personnel user terminal according to the maintenance personnel information in the final maintenance plan, including the fault area, main fault phenomenon, fault parameters, fault cause, recommended maintenance process, maintenance personnel configuration, tool configuration and estimated maintenance time, and adjusts the status of the corresponding maintenance personnel in the maintenance personnel database to busy.

[0065] In another embodiment, after the fault repair is completed, the maintenance personnel clicks on the maintenance completion through the user terminal, and the maintenance record module automatically pops up the maintenance feedback interface, including the fault registration sub-interface, the maintenance record sub-interface and the personnel configuration sub-interface. The fault registration sub-interface involves the fault prediction content in the final maintenance plan, including the fault area, the main fault phenomenon, the fault parameters, and the fault cause. The maintenance personnel judge whether the actual fault situation is consistent with the fault prediction content. If it is consistent, the maintenance personnel clicks on the recommendation. If it is wrong or insufficient, the maintenance personnel clicks on the feedback to adjust the final maintenance plan content as the actual fault situation. The actual fault situation is transmitted to the training data sub-library of the shield fault database through the maintenance record module for storage;

[0066] The maintenance record sub-interface involves a recommended maintenance process, and the recommended maintenance process is updated according to the data input by the maintenance personnel in the fault registration sub-interface. The specific updating process is: if the maintenance personnel confirms in the fault registration sub-interface that the actual fault situation is consistent with the fault prediction content, the maintenance record sub-interface will display the recommended maintenance process in the final maintenance plan; if the maintenance personnel confirms that it is inconsistent in the fault registration sub-interface, the maintenance record module obtains the recommended maintenance process corresponding to the fault situation from the maintenance guidance database based on the actual fault situation registered by the maintenance personnel in the fault registration sub-interface and displays it in the maintenance record sub-interface.

[0067] The maintenance personnel determine whether the updated recommended maintenance process is consistent with the actual workflow. If so, they click to confirm. If not, they can choose to adjust the maintenance process based on the recommended maintenance process, or re-record the maintenance process as the actual maintenance workflow; the maintenance record module feeds back the corresponding data collected by the maintenance record sub-interface to the maintenance guidance database.

[0068] In another embodiment, the maintenance guidance database compares the actual maintenance process fed back by the maintenance personnel with the recommended maintenance process, calculates the similarity, divides the similarity range into three levels: high, medium and low, and stores the similarity and the corresponding actual maintenance process in different databases according to the similarity level, specifically including three databases of high, medium and low; wherein the similarity input at the medium level and the low level requires additional judgment on whether the actual maintenance process corresponding to the similarity corresponds to the predicted fault condition and the actual fault condition, and if they are consistent, the similarity and the corresponding actual maintenance process are placed in the corresponding database; after the data volume in the three databases reaches the basic volume, the average value of all similarities is calculated regularly, and if the average value of all similarities is lower than the basic value, the data in the high, medium and low databases are statistically processed; the statistics include statistically calculating the proportion of the data volume in each database t a ,t b ,t c If t a >α and t c >t b , the similarity of the three databases, the similarity in the low-level database and the corresponding actual maintenance process are transmitted to the maintenance personnel to determine whether the operation of the relevant maintenance personnel is wrong. If the maintenance personnel determine that the operation of the relevant maintenance personnel is wrong, the prompt information is fed back to the user end; if t a <α and t c >β, the similarity of the three databases, the similarity in the lower-level database and the corresponding actual maintenance process are fed back to the maintenance personnel to determine whether the recommended maintenance process needs to be modified; if t a <α and t b >β, the similarity of the three databases, the similarity in the middle-level database and the corresponding actual maintenance process are transmitted to the maintenance personnel to determine whether the recommended maintenance process needs to be modified. Specifically, t a +t b +t c =1; α is the first threshold, β is the second threshold, both of which are set according to the experience of maintenance personnel; in this embodiment, α is 0.8; β is 0.4.

[0069] The statistics also include counting the average similarity value y in each database a ,y b ,y c After the maintenance personnel receive the feedback information from the maintenance guidance database, if they think that the feedback information does not match the actual work, the maintenance guidance database will calculate the average similarity value y of each database. a ,y b ,y c Feedback is provided to maintenance personnel and maintenance personnel are prompted to adjust the similarity ranges at different levels.

[0070] In another embodiment, the personnel configuration sub-interface involves the ratio of the expected maintenance time to the actual maintenance time. If the ratio is not within the qualified range, the current values ​​of various calculation parameters for calculating the expected maintenance time will pop up, including the recommended length of service for fault repair, the recommended number of maintenance personnel, length of service impact parameters, area impact parameters and quantity impact parameters. The maintenance personnel input the expected value with reference to the current value, and the maintenance record module collects the corresponding expected value and transmits it to the solution generation module for storage.

[0071] In another embodiment, the model updating module includes a historical model library, which is used to count the content fed back to the training data sub-library by the maintenance record module in real time, and judge whether the similarity between the content in the training data sub-library and the corresponding content in the fault prediction sub-library meets the standard. If the standard is met, no operation is performed. If the standard is not met, it is judged whether the substandard data in the training data sub-library meets the retraining requirements (for example, the substandard data reaches 100). If the standard is met, relevant substandard data is extracted from the training data sub-library to train and optimize the fault prediction sub-model and the semantic extraction sub-model.

[0072] The historical model library stores the fault inference sub-model and semantic extraction sub-model obtained from historical training. When performing corresponding training, the fault inference sub-model and semantic extraction sub-model versions currently used by the fault identification module are extracted from the historical model library for model optimization. After the optimization is completed, they are deployed to the fault identification module and corresponding backup is performed in the historical model library.

[0073] In another embodiment, the fault identification module includes a fault inference sub-model and a semantic extraction sub-model, and the fault inference sub-model and the semantic extraction sub-model are trained respectively. The specific training process is as follows:

[0074] The fault inference sub-model is built based on CNN, and the CNN is inferred and trained using a fault-cause sample set. The fault cause sample set includes a variety of main fault phenomena, fault parameters and fault causes involved in each fault area, and a logical relationship between the main fault phenomena, fault parameters and the fault area and fault causes is established, so that the trained CNN can infer the fault area and fault cause based on the main fault phenomena and fault parameters.

[0075] The semantic extraction sub-model is built based on Transformer, and Transformer is trained in stages. First, the Transformer is semantically trained using a preprocessed sample set, which includes original fault data and annotated data. The annotated data clarifies the fault area, main fault phenomenon, and fault cause based on the original fault data. The semantically trained Transformer can give accurate corresponding contents of the fault area, main fault phenomenon, fault parameters, and fault cause based on the original fault data.

[0076] The semantically trained Transformer is then inferred and trained using a standard sample set, where the standard sample set includes annotated content data and array data. The annotated content data includes a brief fault area, main fault phenomenon, fault parameters and fault cause for each fault, and the array data includes the annotated content data of each fault and the corresponding array form data. The inference-trained Transformer can process the original fault data into annotated data, and then convert the annotated data into a unified array form.

[0077] The above-mentioned embodiments only express the specific implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the protection scope of the present application. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the technical solution concept of the present application, and these all belong to the protection scope of the present application.

Claims

1. A shield machine maintenance management system, characterized in that: It includes shield fault database, maintenance instruction database, maintenance personnel database, fault identification module, solution generation module, solution selection module, scheduling module, maintenance record module, model update module and user end. The shield machine fault database is used to store the collected historical fault data of the shield machine, and the fault data includes the fault area, main fault phenomenon, fault parameters and fault cause; The maintenance instruction database is used to store recommended maintenance procedures for different fault conditions; The maintenance personnel database is used to store maintenance personnel information, which includes length of service, working status, and experience value for each fault area; The fault identification module is used to generate predicted fault conditions based on the on-site fault data of the shield machine acquired in real time; The solution generation module is used to generate multiple preliminary maintenance solutions according to the predicted fault conditions in combination with the maintenance guidance database and the maintenance personnel database; The scheme selection module is used to calculate the estimated maintenance time of each preliminary maintenance scheme and determine the final maintenance scheme according to the estimated maintenance time; The scheduling module notifies the corresponding maintenance personnel through the user terminal according to the final maintenance plan, and pushes the predicted fault situation and the final maintenance plan to the maintenance personnel; The maintenance record module is used to assist maintenance personnel in feeding back actual fault conditions and actual maintenance processes, and feeding back actual fault conditions and actual maintenance processes to the shield fault database and maintenance guidance database respectively; The model updating module is used to update the fault identification module based on the actual fault situation related data accumulated in the shield fault database; The user end is used for maintenance personnel, maintenance personnel to interact with other modules.

2. A shield machine maintenance management system according to claim 1, characterized in that: The fault identification module includes a fault inference sub-model and a semantic extraction sub-model. The fault identification module has the following working process: The on-site monitoring system collects the working conditions and working parameters of the shield machine in real time. When a warning or alarm message is found in the shield machine, the fault-related information is automatically uploaded to the fault identification module. Then the semantic extraction sub-model identifies and extracts key information including the main fault phenomena and fault parameters based on the information uploaded by the on-site monitoring system and transmits it to the fault inference sub-model. The fault inference sub-model infers the fault area and fault cause based on the main fault phenomena and fault parameters, and then packages the fault area, main fault phenomena and fault causes and transmits them to the semantic extraction sub-model. The semantic extraction sub-model converts them into a unified array form and sends them to the solution generation module as a predicted fault situation.

3. A shield machine maintenance management system according to claim 1, characterized in that: The solution generation module extracts relevant recommended maintenance processes from the maintenance guidance database according to the predicted fault conditions, clarifies the requirements for maintenance personnel and maintenance tools based on the recommended maintenance processes, and then automatically combines and generates multiple preliminary maintenance plans based on the working status of maintenance personnel in the maintenance personnel database. The preliminary maintenance plans include maintenance personnel information, the number of maintenance personnel, and the number of tools; the maintenance personnel information includes the maintenance personnel's age and the maintenance areas that the maintenance personnel are good at.

4. A shield machine maintenance management system according to claim 1, characterized in that: The scheme selection module deletes the preliminary maintenance schemes whose estimated maintenance time is greater than the maintenance time threshold, takes the remaining preliminary maintenance schemes as alternative schemes, then reads the maintenance personnel information in all alternative schemes, sends information to the corresponding maintenance personnel user terminal to confirm whether they have other work arrangements next, thereby determining the real-time available maintenance personnel; if a maintenance personnel in an alternative scheme has other work arrangements next, the maintenance personnel is deleted from the alternative scheme and the estimated maintenance time is recalculated; finally, the scheme with the shortest estimated maintenance time among the alternative schemes is selected as the final maintenance scheme and sent to the allocation module.

5. A shield machine maintenance management system according to claim 1, characterized in that: After confirming that the maintenance task is completed, the maintenance record module automatically pops up the maintenance feedback interface, which includes the fault registration sub-interface, the maintenance record sub-interface and the personnel configuration sub-interface in sequence. The fault registration sub-interface involves the predicted fault situation. The maintenance personnel judge whether the actual fault situation is consistent with the predicted fault situation. If they are consistent, click Recommend. If there is an error or insufficiency, click Feedback to make adjustments based on the final maintenance plan content as the actual fault situation. The actual fault situation is transmitted to the shield fault database for storage through the maintenance record module.

6. A shield machine maintenance management system according to claim 1, characterized in that: The maintenance record sub-interface involves a recommended maintenance process, and the recommended maintenance process is updated according to the data input by the maintenance personnel in the fault registration sub-interface. The specific updating process is as follows: if the maintenance personnel confirms in the fault registration sub-interface that the actual fault situation is consistent with the fault prediction content, the maintenance record sub-interface displays the recommended maintenance process in the final maintenance plan; if the maintenance personnel confirms inconsistency in the fault registration sub-interface, the maintenance record module obtains the recommended maintenance process corresponding to the fault situation from the maintenance guidance database based on the actual fault situation registered by the maintenance personnel in the fault registration sub-interface, and displays it in the maintenance record sub-interface; The maintenance personnel determines whether the updated recommended maintenance process is consistent with the actual work process. If so, they click Confirm. If not, they can choose to adjust the maintenance process based on the recommended maintenance process, or re-record the maintenance process as the actual maintenance process. The maintenance record module feeds back the corresponding data collected by the maintenance record sub-interface to the maintenance guidance database.

7. A shield machine maintenance management system according to claim 1, characterized in that: The maintenance guidance database compares the actual maintenance process fed back by the maintenance personnel with the recommended maintenance process, calculates the similarity, divides the similarity range into three levels: high, medium and low, and stores the similarity and the corresponding actual maintenance process in different databases according to the similarity level, specifically including three databases of high, medium and low; wherein the similarity of the input medium and low levels needs to be additionally judged whether the predicted fault condition corresponding to the actual maintenance process corresponding to the similarity is consistent with the actual fault condition, and if they are consistent, the similarity and the corresponding actual maintenance process are placed in the corresponding database; after the data volume in the three databases reaches the basic volume, the average value of all similarities is calculated regularly, and if the average value of all similarities is lower than the basic value, the data in the high, medium and low databases are statistically processed; the statistics include counting the proportion of the data volume in each database t a ,t b ,t c If t a >α and t c >t b , the similarity of the three databases, the similarity in the low-level database and the corresponding actual maintenance process are transmitted to the maintenance personnel to determine whether the operation of the relevant maintenance personnel is wrong. If the maintenance personnel determine that the operation of the relevant maintenance personnel is wrong, the prompt information is fed back to the user end; if t a <α and t c >β, the similarity of the three databases, the similarity in the lower-level database and the corresponding actual maintenance process are fed back to the maintenance personnel to determine whether the recommended maintenance process needs to be modified; if t a <α and t b >β, the similarities of the three databases, the similarities in the middle-level database and the corresponding actual maintenance process are transmitted to the maintenance personnel to determine whether the recommended maintenance process needs to be modified, where α is the first threshold and β is the second threshold.

8. A shield machine maintenance management system according to claim 7, characterized in that: The statistics also include counting the average similarity value y in each database a ,y b ,y c After the maintenance personnel receive the feedback information from the maintenance guidance database, if they think that the feedback information does not match the actual work, the maintenance guidance database will calculate the average similarity value y of each database. a ,y b ,y c Feedback is provided to maintenance personnel and maintenance personnel are prompted to adjust the similarity ranges at different levels.

9. A shield machine maintenance management system according to claim 2, characterized in that: The model update module includes a historical model library, in which the fault inference sub-model and semantic extraction sub-model obtained from historical training are stored. When performing corresponding training, the fault inference sub-model and semantic extraction sub-model versions currently used by the fault identification module are extracted from the historical model library for model optimization. After the optimization is completed, they are deployed to the fault identification module and corresponding backup is performed in the historical model library.

10. A shield machine maintenance management system according to claim 2, characterized in that: The training process of the fault inference sub-model and the semantic extraction sub-model is as follows: The fault inference sub-model is built based on CNN, and the CNN is inferred and trained using a fault-cause sample set. The fault cause sample set includes a variety of main fault phenomena, fault parameters and fault causes involved in each fault area, and a logical relationship between the main fault phenomena, fault parameters and the fault area and fault cause is established, so that the trained CNN can infer the fault area and fault cause based on the main fault phenomena and fault parameters; The semantic extraction sub-model is built based on Transformer, and the Transformer is trained in stages. First, the Transformer is semantically trained using a preprocessed sample set, and the preprocessed sample set includes original fault data and annotated data. The annotated data clarifies the fault area, main fault phenomenon, and fault cause based on the original fault data. The semantically trained Transformer can give accurate corresponding contents of the fault area, main fault phenomenon, fault parameters, and fault cause based on the original fault data. The semantically trained Transformer is then inferred and trained using a standard sample set, where the standard sample set includes annotated content data and array data. The annotated content data includes a brief fault area, main fault phenomenon, fault parameters and fault cause for each fault, and the array data includes the annotated content data of each fault and the corresponding array form data. The inference-trained Transformer can process the original fault data into annotated data, and then convert the annotated data into a unified array form.

Images