Digital twinning-based artificial intelligence driven smart factory fault prediction and maintenance system
By using digital twin technology and artificial intelligence, the factory production process is simulated and sub-chain weights are assigned, which solves the problem of the unconsidered influence of related processes and improves the accuracy and efficiency of prediction and maintenance.
Patent Information
- Application Number
- CN202510763464.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-09
- Publication Date
- 2025-10-31
AI Technical Summary
Existing technologies fail to effectively consider the influence relationships between related processes in predicting process anomalies, resulting in low accuracy and long prediction times, which affects subsequent maintenance work.
By simulating the factory production process using digital twin technology, the system divides the data into independent and related sub-chains, adjusts the weights of the sub-chains according to the association method, and makes predictions based on anomaly feedback and historical data to plan maintenance schemes and optimize the prediction and maintenance sequence.
It improves the accuracy of weight allocation for associated subchains, enhances prediction performance and maintenance efficiency, reduces prediction time, and optimizes maintenance plans.
Smart Images
Figure CN120875827A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of industrial control technology, and more specifically, to an artificial intelligence-driven fault prediction and maintenance system for smart factories based on digital twins. Background Technology
[0002] With the continuous development of artificial intelligence, more and more factories have achieved industrial transformation, that is, to transform manual production into intelligent production, and to replace manual labor with automated equipment to reduce labor costs.
[0003] During the production process, since most of the processes are replaced by mechanical equipment, and for production work with a large number of processes, if an abnormality occurs in a certain process and the quality inspection cannot identify it in time, it will lead to an abnormality in the final product. In order to avoid the recurrence of abnormalities, it is necessary to predict the cause of the abnormality and locate the abnormal process.
[0004] Traditional forecasting methods mainly combine the location of product anomalies with the failure rate of each process. However, in actual production, the types of processes are different. There are related processes (which work together with other processes to complete production) and independent processes (which complete a certain task on their own). If only the failure rate of a single process at a certain anomaly location is considered, the accuracy of the corresponding forecast results will be greatly reduced, especially for production work with complicated processes. At the same time, the forecasting process requires comparison of multiple processes one by one, which will increase the forecasting time and affect subsequent maintenance work.
[0005] To address the aforementioned issues, there is an urgent need for an AI-driven smart factory fault prediction and maintenance system based on digital twins. Summary of the Invention
[0006] The purpose of this invention is to provide an AI-driven fault prediction and maintenance system for smart factories based on digital twins. Through an anomaly cause prediction module, based on feedback results and located anomaly areas, the system predicts the causes of anomalies. It also establishes sub-chain ranges based on the coordination and influence relationships between various processes, and recalculates the weights of each related sub-chain according to the influence relationships, improving the accuracy of weight allocation and subsequent prediction effectiveness. Simultaneously, a maintenance mode simulation module combines the predicted route to simulate maintenance, pre-planning corresponding maintenance schemes. During maintenance, different sub-chain weight range sets are pre-defined, and the impact of a single sub-chain is calculated according to the unit impact value corresponding to the sub-chain weight range set. The maintenance sequence of each sub-chain range is then determined based on the calculation results, thereby solving the problems mentioned in the background art, namely:
[0007] The prediction process did not consider the influence between related processes, which greatly reduced the accuracy of the prediction results.
[0008] To achieve the above objectives, an AI-driven smart factory fault prediction and maintenance system based on digital twins is provided, including a production chain digital twin mapping module, a relationship chain analysis module, a product anomaly feedback module, an anomaly cause prediction module, and a maintenance mode simulation module.
[0009] The overall workflow of factory product production is collected through the production chain digital twin mapping module, and a virtual workflow is simulated through digital twin technology.
[0010] Because there is mutual cooperation between various processes, such as a production step that requires the cooperation of multiple processes, there are differences in the types of processes in production. The relationship chain analysis module divides each process into a corresponding digital twin chain. According to the process content and combination of each process, the corresponding digital twin chain is divided into independent sub-chains and related sub-chains. For related sub-chains, according to their different association methods, they will be divided into three association methods: sequential connection, branch connection, and convergence connection. Different association methods result in different final weight allocation modes.
[0011] Furthermore, when a product malfunctions, the product malfunction feedback module, combined with the product's quality inspection results, reports the malfunctioning product, pinpoints the malfunctioning area, and uses the malfunction cause prediction module to predict the cause based on the feedback results and the located malfunction area. It updates the weights of the corresponding related sub-chains according to their association methods within each sub-chain range. This is combined with the maintenance mode simulation module to perform maintenance simulations based on the predicted routes, allowing for advance planning of corresponding maintenance schemes. First, the impact on the overall sub-chain range is calculated, and the sub-chains are sorted according to their impact magnitude. The maintenance order is determined based on this sorting, with larger impact magnitudes ranking higher. Then, the maintenance order of each related sub-chain within the sub-chain range is ranked according to its weight, with larger weights ranking higher. The overall maintenance route is planned first, followed by the maintenance route planning for each individual related sub-chain.
[0012] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0013] This AI-driven fault prediction and maintenance system for smart factories, based on digital twins, uses an anomaly cause prediction module to predict the causes of anomalies based on feedback results and located anomaly areas. It also establishes sub-chain ranges based on the coordination and influence relationships between various processes, and recalculates the weights of each related sub-chain according to these relationships, improving the accuracy of weight allocation and enhancing subsequent prediction effectiveness. Simultaneously, a maintenance mode simulation module combines predicted routes to simulate maintenance, allowing for advance planning of corresponding maintenance plans. During maintenance, different sub-chain weight range sets are pre-defined, and the impact of a single sub-chain is calculated based on the unit impact value corresponding to each weight range set. The maintenance sequence for each sub-chain range is then determined based on the calculation results, improving maintenance efficiency. Attached Figure Description
[0014] Figure 1 This is a block diagram of the overall system structure of the present invention;
[0015] Figure 2 This is a simulation diagram of the association method of the present invention.
[0016] The meanings of the labels in the diagram are as follows:
[0017] 10. Production chain digital twin mapping module;
[0018] 20. Relationship Chain Analysis Module;
[0019] 30. Product anomaly feedback module;
[0020] 40. Anomaly Cause Prediction Module;
[0021] 50. Maintenance mode simulation module. Detailed Implementation
[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0023] Please see Figure 1 As shown, an AI-driven smart factory fault prediction and maintenance system based on digital twins is provided, including a production chain digital twin mapping module 10, a relationship chain analysis module 20, a product anomaly feedback module 30, an anomaly cause prediction module 40, and a maintenance mode simulation module 50.
[0024] Among them, the production chain digital twin mapping module 10 is used to collect the overall workflow of the factory's product production and simulate a virtual workflow through digital twin technology;
[0025] The relationship chain analysis module 20 combines the components of the virtual workflow to divide each process into corresponding digital twin chains, and further divides the corresponding digital twin chains into independent sub-chains and related sub-chains according to the process content and combination of each process.
[0026] Independent subchains represent individual processes;
[0027] Related sub-chains are processed together with the matching process;
[0028] The product anomaly feedback module 30 combines the quality inspection results from the product side to provide feedback on the products with anomalies and locate the abnormal areas.
[0029] The anomaly cause prediction module 40 predicts the cause of the anomaly based on the feedback results and the located anomaly area. The prediction method is as follows:
[0030] S401. Collect historical abnormal fault data, obtain the abnormal causes corresponding to different abnormal faults, and obtain the occurrence rate of the corresponding abnormal causes.
[0031] S402. Map the occurrence rate of abnormal causes to the corresponding process and obtain the weight of each individual subchain;
[0032] S403. Obtain the range of sub-chains composed of various related sub-chains and establish a historical database for data storage;
[0033] S404. During the prediction process, the range of sub-chains is first located, and then predictions are made in batches based on the abnormal weights of single sub-chains in the historical database to obtain the corresponding prediction routes.
[0034] The maintenance mode simulation module 50 combines the predicted route to perform maintenance simulation and plan the corresponding maintenance scheme in advance.
[0035] In practical use, the specific details are as follows:
[0036] Because the production processes and their sequences differ across production lines, determining a unified production process only after an anomaly occurs would affect subsequent forecasting and maintenance. Therefore, it is necessary to collect the overall workflow of the factory's product production in advance using the production chain digital twin mapping module 10, and then simulate a virtual workflow using digital twin technology, thus virtualizing the entire production process. The specific steps are as follows:
[0037] First, the overall workflow is collected to obtain the production sequence of each process. The process is then marked according to the production sequence to obtain the work content of each process. Corresponding monitoring equipment is configured according to the work content, and real-time monitoring data from the monitoring equipment, such as pressure sensors and temperature sensors, is collected to obtain the real-time status of each process, which serves as the basis for subsequent weighting.
[0038] Furthermore, since there is mutual cooperation between various processes, for example, a certain production step requires the cooperation of multiple processes to complete, the types of processes in production vary. Therefore, the relationship chain analysis module 20, combined with the components of the virtual workflow, divides each process into corresponding digital twin chains. Based on the process content and combination method of each process, the corresponding digital twin chains are further divided into independent sub-chains and related sub-chains. During the division process, the processing mode of each process must first be considered, the processing position of different processes must be obtained, and it must be determined whether the processing positions of each process are the same. If the processing positions of two processes are the same, it indicates that... Two processes are related sub-chains. If the processing positions of the two processes are not the same, it is necessary to determine whether there is a mutual influence relationship between them. If there is an influence relationship, the two are related sub-chains. Conversely, if the processing position of a certain process is not the same as that of other processes and there is no mutual influence relationship, the process will be marked as an independent sub-chain. For example, for drilling, process 1 drills holes with a drill bit, while process 2 needs to blow away impurities from the inside of the drill hole. In this case, process 1 and process 2 are related and both are marked as related sub-chains. For the packaging process, it is an independent process and does not need to cooperate with other processes, so it will be marked as an independent sub-chain.
[0039] For related subchains, based on their association methods, they will be divided into three types: sequential connection, branch connection, and convergence connection. Sequential connection involves multiple processes existing in a sequential positional relationship. Figure 2 As shown, A, B, and C are different processes, and the corresponding order is ABC. For example, to coat a car door, it needs to go through soaking, settling, and drying. The processing positions are the same, all of which are on the outside of the car door. At this time, the three processes are related sub-chains, and the association method is sequential.
[0040] For sequential processes, the correlation occurs when one process affects multiple other processes. Figure 2 As shown, the processing content of process C will affect both process B and process C. For example, the pre-fixing of the two parts will affect the subsequent bolt fixing and welding.
[0041] For consolidation, the correlation method is that multiple processes affect one process, by... Figure 2 As shown, the processing content of process A and process B will affect process C. At the same time, multiple related sub-chains formed by different association methods together constitute the sub-chain range, which serves as the basis for later prediction.
[0042] When a product malfunctions, the product malfunction feedback module 30, combined with the product's quality inspection results, reports the malfunctioning product and locates the malfunction area. The malfunction cause prediction module 40 then predicts the cause of the malfunction based on the feedback results and the located malfunction area. During the prediction process, historical malfunction data needs to be collected in advance to obtain the malfunction causes corresponding to different malfunctions and the corresponding malfunction cause occurrence rates, i.e., the probability of different malfunction causes occurring under the same malfunction, corresponding to different associated sub-chains. The malfunction cause occurrence rates are mapped to the corresponding processes to obtain the weights of each individual sub-chain. The malfunction cause occurrence rate is proportional to the malfunction weight. The range of the sub-chains composed of each associated sub-chain is also obtained, and a historical database is established for data storage. During the prediction process, the range of the sub-chains is first located, and then predictions are performed in batches based on the malfunction weights of individual sub-chains in the historical database to obtain the corresponding prediction routes.
[0043] For different association methods, the allocation of anomaly weights for the corresponding associated sub-chains differs. For sequential association, the weights of each associated sub-chain remain unchanged and are related to the occurrence rate of the anomaly cause. For example, if the occurrence rate of the anomaly cause for a certain associated sub-chain is 35%, then the corresponding anomaly weight is 0.35.
[0044] In the case of a split-chain approach, since one associated subchain affects the other associated subchains, the associated subchain that causes the effect is called the influencing associated subchain. Figure 2 In the connection graph, C represents the affected sub-chain, which is the linked sub-chain that is impacted. Figure 2 In the process of assigning abnormal weights to A and B in the connection diagram, it is first necessary to determine the occurrence rate of abnormal causes of the affected related sub-chains. When the occurrence rate of abnormal causes is zero, it indicates that the affected related sub-chain is not related to the abnormal cause, and the abnormal weight of the corresponding affected related sub-chain remains unchanged.
[0045] When the occurrence rate of anomalies affecting related subchains is greater than zero, it indicates that the affected related subchain is related to anomalies. In this case, it is necessary to determine the occurrence rates of anomalies affecting other affected related subchains. When the occurrence rate of anomalies affecting related subchains is zero, it does not need to be considered. When the occurrence rate of anomalies affecting related subchains is greater than zero, the affected related subchains will share the occurrence rate of anomalies affecting the affected subchain. The corresponding calculation formula is as follows:
[0046] P a =(1+P) c )×P a1 ;
[0047] Where P c P represents the occurrence rate of abnormal causes affecting associated subchains in the delimited chaining method. a1P represents the initial occurrence rate of the cause of anomalies in the affected associated subchains. a The occurrence rate of abnormal causes affecting associated subchains after sharing;
[0048] In the confluence method, multiple influencing related subchains jointly affect a single affected related subchain. Figure 2 The graph shows the connection relationships, where A and B are different influencing sub-chains, and C is the affected sub-chain. In the process of allocating abnormal weights to the sub-chains, it is first necessary to determine the occurrence rate of abnormal causes in the affected sub-chains.
[0049] When the occurrence rate of anomalies in the affected associated subchain is zero, the impact of other factors on the associated subchain is not considered, and the original occurrence rate of anomalies remains unchanged.
[0050] When the occurrence rate of anomalies in the affected associated subchain is greater than zero, the affected associated subchain will share the occurrence rate of anomalies of the other affected associated subchains. The calculation formula is as follows:
[0051] P C1 =(1+P) A1 +P B1 +…+P N1 )×P C2 ;
[0052] Where P A1 -P N1 P represents the occurrence rate of each anomaly affecting the associated subchain in the concatenation method. C2 P represents the initial occurrence rate of the cause of anomalies in the affected associated subchains. C1 This represents the occurrence rate of anomalies in the affected associated subchains after the update.
[0053] Finally, maintenance simulation is performed using the maintenance mode simulation module 50 in conjunction with the predicted route to plan the corresponding maintenance scheme in advance. The specific maintenance method is as follows:
[0054] First, since the individual subchain weights of each associated subchain within different subchain ranges differ and they influence each other, this invention defines three sets of subchain weight ranges to improve later maintenance efficiency: a first-level subchain weight range set, a second-level subchain weight range set, and a third-level subchain weight range set. The first-level subchain weight range set is (0, 45%), the second-level subchain weight range set is (45%, 75%), and the third-level subchain weight range set is (75%, 99%). Each set is defined by a unit influence value: the first-level subchain weight range set has a single unit influence value, the second-level subchain weight range set has three units of influence, and the third-level subchain weight range set has five units of influence. In the specific prediction process, the individual subchain weights of each associated subchain are considered together. Range matching involves calculating the influence of a sub-chain range. The influence of a sub-chain range is the sum of the unit influence of each associated sub-chain. For example, if a sub-chain range includes four associated sub-chains, m1, m2, m3, and m4, where m1 and m3 belong to the second-level sub-chain weight range set, m2 belongs to the third-level sub-chain weight range set, and m4 belongs to the first-level sub-chain weight range set, then the influence of this sub-chain range is 2 × 3 + 6 + 1 = 13. The influence of each sub-chain range is calculated and sorted according to the magnitude of the influence. The maintenance order is determined according to the sorting result, with larger influence values ranking higher in the maintenance order. The maintenance order of each associated sub-chain within a sub-chain range is sorted according to its sub-chain weight, with larger sub-chain weights ranking higher in the maintenance order.
[0055] This invention uses an anomaly cause prediction module 40 to predict the cause of anomalies based on feedback results and located anomaly areas. It also establishes sub-chain ranges based on the coordination and influence relationships between various processes, and recalculates the weights of each associated sub-chain according to these relationships, improving the accuracy of weight allocation and subsequent prediction effectiveness. Simultaneously, it works with a maintenance mode simulation module 50 to simulate maintenance along the predicted route, pre-planning corresponding maintenance schemes. During maintenance, different sub-chain weight range sets are pre-defined, and the influence of a single sub-chain is calculated based on the unit influence corresponding to each sub-chain weight range set. The maintenance sequence for each sub-chain range is then determined based on the calculation results, improving maintenance efficiency.
[0056] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. A digital twin-based AI-driven smart factory fault prediction and maintenance system, characterized by: It includes a production chain digital twin mapping module (10), a relationship chain analysis module (20), a product anomaly feedback module (30), an anomaly cause prediction module (40), and a maintenance mode simulation module (50); The production chain digital twin mapping module (10) is used to collect the overall workflow of the factory's production of products and simulate a virtual workflow through digital twin technology. The relationship chain analysis module (20) combines the components of the virtual workflow to divide each process into corresponding digital twin chains, and divides the corresponding digital twin chains into independent sub-chains and related sub-chains according to the process content and matching method of each process. Independent subchains represent individual processes; Related sub-chains are processed together with the matching process; The product anomaly feedback module (30) combines the quality inspection results from the product end to provide feedback on the products that have anomalies and locate the abnormal areas; The anomaly cause prediction module (40) predicts the anomaly cause based on the feedback results and the located anomaly area. The prediction method is as follows: S401. Collect historical abnormal fault data, obtain the abnormal causes corresponding to different abnormal faults, and obtain the occurrence rate of the corresponding abnormal causes. S402. Map the occurrence rate of abnormal causes to the corresponding process and obtain the weight of each individual subchain; S403. Obtain the range of sub-chains composed of various related sub-chains and establish a historical database for data storage; S404. During the prediction process, the range of sub-chains is first located, and then predictions are made in batches based on the abnormal weights of single sub-chains in the historical database to obtain the corresponding prediction routes. The maintenance mode simulation module (50) combines the predicted route to perform maintenance simulation and plan the corresponding maintenance scheme in advance.
2. The AI-driven fault prediction and maintenance system for smart factories based on digital twins as described in claim 1, characterized in that: The method for simulating a virtual workflow using digital twin technology in the production chain digital twin mapping module (10) includes the following steps: S101. Collect the overall workflow and obtain the production sequence of each process; S102. Mark the process according to the production sequence to obtain the work content of each process. S103. Configure the corresponding monitoring equipment according to the work content, and collect the real-time monitoring data of the monitoring equipment to obtain the real-time status of each process.
3. The AI-driven fault prediction and maintenance system for smart factories based on digital twins as described in claim 1, characterized in that: The method for dividing each process in the relationship chain analysis module (20) includes the following steps: S201. Obtain the processing positions of different processes; S202. Determine whether the processing positions of each process are the same; If two processes are located at the same position, it indicates that the two processes are related sub-chains. If the processing positions of the two processes are not the same, it is necessary to determine whether there is a mutual influence relationship between them. If an influence relationship exists, the two are related subchains; Conversely, if a process is located at a different position than other processes and there is no mutual influence between them, then the process will be marked as an independent sub-chain.
4. The AI-driven fault prediction and maintenance system for smart factories based on digital twins as described in claim 1, characterized in that: The association methods of the associated sub-chains in S202 include sequential association, branch association, and convergence association.
5. The AI-driven fault prediction and maintenance system for smart factories based on digital twins as described in claim 1, characterized in that: The method for obtaining the weights of each individual subchain in S402 includes the following steps: S4021. Distribute the weight of a single subchain according to the association method; S4022. For the range of sub-chains in the sequential connection method, the initial weights of each associated sub-chain do not change, and the corresponding initial weights are the occurrence rates of abnormal causes. S4023. For the sub-chain range of the split-chain method, determine the occurrence rate of abnormal causes affecting the associated sub-chains; When the occurrence rate of its abnormal cause is zero, the abnormal weight of the corresponding affected related subchain remains unchanged. When the occurrence rate of abnormal causes affecting related sub-chains is greater than zero, the occurrence rate of abnormal causes for other affected related sub-chains is determined. If the occurrence rate of anomalies in the affected related subchains is zero, then no consideration is needed; When the occurrence rate of anomalies in the affected associated subchain is greater than zero, the affected associated subchains will share the occurrence rate of anomalies affecting the associated subchains. S4024. For the range of sub-chains in the concatenation method, determine the occurrence rate of abnormal causes for the affected associated sub-chains. When the occurrence rate of anomalies in the affected associated subchain is zero, the impact of other factors on the associated subchain is not considered, and the original occurrence rate of anomalies remains unchanged. When the occurrence rate of anomalies in the affected associated subchain is greater than zero, the affected associated subchain will share the occurrence rate of anomalies in the other affected associated subchains.
6. The AI-driven fault prediction and maintenance system for smart factories based on digital twins as described in claim 5, characterized in that: The algorithm for the occurrence rate of anomalies affecting associated subchains in S4023 is as follows: P a =(1+P c )×P a1 ; P c P represents the occurrence rate of abnormal causes affecting associated subchains in the delimited chaining method. a1 P represents the initial occurrence rate of the cause of anomalies in the affected associated subchains. a The occurrence rate of abnormal causes affecting associated subchains after sharing.
7. The AI-driven fault prediction and maintenance system for smart factories based on digital twins as described in claim 5, characterized in that: The algorithm for sharing the occurrence rate of other anomalies affecting associated subchains in S4024 is as follows: P C1 =(1+P A1 +P B1 +…+P N1 )×P C2 ; Where P A1 -P N1 P represents the occurrence rate of each anomaly affecting the associated subchain in the concatenation method. C2 P represents the initial occurrence rate of the cause of anomalies in the affected associated subchains. C1 This represents the occurrence rate of anomalies in the affected associated subchains after the update.
8. The AI-driven fault prediction and maintenance system for smart factories based on digital twins according to claim 1, characterized in that: The method for pre-planning the corresponding maintenance scheme in the maintenance mode simulation module (50) includes the following steps: S501. Define three sets of sub-chain weight ranges: the first-level sub-chain weight range set, the second-level sub-chain weight range set, and the third-level sub-chain weight range set. S502. Define each set by unit influence. Among them, the set of weight ranges of the first-level subchains represents the influence of a single unit; The set of weight ranges for the second-level subchains consists of three units of influence. The set of weight ranges for the third-level subchain consists of five units of influence. S503. Combine the single subchain weight range matching of each related subchain to calculate the influence of the subchain range; S504. The influence of a subchain range is the sum of the unit influence of each of its associated subchains; S505. Sort according to the magnitude of the impact, and determine the maintenance order according to the sorting result. The greater the impact, the higher the maintenance order. For each related sub-chain within the sub-chain range, the maintenance order is sorted according to the weight of the sub-chain. The greater the weight of the sub-chain, the higher the maintenance order of the corresponding related sub-chain.
9. The AI-driven fault prediction and maintenance system for smart factories based on digital twins as described in claim 8, characterized in that: The set of weight ranges for the first-level subchains in S501 is (0, 45%). The set of weight ranges for the second-level subchains is (45%, 75%). The set of weight ranges for the third-level subchain is (75%, 99%).