Intelligent maintenance decision method and system based on knowledge graph

Abstract

The application relates to an intelligent maintenance decision method and system based on a knowledge graph, relates to the fields of intelligent operation and maintenance and decision support and information technology, and comprises the following steps: acquiring industrial equipment operation state data and historical maintenance data; then, constructing an industrial knowledge graph and generating a correlation relationship model; based on the analysis of the relationship between fault types and maintenance paths, an optimal maintenance path is obtained; through the optimal maintenance path, the correlation relationship model and maintenance resource constraint conditions, a maintenance task allocation scheme is optimized; then, optimal maintenance strategies are obtained through dynamic adjustment analysis of maintenance time windows; finally, multi-objective optimization iteration analysis is performed to obtain a final maintenance decision; the scheme can comprehensively consider various factors, and the scientificity and accuracy of the maintenance decision of the industrial equipment are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent operation and maintenance and decision support, in particular to an intelligent maintenance decision method and system based on a knowledge graph. BACKGROUND

[0002] At present, in the field of industrial equipment maintenance, efficient handling of complex faults is a key link to ensure production continuity. With the development of industrial systems towards high integration and intelligence, the operating environment of equipment is becoming more and more complex, and the traditional maintenance mode relying solely on human experience has been difficult to meet the demand for rapid response. However, factors such as scattered storage of maintenance data, difficulty in expert experience inheritance, and low utilization efficiency of equipment manuals make the traditional maintenance method insufficient in dealing with complex scenarios such as multi-source heterogeneous information, resulting in prolonged fault handling time and increased downtime loss.

[0003] In one prior art, equipment maintenance decision mainly relies on manual consultation of equipment manuals combined with historical maintenance records, and maintenance plans are formulated by manually searching for related cases and referring to expert suggestions, lacking deep correlation analysis of multi-source data. At the same time, existing methods mostly use static knowledge bases based on rules for fault matching, without fully considering the impact of dynamic changes in equipment state on maintenance strategies, resulting in insufficient adaptability and deviation in new types of faults or complex scenarios. Therefore, the existing technology of equipment maintenance decision has the problem of low accuracy. SUMMARY

[0004] The main purpose of the present application is to provide an intelligent maintenance decision method and system based on a knowledge graph, which can improve the accuracy of maintenance decision of industrial equipment.

[0005] To achieve the above-mentioned purpose, the embodiments of the present application provide an intelligent maintenance decision method based on a knowledge graph, which comprises:

[0006] obtaining equipment operating state data and historical maintenance data of industrial equipment;

[0007] constructing an industrial knowledge graph according to the equipment operating state data and the historical maintenance data, and generating a correlation relationship model of equipment operating state and maintenance behavior according to the industrial knowledge graph;

[0008] analyzing the relationship between fault types and maintenance paths according to the correlation relationship model and the equipment operating state data to obtain an optimal maintenance path;

[0009] performing maintenance task allocation optimization according to the optimal maintenance path, the correlation relationship model and a preset maintenance resource constraint condition to obtain an optimal maintenance task allocation scheme;

[0010] According to the optimal maintenance path, the optimal maintenance task allocation scheme and the correlation model, dynamic adjustment analysis of a maintenance time window is performed to obtain an optimal maintenance strategy.

[0011] According to the optimal maintenance strategy, multi-objective optimization iterative analysis is performed to obtain a final maintenance decision of the industrial equipment.

[0012] Correspondingly, the embodiment of the application further provides an intelligent maintenance decision system based on a knowledge graph, comprising:

[0013] An acquisition module is configured to acquire equipment operation state data and historical maintenance data of an industrial equipment.

[0014] A graph construction module is configured to construct an industrial knowledge graph according to the equipment operation state data and the historical maintenance data, and generate a correlation model of equipment operation state and maintenance behavior according to the industrial knowledge graph.

[0015] A relationship analysis module is configured to analyze a relationship between a fault type and a maintenance path according to the correlation model and the equipment operation state data, and obtain an optimal maintenance path.

[0016] An optimization module is configured to perform maintenance task allocation optimization according to the optimal maintenance path, the correlation model and a preset maintenance resource constraint condition, and obtain an optimal maintenance task allocation scheme.

[0017] A window adjustment module is configured to perform dynamic adjustment analysis of a maintenance time window according to the optimal maintenance path, the optimal maintenance task allocation scheme and the correlation model, and obtain an optimal maintenance strategy.

[0018] A decision module is configured to perform multi-objective optimization iterative analysis according to the optimal maintenance strategy, and obtain a final maintenance decision of the industrial equipment.

[0019] To sum up, by using the technical solution of the application, firstly, the industrial equipment running state data and historical maintenance data are acquired; then, the industrial knowledge graph is constructed and the correlation relationship model is generated, the internal relationship between the equipment running and the maintenance behavior can be deeply mined; based on this, the optimal maintenance path is obtained by analyzing the relationship between the fault type and the maintenance path, which can effectively improve the maintenance efficiency and reduce the fault troubleshooting time; the maintenance task allocation scheme is optimized by the optimal maintenance path, the correlation relationship model and the maintenance resource constraint condition, which can reasonably allocate the maintenance resources and avoid resource waste; the optimal maintenance strategy is obtained by the dynamic adjustment analysis of the maintenance time window, which makes the maintenance plan more flexible and adaptive; finally, the final maintenance decision is obtained by the multi-objective optimization iterative analysis; the scheme can comprehensively consider various factors, improves the scientificity and accuracy of the maintenance decision of the industrial equipment, especially in the face of complex fault scenarios, further improves the overall benefit of the industrial equipment maintenance, guarantees the stable and efficient operation of the industrial equipment, reduces the influence of equipment failure on production and reduces the maintenance cost. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0021] Figure 1 is a scene schematic diagram of the intelligent maintenance decision method based on the knowledge graph in the embodiments of the application;

[0022] Figure 2 is a flowchart of the intelligent maintenance decision method based on the knowledge graph provided in the embodiments of the application;

[0023] Figure 3 is a flowchart of the knowledge graph construction provided in the embodiments of the application;

[0024] Figure 4 is a flowchart of the optimal maintenance path determination provided in the embodiments of the application;

[0025] Figure 5 is a flowchart of the path optimization provided in the embodiments of the application;

[0026] Figure 6 is another flowchart of the path optimization provided in the embodiments of the application;

[0027] Figure 7 is a flowchart of the multi-objective optimization iterative analysis provided in the embodiments of the application;

[0028] Figure 8A structural schematic diagram of an intelligent maintenance decision system based on a knowledge graph is provided for an embodiment of the present application.

[0029] Figure 9 A structural schematic diagram of a computer device is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person skilled in the art without creative work fall within the scope of protection of the present application.

[0031] The present application provides an intelligent maintenance decision method and system based on a knowledge graph, which will be described in detail below.

[0032] In the embodiments of the present application, the intelligent maintenance decision method based on a knowledge graph is an innovative method specially for industrial equipment maintenance decision. It first collects equipment running state data and historical maintenance data of industrial equipment, which is the basic data source of the entire method. Then it uses these data to build an industrial knowledge graph, thereby generating a correlation model of equipment running state and maintenance behavior, which can mine deep connections between equipment running and maintenance. On this basis, the optimal maintenance path is determined by analyzing the relationship between fault types and maintenance paths, improving the pertinence and efficiency of maintenance. Then the optimal maintenance path, the correlation model and the preset maintenance resource constraint condition are combined to optimize the maintenance task allocation scheme, realizing the rational use of resources. Then the optimal maintenance strategy is obtained by dynamic adjustment analysis of the maintenance time window based on the above results, enhancing the ability to respond to actual production changes. Finally, the final maintenance decision is obtained by multi-objective optimization iteration analysis, improving the scientificity and rationality of the industrial equipment maintenance decision as a whole, ensuring stable operation of the equipment and reducing maintenance cost, etc.

[0033] As shown in FIG. 1, a knowledge graph-based intelligent maintenance decision method scenario is provided. In the knowledge graph-based intelligent maintenance decision scenario, it mainly includes data acquisition equipment, knowledge graph construction equipment, path optimization equipment, task allocation equipment, intelligent decision equipment, etc., and the data interaction between each equipment is realized through the network.

[0034] The data acquisition equipment is used to acquire equipment running state data and historical maintenance data, which are derived from equipment real-time sensor data, equipment running log data, and maintenance operation steps, maintenance consumable usage and maintenance man-hour data in historical maintenance records.

[0035] The data acquisition device converts the above data into a unified format through standardization processing for subsequent module use. The standardization processing includes data cleaning and normalization to ensure data consistency and availability. For example, for sensor data of different devices, the maximum and minimum value normalization method is used to map the numerical range to 0 to 1, and the missing values are interpolated to ensure data integrity.

[0036] The knowledge graph construction device is used to process the standardized data based on a domain knowledge extraction algorithm to generate an industrial knowledge graph. The knowledge graph construction device first extracts initial knowledge graph nodes and edges from the equipment operation state data and historical maintenance data through entity recognition and relationship extraction methods to form a preliminary knowledge graph structure. Subsequently, the correlation strength distribution of the initial knowledge graph is obtained by calculating the relationship weight between nodes, and the knowledge graph is mapped to a low-dimensional space using a graph embedding algorithm to enhance the expression ability of the graph. Finally, a hierarchical clustering method is used to further optimize node grouping and edge weight to ultimately form a correlation model of equipment operation state and maintenance behavior. This model can clearly reflect the internal relationship between equipment operation state and maintenance behavior, providing a basis for subsequent analysis.

[0037] The path optimization device is used to analyze the relationship between fault types and maintenance paths and optimize them based on the correlation model generated by the knowledge graph construction device and the real-time equipment operation state data. The path optimization device first analyzes the dynamic response relationship between fault type occurrence probability and maintenance path execution order to obtain the influence weight of fault type on maintenance path. Then, based on the influence weight and the correlation model, a historical matching degree matrix of fault type and maintenance path is calculated to determine the path selection constraint condition. On this basis, a simulated annealing algorithm is used to comprehensively evaluate the candidate maintenance path set, calculate the execution efficiency index of each path, and sort the paths. If the highest ranked path meets the preset efficiency threshold, it is determined as the optimal maintenance path; otherwise, the candidate path set is locally searched and optimized until the optimal solution is found.

[0038] The task allocation device is used to allocate maintenance tasks based on the optimal maintenance path, the association relationship model, and the pre-set maintenance resource constraints. First, the nonlinear relationship between maintenance task allocation and maintenance resource utilization is analyzed, and a nonlinear relationship model of maintenance resources is constructed. Then, based on this model and the optimal maintenance path, the maintenance task allocation optimization function is decomposed, and the particle swarm optimization algorithm is used for iterative calculation to obtain a set of maintenance task allocation solutions that meet the maintenance resource constraints. Through the balance analysis of each solution in the solution set, the solution with the best balance is finally selected as the optimal maintenance task allocation scheme. For example, in a certain task allocation, the task allocation device will give priority to the matching degree of maintenance personnel skill level and task complexity, while taking into account the inventory situation of maintenance consumables, to ensure that the allocation result is both efficient and reasonable.

[0039] The plan optimization device is responsible for dynamic adjustment analysis of the maintenance time window to optimize the maintenance plan. First, the dynamic coupling relationship between the maintenance time window and the equipment running state is analyzed, and the equipment running state change characteristics are extracted. Then, the time series prediction algorithm is used to analyze the maintenance time window fluctuation to determine the adjustment interval. On this basis, the priority sorting function of the maintenance time window adjustment strategy is constructed, and the dynamic programming algorithm is used to analyze the global optimization result. Finally, the plan optimization device optimizes the maintenance plan by combining the optimal maintenance path, the optimal maintenance task allocation scheme, and the maintenance time window adjustment strategy, and generates the optimal maintenance plan and the optimal maintenance strategy. For example, in a certain scenario, the plan optimization device will flexibly adjust the maintenance time window according to the change of equipment running load to avoid production interruption caused by maintenance.

[0040] The intelligent decision-making device is used to generate the optimal maintenance strategy based on the optimal maintenance strategy generated by the plan optimization device, and to perform iterative analysis using a multi-objective optimization algorithm to output the final maintenance decision. The intelligent decision-making device first constructs a maintenance decision optimization space containing the maintenance path, the maintenance task allocation, and the maintenance time window, and combines the pre-set maintenance cost threshold and the maintenance efficiency threshold for collaborative constraint to generate the first maintenance decision combination. Subsequently, the multi-objective optimization algorithm is initialized and iteratively calculated, and after each iteration, it is determined whether the second maintenance decision combination is better than the first maintenance decision combination. If it is better, update the first maintenance decision combination, otherwise discard the current combination. When the number of iterations reaches the pre-set value, the current first maintenance decision combination is output as the final maintenance decision. For example, during a certain iteration process, the intelligent decision-making device finds that a certain maintenance decision combination significantly improves the maintenance efficiency while reducing the maintenance cost, so it is selected as the final decision. The above devices work closely together to realize the overall function of the intelligent maintenance decision-making system based on the knowledge graph.

[0041] Reference Figure 2 , Figure 2FIG. 1 is a flowchart of a method for intelligent maintenance decision-making based on a knowledge graph according to an embodiment of the present application. The method can be executed by a computer device, which can be a single computer device or a cluster of multiple computer devices. The computer device can be a terminal device or a server. The method includes the following steps.

[0042] Step S10: Obtain device running state data and historical maintenance data of the industrial equipment.

[0043] The device running state data refers to a set of information reflecting the running state of the industrial equipment during operation. For example, for a large industrial compressor, the device running state data includes, but is not limited to, the inlet pressure, outlet pressure, running temperature, rotation speed, vibration frequency, and amplitude of the compressor. The device running state data can be obtained by various sensors installed at key parts of the equipment, such as pressure sensors, temperature sensors, rotation speed sensors, and acceleration sensors.

[0044] The historical maintenance data is a detailed record of the maintenance activities experienced by the industrial equipment in the past, including maintenance date, maintenance reason (such as fault phenomenon, fault diagnosis result), maintenance location, maintenance measures (such as replaced parts, maintenance process), and maintenance personnel information. These data are usually stored in the enterprise equipment management information system (EMIS) database and can be obtained from the EMIS.

[0045] In an embodiment, the Internet of Things technology can be used to obtain the state data and historical data. An Internet of Things module is installed on the industrial equipment, and the device running state data is transmitted in real time to the cloud platform through a wireless communication protocol. At the same time, a data mining tool is deployed on the enterprise's equipment maintenance database to extract historical maintenance data from the database. This helps to improve the efficiency of data collection and facilitates centralized management of data from multiple devices.

[0046] Step S20: Construct an industrial knowledge graph based on the device running state data and the historical maintenance data, and generate a correlation model of device running state and maintenance behavior based on the industrial knowledge graph.

[0047] The industrial knowledge graph is a graph-based data structure used to represent entities, concepts, and their relationships in the field of industrial equipment. In this application, entities can be equipment components (such as motors, bearings, valves, etc.), equipment operating states (such as overheating, abnormal vibration, etc.), maintenance actions (such as replacing parts, adjusting parameters, etc.), and maintenance resources (such as maintenance tools, maintenance personnel skills, etc.). Relationships describe the connections between these entities, for example, "motor overheating" may have a causal relationship with the maintenance action of "replacing heat sinks".

[0048] In the embodiments of the present application, when constructing the industrial knowledge graph, first, the equipment operating state data and historical maintenance data are pre-processed, including data cleaning (removing noise data, duplicate data, etc.), data labeling (adding semantic labels to the data), and other operations. Then, the entities and relationships in the data are identified through knowledge extraction techniques, and an initial knowledge graph structure is constructed.

[0049] The association relationship model is a representation form that quantifies and models the relationships between entities based on the industrial knowledge graph. For example, a model reflecting the quantitative relationship between equipment operating states and maintenance actions can be constructed by calculating indicators such as the frequency of occurrence and the strength of association between entities.

[0050] In an embodiment, a graph neural network (GNN) technique is used to construct the industrial knowledge graph and generate the association relationship model. The pre-processed equipment operating state data and historical maintenance data are converted into a graph structure data representation, where nodes represent entities and edges represent relationships between entities. Then, this graph structure data is input into the graph neural network. The graph neural network automatically learns the relationship features between entities by performing convolution operations on the graph structure. For the generation of the association relationship model, during the training process of the graph neural network, a suitable loss function is designed to enable the network to learn the quantitative representation of the relationships between entities. For example, the loss function can be designed based on the prediction accuracy of the relationships between entities. This implementation based on the graph neural network has strong learning ability and can automatically mine complex relationships in the data, making it more suitable for handling large-scale and complex industrial equipment data compared to traditional rule-based and handcrafted feature-based methods.

[0051] In an embodiment, to improve the accuracy of the industrial knowledge graph and maintenance decision-making, as shown in FIG. 2, the following methods can be used to construct the knowledge graph: Figure 3

[0052] Step S201: According to the equipment operating state data and the historical maintenance data, the nodes and edges of the initial knowledge graph are constructed, and a preliminary knowledge graph structure is obtained.

[0053] ​In this step, the construction of nodes and edges is the basis for building the industrial knowledge graph. Among them, the nodes represent various entities related to industrial equipment maintenance. For example, in the maintenance scenario of industrial robots, the nodes can be the joints of the robot (such as shoulder joint, elbow joint, etc.), key components (such as motor, reducer, etc.), possible faults (such as motor overheating, joint jamming, etc.), and maintenance operations (such as replacing the motor, lubricating the joint, etc.). The edges represent the relationships between these entities. For example, there is a causal relationship between the fault node "motor overheating" and the maintenance operation node "replace the motor", which can be represented by an edge.

[0054] In the embodiments of the present application, the nodes and edges of the initial knowledge graph need to be constructed by deeply mining the equipment running state data and historical maintenance data. First, the data needs to be analyzed semantically to identify meaningful entities and relationship descriptions. For example, from the maintenance record description "due to bearing wear, the equipment vibration is abnormal, and bearing replacement maintenance is performed" in the historical maintenance data, the entities "bearing wear", "equipment vibration abnormality", and "bearing replacement" can be identified, as well as the relationships "bearing wear - causes - equipment vibration abnormality" and "equipment vibration abnormality - requires - bearing replacement", thereby constructing the corresponding nodes and edges.

[0055] In an embodiment, semantic template matching technology can be used. Specifically, a series of semantic templates related to industrial equipment maintenance are defined in advance, such as "[component name] - fault cause - [fault phenomenon]", "[fault phenomenon] - maintenance operation - [maintenance measure]", etc. Then the text information in the equipment running state data and historical maintenance data is matched with these semantic templates, and the entities and relationships are extracted according to the matching results, and the nodes and edges of the initial knowledge graph are constructed. The advantage of this method is simple and direct, and it has good effect on relatively fixed structure data.

[0056] Step S202: According to the preliminary knowledge graph structure, the relationship weight between nodes is calculated to obtain the correlation strength distribution of the initial knowledge graph.

[0057] Among them, the correlation strength distribution is used to reflect the closeness of the relationships between the nodes in the knowledge graph. For example, in the knowledge graph of a certain chemical production equipment, the correlation between "reaction kettle temperature too high" and "cooling system failure" may be very close, while the correlation between "stirring paddle damage" is relatively weak. The calculation of this correlation strength is based on multiple factors, such as the frequency of the simultaneous occurrence of the two in historical data, the certainty of the causal relationship, etc.

[0058] In an embodiment, the inter-node relationship weight needs to consider various information in the data. Based on statistical methods, for example, the proportion of the number of times that "reactor temperature is too high" and "cooling system failure" appear at the same time in all maintenance records to the total number of times that "reactor temperature is too high" appears can be calculated as a weight index of the relationship between the two. At the same time, domain expert knowledge can also be introduced, and for some node pairs that are judged to be closely related according to experience, appropriate weight addition can be given.

[0059] In an embodiment, Bayesian network technology can also be used to calculate the inter-node relationship weight. The preliminary knowledge graph structure is converted into a Bayesian network structure, and the nodes are used as variables, and the edges represent the conditional dependence relationship between variables. According to the sample data in the device operating state data and historical maintenance data, the conditional probability distribution in the Bayesian network is estimated, and this conditional probability distribution can be used as the weight of the relationship between nodes. The advantage of this method is that it can well handle uncertain relationships and can easily combine prior knowledge.

[0060] Step S203: enhancing the preliminary knowledge graph structure according to the correlation strength distribution to obtain an industrial knowledge graph.

[0061] Among them, the enhancement processing is the optimization of the preliminary knowledge graph structure, so that the knowledge graph can more accurately reflect the entity relationship in the industrial equipment maintenance. According to the correlation strength distribution, for the node relationship with high correlation strength, it can be represented in a more prominent way in the knowledge graph, such as bold display of the edge, increase of the connection tightness of the node, etc. (here is only a conceptual representation, and there will be corresponding processing method in the actual data structure).

[0062] In an embodiment, the data structure of the knowledge graph can also be adjusted. For example, if a node has a strong correlation relationship with multiple other nodes, clustering analysis can be performed on the node to group it with the strongly correlated nodes for better reflecting the relationship structure between them.

[0063] In an embodiment, a graph convolutional neural network (GCN) can be used to enhance the preliminary knowledge graph structure. The knowledge graph is input into the GCN as graph structure data, and the GCN learns the feature representation of the nodes according to the correlation strength distribution between the nodes. In the training process, the GCN will automatically adjust the representation of the nodes, so that the nodes with high correlation strength are closer in the feature space, thereby enhancing the knowledge graph structure. This method can take advantage of the powerful graph data processing ability of the GCN, and can effectively improve the representation ability of the knowledge graph for the relationship in the industrial equipment maintenance.

[0064] In an embodiment, the enhanced knowledge graph representation can be obtained by mapping the association strength distribution and the preliminary knowledge graph structure to a low-dimensional space using a graph embedding algorithm.

[0065] In the embodiments of the present application, the graph embedding algorithm aims to convert the high-dimensional graph structure data (here, the preliminary knowledge graph structure and its association strength distribution) into vector representation in a low-dimensional space. The advantage of doing so is that it can preserve the graph structure information while facilitating subsequent calculation and analysis operations.

[0066] For the knowledge graph of industrial equipment maintenance, the preliminary knowledge graph structure contains numerous nodes and complex edge relationships, which are in a high-dimensional space and are relatively complex to process. For example, in a knowledge graph of a large industrial production system containing various equipment components, fault types, maintenance operations, etc., each entity has its own attributes and multiple relationships with other entities, which collectively form a high-dimensional data structure. Through the graph embedding algorithm, these complex structures can be mapped to a low-dimensional space.

[0067] The graph embedding algorithm adjusts the mapping results according to the association strength distribution during the mapping process. The vector representation of nodes with high association strength in the low-dimensional space will be closer, thus better reflecting the internal relationships in the knowledge graph. For example, if the association strength between “equipment component failure” and “specific maintenance operation” is high, the corresponding node vectors in the low-dimensional space will be closer in distance.

[0068] In an embodiment, the Node2Vec graph embedding algorithm can be used. Node2Vec is a graph embedding algorithm based on random walk. First, random walk is performed on the preliminary knowledge graph structure, and the probability of random walk is determined according to the association strength distribution. For example, edges with high association strength have a higher probability of being selected for random walk. Then, by learning a large number of random walk paths, each node is mapped to a vector representation in a low-dimensional space. The advantage of this algorithm is that it can effectively capture local and global relationships in the graph structure, and can perform biased random walk according to the different association strengths, thus better reflecting the relationship structure in the knowledge graph.

[0069] In an embodiment, the step S20 of generating an association relationship model between equipment operation state and maintenance behavior according to the industrial knowledge graph can include: using a hierarchical clustering method to optimize node grouping and edge weight according to the enhanced knowledge graph representation, to obtain the association relationship model between equipment operation state and maintenance behavior.

[0070] Hierarchical clustering is a clustering method that forms a tree structure by analyzing data at different levels based on the similarity between clusters. Based on the enhanced knowledge graph representation, hierarchical clustering can be used to group nodes according to their similarity in vector representation (obtained by graph embedding algorithm). For example, in the knowledge graph of industrial equipment, nodes related to the running state of the equipment (such as different running parameter nodes of the equipment) may be grouped into a group according to their similarity, and nodes related to maintenance behavior (such as different maintenance operation nodes) may also be grouped into a group. Such grouping helps to more clearly present the relationship between the running state of the equipment and the maintenance behavior.

[0071] At the same time, the hierarchical clustering method can also optimize the edge weight. In the clustering process, the edge weight between nodes in the same group may be adjusted according to the clustering result. If two nodes are determined to be in a closer relationship after clustering, the edge weight between them may be increased; on the contrary, if the relationship between nodes is considered relatively weak after clustering analysis, the edge weight may be reduced. In this way, the structure of the knowledge graph is further optimized, so that the final obtained association relationship model between the running state of the equipment and the maintenance behavior more accurately reflects the actual situation.

[0072] In an embodiment, specifically, each node can be regarded as a separate class. Then, the distance between each pair of classes (nodes) is calculated, which can be based on the Euclidean distance or cosine similarity of node vector representation and other measurement methods. According to the calculated distance, the two classes with the closest distance are merged into a new class. Repeat this process until a predetermined stopping condition is reached, such as reaching a specified number of clusters or the distance between classes being greater than a certain threshold. In the process of merging classes each time, the edge weight is adjusted according to the new clustering structure. For example, if the relationship between nodes in a class becomes closer after the two classes are merged, the corresponding edge weight is increased. This way can naturally form a hierarchical structure, which is suitable for optimizing complex knowledge graph structures.

[0073] Step S30: According to the association relationship model and the equipment running state data, analyze the relationship between fault types and maintenance paths to obtain the optimal maintenance path.

[0074] Among them, the fault type refers to the result of classifying industrial equipment according to factors such as fault phenomenon and fault cause. For example, the fault type of equipment on an automated production line can be divided into mechanical failure (such as transmission component wear, connection loose, etc.), electrical failure (such as circuit short circuit, motor burnout, etc.), and control system failure (such as program error, sensor failure, etc.).

[0075] The maintenance path refers to an ordered combination of a series of maintenance operations taken for a specific fault type. In analyzing the relationship between the fault type and the maintenance path, the relationship can be analyzed based on the association relationship model and the equipment operating state data. Specifically, in an embodiment, first, knowledge in the association relationship model is used to infer possible fault types according to the current equipment operating state data. For example, if the equipment operating state data shows that the current of a certain electrical equipment suddenly increases and the voltage fluctuates abnormally, in combination with the characteristic relationship about electrical faults in the association relationship model, it can be judged that the possible fault type is a circuit short circuit. Then, the maintenance operations related to the fault type and the order relationship thereof are found in the association relationship model to form a candidate set of maintenance paths. Finally, the optimal maintenance path is determined by evaluating factors such as the cost, maintenance time, and maintenance success rate of each maintenance path.

[0076] In an embodiment, reinforcement learning techniques can be used to analyze the relationship between the fault type and the maintenance path and obtain the optimal maintenance path. The fault diagnosis and maintenance path selection process is constructed as a reinforcement learning environment, in which the state represents the equipment operating state data and the inferred fault type, the action represents the selected maintenance operation, and the reward function is designed based on factors such as maintenance cost, maintenance time, and maintenance success rate. For example, a higher reward is given when the maintenance operation successfully repairs the fault and the cost is low and the time is short. The agent learns the optimal maintenance path strategy for different fault types in this environment by constantly interacting with the environment (selecting maintenance operations and observing the results). This implementation of reinforcement learning can continuously optimize the maintenance path selection strategy according to actual maintenance feedback, does not need to pre-set complex rules, and can adapt to dynamic changes in equipment operating state and fault type.

[0077] Step S40: performing maintenance task allocation optimization according to the optimal maintenance path, the association relationship model, and a preset maintenance resource constraint condition to obtain an optimal maintenance task allocation scheme.

[0078] The maintenance resource constraint condition refers to various resource limitation factors faced in equipment maintenance, including but not limited to the number of maintenance personnel, skill level, types and number of maintenance tools, inventory situation of maintenance parts, etc. In performing maintenance task allocation optimization, the demand of each maintenance operation in the optimal maintenance path for maintenance resources and the order and dependency relationship between maintenance operations in the association relationship model are comprehensively considered. For example, certain maintenance operations require maintenance personnel with specific skills and special maintenance tools, and must be performed after other maintenance operations are completed.

[0079] In an embodiment, an integer programming algorithm can be employed for optimization. The maintenance tasks are regarded as variables, the maintenance resource constraints as constraint equations, and the relationships between the maintenance tasks (based on the association relationship model) as part of the objective function. For example, if maintenance task A must be completed before maintenance task B, this order relationship is reflected in the objective function. By solving with the integer programming algorithm, a maintenance task allocation scheme is obtained that satisfies the maintenance resource constraints and optimizes the objective function (such as the shortest maintenance time or the lowest maintenance cost). This approach can obtain an optimal maintenance task allocation scheme through mathematical optimization methods while satisfying various constraints, improving the utilization efficiency of maintenance resources.

[0080] Step S50: Based on the optimal maintenance path, the optimal maintenance task allocation scheme, and the association relationship model, dynamic adjustment analysis of the maintenance time window is performed to obtain an optimal maintenance strategy.

[0081] The maintenance time window refers to the time period suitable for maintenance operations during the operation of the industrial equipment. The determination of this time period needs to consider various factors such as the equipment operation plan, production task arrangement, and equipment operation state. For example, for a continuous production chemical equipment, the maintenance time window may only be selected during the regular maintenance shutdown period or the short gap of production task switching when the production task is tight; while when the production task is relatively loose, maintenance operations can be performed during the low load stage of equipment operation. When performing dynamic adjustment analysis of the maintenance time window based on the optimal maintenance path, the optimal maintenance task allocation scheme, and the association relationship model, the equipment operation state data is used to predict the equipment operation state in the future period, combined with the characteristics of the maintenance operations (such as some maintenance operations requiring equipment shutdown, while some can be performed while the equipment is running) and the maintenance sequence in the maintenance task allocation scheme, to determine the best time window for each maintenance operation, thereby obtaining the optimal maintenance strategy.

[0082] In an embodiment, a long short-term memory network (LSTM) or a gated recurrent unit (GRU) can be employed for dynamic adjustment analysis of the maintenance time window. The equipment operation state data is input into the network in time series, and the network learns the variation law of the equipment operation state based on historical data. At the same time, the relationships between maintenance operations and equipment operation states are modeled based on the relevant information in the optimal maintenance path, the optimal maintenance task allocation scheme, and the association relationship model. Through the prediction ability of the network, the equipment operation state in the future different time periods is predicted in advance, and the best maintenance time window for each maintenance operation is determined, thereby obtaining the optimal maintenance strategy. This deep learning-based time series prediction method can better capture the dynamic variation characteristics of the equipment operation state, improving the accuracy of maintenance time window prediction.

[0083] Step S60: According to the optimal maintenance strategy, multi-objective optimization iterative analysis is performed to obtain the final maintenance decision of the industrial equipment.

[0084] Multi-objective optimization refers to an optimization process in which multiple objective functions are considered simultaneously in an optimization problem. In industrial equipment maintenance decision-making, multiple objectives often include maintenance cost, maintenance time, equipment reliability, equipment service life, etc. There is often a mutually restrictive relationship between these objectives, for example, reducing maintenance cost may result in longer maintenance time or lower equipment reliability. When performing multi-objective optimization iterative analysis based on the optimal maintenance strategy, an optimal solution that balances all objectives is sought under the premise of meeting all objective constraint conditions.

[0085] In an embodiment, an artificial intelligence-based evolutionary multi-objective optimization algorithm such as the non-dominated sorting genetic algorithm (NSGA-II) can be used for multi-objective optimization iterative analysis. Each scheme in the optimal maintenance strategy is encoded as a chromosome, and each chromosome represents a possible maintenance decision. The multi-objectives of maintenance cost, maintenance time, equipment reliability, etc. are used as components of the fitness function. Through selection, crossover and mutation operations of the NSGA-II algorithm, the population is constantly evolving, and in each generation of population, excellent individuals (maintenance decision schemes) are selected according to non-dominated sorting and crowding distance calculation. After multiple iterations, a set of Pareto optimal solutions is obtained, i.e. these solutions cannot be improved in any one objective without degrading other objectives. Finally, according to the actual needs of the enterprise (such as the degree of emphasis on maintenance cost, the minimum requirement for equipment reliability, etc.), a final maintenance decision is selected from the Pareto optimal solutions. The NSGA-II algorithm can effectively handle the complex relationships between multiple objectives, provide multiple representative optimal solutions for the enterprise to choose, and better meet the multi-objective optimization needs in actual industrial equipment maintenance decision-making.

[0086] In an embodiment, reference is made to Figure 4 To further improve the accuracy of the decision, step S30 of the present application can specifically include:

[0087] Step S301: According to the correlation relationship model and the equipment operating state data, analyze the dynamic response relationship between the failure type occurrence probability and the maintenance path execution sequence to obtain the influence weight of the failure type on the maintenance path.

[0088] Among them, the association relationship model is a data model constructed based on the relevant knowledge of industrial equipment, which describes the complex relationships between entities such as equipment operating state, fault type, maintenance behavior, etc. These relationships are represented through quantification and logical connection, and are an important basis for analyzing equipment maintenance decisions. For example, it may indicate the likelihood of a particular fault type occurring under a certain equipment operating state, as well as the degree of association with the corresponding maintenance behavior.

[0089] Among them, the fault type occurrence probability refers to the likelihood of a certain fault type occurring under a particular equipment operating state. The calculation of this probability is based on multiple factors, including historical fault data of the equipment, current operating parameters, environmental factors, etc. For example, for a motor that often operates in a high-temperature environment, the occurrence probability of winding short-circuit fault may increase with the increase of operating time and temperature.

[0090] Among them, the maintenance path execution sequence is the order of a series of maintenance operations planned for different fault types. This sequence is determined according to the structure of the equipment, the fault occurrence mechanism, and the effectiveness and efficiency of maintenance, etc. For example, for a complex equipment that contains multiple components, if a fault occurs, it may need to first check the external connecting components (such as line connections, interfaces, etc.), and then further check the internal core components (such as circuit boards, sensors, etc.), which is a maintenance path execution sequence.

[0091] In the embodiments of the present application, the association relationship model provides basic relationship information between fault types and maintenance paths, and the equipment operating state data reflects the current actual operating condition of the equipment. There is a dynamic response relationship between the fault type occurrence probability and the maintenance path execution sequence. For example, when the equipment operating state data shows that the temperature of a certain component of the equipment suddenly rises, the occurrence probability of a certain temperature-related fault type (such as overheating fault) will increase. This increase in probability may cause the maintenance path execution sequence to change, and the inspection or processing operation for the overheating fault that was originally in the later part of the maintenance path may be advanced.

[0092] By analyzing this dynamic response relationship, the embodiments of the present application can obtain the influence weight of the fault type on the maintenance path. This weight is a quantitative value that reflects the importance of the fault type in influencing the selection of the maintenance path. For example, if a certain fault type, once it occurs, will cause serious damage to the equipment and develop rapidly, then its influence weight on the maintenance path will be high, meaning that the maintenance operation for this fault type needs to be given priority in determining the maintenance path.

[0093] In an embodiment, a Markov chain model is constructed based on the association relationship model and the equipment operation state data. In this model, the state space represents different fault types and maintenance path execution stages. For example, different states of fault types (such as the initial stage, the middle stage, and the serious stage of a fault) and each operation step in the maintenance path execution process are regarded as different states. Then, simulation is performed through a MCMC (Markov Chain Monte Carlo) algorithm, and the transition of the fault type occurrence probability and the corresponding adjustment of the maintenance path execution order are estimated according to the dynamic information (such as the change trend of the equipment operation parameters) in the equipment operation state data. In the simulation process, the change frequency and amplitude of the maintenance path execution order when the fault type occurrence probability changes are counted, so as to calculate the influence weight of the fault type on the maintenance path. This method can effectively handle complex dynamic relationships and can take into account the uncertainty in the data.

[0094] Step S302: According to the influence weight and the association relationship model, a historical matching degree matrix of the fault type and the maintenance path is analyzed to obtain a path selection constraint condition.

[0095] The influence weight is a quantitative value reflecting the importance degree of the fault type in influencing the selection of the maintenance path. This value is obtained by analyzing the dynamic response relationship between the fault type occurrence probability and the maintenance path execution order, and is used as an important reference factor in subsequent analysis.

[0096] The historical matching degree matrix is a matrix constructed based on historical data, which records the matching success between different fault types and various maintenance paths in the past equipment maintenance. The rows of the matrix represent different fault types, the columns represent different maintenance paths, and the elements in the matrix represent the matching success number, success rate, or other related matching degree values between a specific fault type and a maintenance path. For example, if the number of times that the maintenance path B is used to successfully repair the equipment for the fault type A is relatively high in history, then the element value corresponding to the A row and the B column in the historical matching degree matrix is relatively high.

[0097] In the embodiments of the present application, the influence weight reflects the importance degree of the fault type on the maintenance path, the association relationship model contains rich device-related entity relationship information, and the historical matching degree matrix of the fault type and the maintenance path provides historical experience data. By analyzing the influence weight and the association relationship model, the internal relationship between the fault type and the maintenance path can be better understood.

[0098] Then, the path selection constraints can be determined by combining the historical matching degree matrix. These constraints include that the repair path must satisfy certain historical success rate requirements, repair operation sequence restrictions for specific fault types, etc. For example, if a fault type has a high impact weight on the repair path, and the historical matching degree matrix shows that only a specific repair path has a high success rate for this fault type, then this repair path becomes an important constraint, i.e. the repair path is prioritized or other irrelevant repair paths are excluded.

[0099] In an embodiment, the impact weight, related information in the association relationship model, and data in the historical matching degree matrix are integrated to form a data set suitable for mining. Then, an association rule mining algorithm such as the Apriori algorithm is applied to mine the association rules between fault types, impact weights, historical matching degrees, and repair paths. These association rules are the path selection constraints. For example, through mining, the rule "if the impact weight of fault type A is greater than a certain threshold, and the matching degree with repair path B is the highest in the historical matching degree matrix, then repair path B is the preferred path" is obtained. This technical implementation can automatically discover hidden relationships from a large amount of historical data, providing a basis for determining path selection constraints.

[0100] Step S303: Obtain candidate repair paths according to the path selection constraints, and obtain a candidate repair path set.

[0101] The path selection constraints are a set of conditions for limiting and screening repair paths. These conditions are derived by considering factors such as the impact weight of fault types on repair paths, historical matching degree matrices, and association relationship models, and include various requirements that repair paths must satisfy, such as repair operation sequence, historical success rate, etc.

[0102] In this application, the path selection constraints are the basis for determining candidate repair paths. These constraints limit the range of possible repair paths. For example, according to the constraints obtained in the previous steps, if it is specified that certain specific detection operations must be performed first for a certain fault type, then repair paths that do not include these detection operations will not be included in the candidate range.

[0103] In actual industrial equipment maintenance scenarios, there may be multiple repair paths that satisfy the path selection constraints. For example, for a large device with multiple subsystems, when a fault occurs in a certain subsystem, there may be different repair paths that satisfy the constraints determined based on fault types, impact weights, and historical matching degrees. These repair paths that satisfy the constraints constitute the candidate repair path set.

[0104] In an embodiment, the path selection constraints can be inputted into the inference engine in the form of rules. The inference engine traverses the pre-defined maintenance path library and filters out the maintenance paths that satisfy the path selection constraints according to the rules, forming a candidate maintenance path set. For example, if the rule specifies that “for fault type X, the maintenance path must contain operation Y and cannot contain operation Z”, the inference engine will search and filter in the maintenance path library according to this rule. This way is simple and direct, and can quickly and accurately obtain the candidate maintenance path according to the rules.

[0105] In an embodiment, the candidate maintenance path set can also be obtained by using simulated annealing algorithm to calculate the comprehensive evaluation function of maintenance path cost and maintenance effect according to the path selection constraints

[0106] Simulated annealing algorithm is a kind of random search algorithm based on the principle of physical annealing process. In this scenario, it can be used to optimize the search of the comprehensive evaluation function of maintenance path cost and maintenance effect under the condition of satisfying the path selection constraints. Simulated annealing algorithm accepts worse solutions with a certain probability to jump out of local optimal solution, so as to have a greater chance to find the global optimal solution. Just like the metal annealing process, atoms have enough energy to jump out of the local minimum state at high temperature to find a more stable global minimum state, the “temperature” parameter in simulated annealing algorithm controls the probability of accepting worse solutions, and as the “temperature” gradually decreases, the algorithm gradually converges to the optimal solution.

[0107] The maintenance path cost includes various expenses related to the maintenance path. It can include direct costs such as the purchase cost of spare parts required for maintenance, the use cost of maintenance tools, the labor cost of maintenance personnel, etc.; and indirect costs such as the production loss cost caused during equipment downtime, the energy consumption cost during maintenance, etc. For example, for a large industrial equipment, if a kind of maintenance path needs to use expensive imported spare parts and has a long maintenance time, resulting in long equipment downtime, then the cost of this maintenance path is relatively high.

[0108] The maintenance effect is a measure of whether the maintenance path can effectively repair the equipment fault and the degree of performance recovery after repair. It can be evaluated from multiple aspects, such as the success rate of fault repair, the stability of equipment operation after maintenance, the interval time of equipment failure again, etc. For example, if a kind of maintenance path can completely solve the equipment fault and make the equipment run stably for a long time, then its maintenance effect is better.

[0109] The comprehensive evaluation function is a mathematical function that comprehensively considers the maintenance path cost and the maintenance effect. Its purpose is to compare and select between different maintenance paths by assigning appropriate weights to the maintenance cost and the maintenance effect and unifying the two factors into a quantitative index. For example, the maintenance effect can be quantified in a certain way and then weighted and summed with the maintenance cost to obtain a comprehensive score, which is the output result of the comprehensive evaluation function.

[0110] Specifically, the search space is determined according to the path selection constraints. This means that among all possible maintenance paths, the maintenance paths that meet the constraints such as maintenance operation sequence, required resources, etc. are screened out as the initial search range. For example, if the constraint requires that the maintenance of a certain fault must first perform a detection operation, then the maintenance path that does not contain this detection operation is not within the search range.

[0111] Then, the comprehensive evaluation function of the maintenance path cost and the maintenance effect is constructed. This function needs to determine the quantification method of the maintenance cost and the maintenance effect and their weights in the comprehensive evaluation according to the actual situation. For example, the maintenance cost can be set as C, the maintenance effect as E, and the comprehensive evaluation function F = w1 × C + w2 × E, where w1 and w2 are the weights of the maintenance cost and the maintenance effect, respectively, and w1 + w2 = 1. The selection of the weights depends on the importance of the enterprise to the cost and the effect. If the enterprise pays more attention to cost control, it may assign a larger value to w1; if it pays more attention to the rapid and effective repair of the equipment, it may assign a larger value to w2.

[0112] The simulated annealing algorithm searches in this search space and comprehensive evaluation function. At the beginning of the algorithm, a higher "temperature" value is set, at which the algorithm will accept worse solutions with a higher probability in order to explore in a larger search space. As the "temperature" gradually decreases, the probability of accepting worse solutions also decreases, and the algorithm gradually converges to better solutions. During the search process, a new maintenance path is generated each time (by randomly adjusting the current path, such as changing the sequence of maintenance operations or replacing maintenance tools, etc.), and its comprehensive evaluation function value is calculated. If the evaluation function value of the new path is better than the current optimal solution or meets a certain acceptance probability (according to the probability acceptance criterion of the simulated annealing algorithm), the new path is accepted as the current optimal solution.

[0113] Through continuous iteration of this process, the algorithm eventually obtains a series of maintenance paths that meet the path selection constraints and have better comprehensive evaluation function values, which constitute the candidate maintenance path set.

[0114] Step S304: determining the optimal maintenance path according to the candidate maintenance path set.

[0115] The candidate maintenance path set is a set composed of maintenance paths that meet the path selection constraint conditions. Each element (maintenance path) in this set is screened after considering factors such as the influence weight of fault types on maintenance paths, the historical matching degree matrix, and the association relationship model, and is a potential maintenance path that can be used to repair the equipment fault.

[0116] The candidate maintenance path set contains multiple possible maintenance paths, but the optimal maintenance path needs to be determined from them. Determining the optimal maintenance path needs to consider multiple factors, such as maintenance cost, maintenance time, maintenance success rate, etc. For example, a candidate maintenance path may have a lower maintenance cost but a longer maintenance time; while another candidate maintenance path may have a shorter maintenance time but a relatively lower maintenance success rate.

[0117] In order to determine the optimal maintenance path, each candidate maintenance path can be quantitatively evaluated. For example, different weights are assigned to factors such as maintenance cost, maintenance time, and maintenance success rate, and then the comprehensive score of each candidate maintenance path is calculated. The candidate maintenance path with the highest comprehensive score is the optimal maintenance path. In addition, factors such as the current production task arrangement of the equipment and the remaining service life of the equipment can also be considered to affect the maintenance path.

[0118] In an embodiment, a hierarchical structure model can be constructed first, with factors such as maintenance cost, maintenance time, and maintenance success rate as the criterion layer, and the candidate maintenance paths as the scheme layer. Then, a judgment matrix is constructed by pairwise comparison to determine the relative importance between factors, i.e., the weight. Next, the weight vector of each candidate maintenance path relative to each criterion is calculated, and finally the comprehensive score of each candidate maintenance path is obtained by weighted summation. The candidate maintenance path with the highest comprehensive score is the optimal maintenance path. This method can systematically consider the relationship between multiple factors, thereby scientifically determining the optimal maintenance path.

[0119] In an embodiment, step S304 can specifically include:

[0120] According to the candidate maintenance path set, the execution efficiency index of each path is calculated to obtain a path execution efficiency ranking; when the path execution efficiency ranking meets a preset efficiency threshold, the candidate maintenance path with the highest ranking is determined as the optimal maintenance path; when the path execution efficiency ranking does not meet the preset efficiency threshold, the candidate maintenance path set is subjected to local search optimization to obtain the optimal maintenance path.

[0121] The execution efficiency indicator is a quantitative indicator that comprehensively measures the efficiency of the maintenance path in actual execution. It can include multiple factors such as maintenance time, resource utilization, operation complexity, etc. Maintenance time refers to the time spent from the start of maintenance operation to the recovery of normal operation of the equipment, including the duration of maintenance operation and possible waiting time (such as waiting for parts to arrive, waiting for specific maintenance personnel to arrive, etc.); resource utilization refers to the degree of effective utilization of various resources (such as maintenance tools, maintenance personnel, spare parts, etc.) during maintenance, for example, if a maintenance path can fully utilize existing maintenance tools and personnel skills, without causing resource idling or waste, it will score higher in resource utilization; operation complexity reflects the difficulty of maintenance operation, simpler operation is usually easier to execute and has lower error probability, which also affects the execution efficiency indicator.

[0122] The execution efficiency indicator of each maintenance path in the candidate maintenance path set is calculated by the embodiment of the application, which is a comprehensive evaluation process. First, consider the maintenance time factor. Assuming that in the maintenance scenario of an automated production equipment, a candidate maintenance path needs to troubleshoot and repair the control system of the equipment. If this path involves multiple complex detection steps, and there is a long waiting time between each step (such as waiting for the feedback of a certain detection result before proceeding to the next step), its maintenance time is relatively long. On the contrary, another maintenance path may use a more direct detection and repair method, which can quickly locate and solve the problem, and its maintenance time is shorter.

[0123] In terms of resource utilization, take a maintenance scenario that includes a variety of professional tools and maintenance personnel with different skills as an example. If a maintenance path requires the use of a specific professional tool that is in short supply within the enterprise, and requires the full-time participation of maintenance personnel with specific skills, it may cause other maintenance personnel and tools to be idle, resulting in low resource utilization. Another maintenance path can flexibly allocate existing maintenance personnel and tools to fully utilize them, so this path has an advantage in resource utilization.

[0124] For some complex industrial equipment, such as large-scale chemical reaction equipment, if a maintenance path contains a large number of maintenance steps that require high-precision operation, the skill requirements for maintenance personnel are extremely high and the operation is prone to error, so its operation complexity is high. In contrast, a maintenance path that is relatively simple to operate and has ordinary skill requirements for maintenance personnel scores higher in operation complexity.

[0125] The execution efficiency index of each candidate maintenance path is obtained by comprehensively calculating the maintenance time, resource utilization rate, operation complexity and other factors according to certain weights (for example, the weights of different factors can be determined according to the actual situation of the enterprise, such as 40% for maintenance time, 30% for resource utilization rate, and 30% for operation complexity). Then, the candidate maintenance paths are sorted according to these indexes to obtain the path execution efficiency ranking.

[0126] The preset efficiency threshold is a standard value preset by the enterprise according to its production requirements, equipment importance, cost control and other factors. It is used to measure whether the execution efficiency of the candidate maintenance path meets the acceptable level. For example, if the enterprise requires that the equipment maintenance must be completed in a short time to reduce the impact on production, the efficiency threshold may be set higher, requiring the execution efficiency index of the maintenance path to reach a higher value; if the equipment is not a critical equipment and the enterprise has a higher tolerance for maintenance time, the efficiency threshold may be relatively lower.

[0127] In the embodiments of the present application, local search optimization is an optimization method that searches and improves in the local range of the candidate maintenance path set. It is based on the current candidate maintenance path set and attempts to improve the execution efficiency index by adjusting part of the maintenance path. This adjustment is usually to optimize some local operations, resource allocation, etc. without changing the overall framework of the maintenance path. For example, in a maintenance path, if it is found that a maintenance operation can use more efficient tools or can be performed in parallel with other operations, the execution efficiency of the entire maintenance path can be improved through this local adjustment.

[0128] After obtaining the path execution efficiency ranking, the present application first compares the ranking with the preset efficiency threshold. If the path execution efficiency ranking meets the preset efficiency threshold, it means that some maintenance paths in the current candidate maintenance path set have reached the acceptable execution efficiency level of the enterprise. In this case, it is reasonable to directly determine the candidate maintenance path with the highest ranking (i.e. the highest execution efficiency index) as the optimal maintenance path. Because this maintenance path not only meets the various constraint conditions (such as path selection constraint condition, cost and effect comprehensive evaluation, etc.) in the front, but also has the highest execution efficiency, it is the most suitable maintenance solution that can be found at present.

[0129] However, if the path execution efficiency ranking does not meet the preset efficiency threshold, it means that the execution efficiency of the current candidate maintenance path set as a whole is not ideal and needs to be further optimized. At this time, a local search optimization method is used to process the candidate maintenance path set. For example, for an electronic equipment maintenance scene, some maintenance paths in the candidate maintenance path set have low execution efficiency, which may be because the order of some maintenance operations is not reasonable or the utilization of some resources is not sufficient. Through local search optimization, the order of maintenance operations in these maintenance paths can be adjusted to see if the maintenance time can be shortened, or the resource allocation can be optimized to improve resource utilization and thus improve the execution efficiency index.

[0130] In the local search optimization process, various technical means can be used. For example, detailed operation flow analysis can be performed on each candidate maintenance path to find out possible bottlenecks (such as a long time-consuming operation or an idle resource), and then an improvement scheme is proposed for these bottlenecks. Some heuristic algorithms, such as simulated annealing algorithm (applied in a local range) or local search strategy of genetic algorithm, can be used to locally adjust the maintenance path without changing the overall framework of the maintenance path, and new schemes are constantly tried until the optimal maintenance path that meets the preset efficiency threshold is found.

[0131] In an embodiment, the local search optimization can be based on a greedy algorithm

[0132] For the candidate maintenance path set that does not meet the preset efficiency threshold, the first maintenance operation of each maintenance path is analyzed. For each maintenance operation, it is checked whether there is an alternative operation that can improve the execution efficiency or whether the operation order can be adjusted to improve the efficiency. For example, if a maintenance operation needs to wait for a long time before the next operation can be performed, it can be checked whether other operations can be inserted into this waiting time to be performed in advance. If an adjustment scheme that can improve the efficiency is found, the maintenance path is locally modified. Then the execution efficiency index of the maintenance path is recalculated.

[0133] The foregoing steps are repeated until each maintenance path in the set has undergone a round of local search optimization.

[0134] If after a round of optimization, there is a maintenance path that meets the preset efficiency threshold, the maintenance path with the highest execution efficiency index is selected as the optimal maintenance path; if there is still no maintenance path that meets the threshold, the next round of local search optimization is continued until the optimal maintenance path that meets the requirements is found or the maximum number of search rounds is reached.

[0135] In an embodiment, with reference to Figure 5 , step S40 can specifically include the following steps:

[0136] Step S401: According to the association relationship model, the nonlinear relationship between maintenance task allocation and maintenance resource utilization is analyzed to obtain a maintenance resource nonlinear relationship model.

[0137] Maintenance task allocation refers to the process of decomposing the entire maintenance work into multiple specific tasks and determining which maintenance resources (such as personnel, tools, spare parts, etc.) are used to perform each task. For example, for the maintenance of a large device, it may include fault detection tasks, spare parts replacement tasks, device debugging tasks, etc., and each task requires specific maintenance resources.

[0138] Maintenance resource utilization involves the use of various resources during the maintenance process, including resource input, usage efficiency, idle time, etc. For example, maintenance personnel's work schedule, maintenance tool usage frequency, and spare parts inventory consumption are all manifestations of maintenance resource utilization.

[0139] Nonlinear relationship means that the relationship between maintenance task allocation and maintenance resource utilization is not simply linearly proportional. For example, increasing the number of maintenance personnel does not necessarily linearly reduce maintenance time, and there may be an optimal personnel configuration range. If the range is exceeded or below, the maintenance efficiency will decrease. This is because there are complex factors such as task collaboration, resource competition, and operation space limitations in the maintenance process.

[0140] The embodiments of the present application can extract information related to maintenance tasks and maintenance resources from the association relationship model. For example, the model may contain the skill types and resource categories required for maintenance tasks corresponding to different fault types. If it is an electrical fault maintenance, it may require electrician personnel and electrical detection tools; mechanical fault maintenance requires mechanical maintenance personnel and corresponding mechanical tools.

[0141] Then, consider the impact of the sequence and dependency relationship between maintenance tasks on maintenance resource utilization. Suppose a device maintenance includes multiple tasks, some of which must be completed after other tasks. This sequence relationship will affect the scheduling and utilization efficiency of maintenance resources. For example, in a complex industrial production line device maintenance, if a key component needs to be replaced (task A), the device may need to be disassembled first (task B), and specific tools and personnel are required during the disassembly process. If the maintenance resource allocation of task A and task B is unreasonable, it may cause tools to be idle or personnel to wait, thus reflecting the nonlinear resource utilization relationship.

[0142] By analyzing a large number of maintenance task allocation situations and corresponding maintenance resource utilization data (which can be obtained from historical maintenance records, monitoring of actual maintenance scenes, etc.), the complex relationship between various factors is determined. For example, the relationship between the number of maintenance personnel and the maintenance time may not be a simple inverse relationship, and there may be a critical value of the number of people near which the trend of change in maintenance time will change. Based on these analysis results, a nonlinear relationship model of maintenance resources is constructed. This model can be a mathematical equation, a set of logical rules, or a prediction model based on a machine learning algorithm, which is used to describe the nonlinear relationship between maintenance task allocation and maintenance resource utilization.

[0143] In an embodiment, nonlinear relationship modeling based on neural networks can be used, specifically: a large number of maintenance task allocation and maintenance resource utilization data are collected as training samples. These data should include the types of maintenance tasks, the allocation of tasks (such as the number of personnel, tools, etc.), and the corresponding maintenance resource utilization indicators (such as maintenance time, resource idle rate, etc.). A neural network model is constructed, for example, a multi-layer perceptron (MLP). The input layer nodes correspond to the relevant variables of the maintenance task allocation (such as task type, number of personnel, tool type, etc.), and the output layer nodes correspond to the relevant indicators of the maintenance resource utilization (such as maintenance time, resource idle rate, etc.). The collected training samples are used to train the neural network. During the training process, the neural network learns the nonlinear mapping relationship between maintenance task allocation and maintenance resource utilization by adjusting the internal weights and biases. After sufficient training, the obtained neural network model is the nonlinear relationship model of maintenance resources. This model can predict the corresponding maintenance resource utilization results according to the input maintenance task allocation situation. The technical implementation based on neural networks can automatically learn complex nonlinear relationships without prior assumptions about the specific form of the relationship, and is suitable for handling maintenance task allocation and maintenance resource utilization scenarios with a large amount of data and complex relationships.

[0144] Step S402: According to the nonlinear relationship model of maintenance resources and the optimal maintenance path, function decomposition is performed in combination with the preset maintenance resource constraint conditions to obtain a maintenance task allocation optimization function.

[0145] Optimal maintenance path: It is the best maintenance operation sequence and process determined for equipment failure. It can include a series of ordered steps from fault detection, maintenance operation to equipment recovery to normal operation, and is the optimal solution determined after considering factors such as fault type, maintenance cost, maintenance effect, etc.

[0146] The preset maintenance resource constraint condition is a restriction condition set by the enterprise according to the maintenance resource status (such as the number of personnel, the availability of tools and equipment, the inventory of spare parts, etc.) of the enterprise. For example, the enterprise can stipulate that the number of maintenance personnel with specific skills cannot exceed a certain number when maintaining a certain type of equipment, or that a certain expensive maintenance tool can only be used a limited number of times.

[0147] Function decomposition is a method of converting a complex relationship or model into a function form that is easier to analyze and optimize. In this step, the maintenance resource nonlinear relationship model, the optimal maintenance path, and the preset maintenance resource constraint condition are integrated to decompose a function that is specifically used for maintenance task allocation optimization.

[0148] In the embodiments of the present application, the maintenance resource nonlinear relationship model provides a basic relationship framework for maintenance task allocation and maintenance resource utilization, the optimal maintenance path determines the operation sequence of maintenance work, and the preset maintenance resource constraint condition sets the boundary for resource use.

[0149] First, each maintenance task in the optimal maintenance path is matched with the relevant part in the maintenance resource nonlinear relationship model according to the order and type. For example, if the first maintenance task in the optimal maintenance path is electrical fault detection, then find the relevant part of the maintenance resource utilization in the maintenance resource nonlinear relationship model, such as the number of electrician personnel required, the use of electrical detection tools, etc. Then, combined with the preset maintenance resource constraint condition, integrate these information into a function. For example, let the number of maintenance personnel be x, the number of maintenance tool usage be y, and the execution time of the maintenance task be z. According to the maintenance resource constraint condition, there can be constraints such as x ≤ X (X is the preset upper limit of the number of personnel), y ≤ Y (Y is the preset upper limit of the number of tool usage), etc. According to the maintenance resource nonlinear relationship model, there can be a relationship z = f (x, y) (f is a nonlinear function). Considering the time sequence relationship between the maintenance tasks in the optimal maintenance path and other factors, a maintenance task allocation optimization function containing all these information is constructed. The goal of this function is to optimize the maintenance task allocation under the premise of meeting the maintenance resource constraint condition, in order to achieve certain optimal effect (such as shortest maintenance time, lowest maintenance cost, etc.).

[0150] In an embodiment, a mathematical expression between the maintenance task allocation and the maintenance resource utilization variables can be determined according to the maintenance resource nonlinear relationship model. For example, if there is a certain nonlinear relationship between the maintenance time and the number of maintenance personnel and the number of tool usage, it can be expressed as t = g (p, t), where t is the maintenance time, p is the number of maintenance personnel, t is the number of tool usage, and g is a nonlinear function.

[0151] The sequence and time relationship of maintenance tasks in the optimal maintenance path are converted into mathematical constraint conditions. For example, if maintenance task A must be completed before maintenance task B, and the execution time of maintenance task A is t1 and the execution time of maintenance task B is t2, the constraint condition t1 + Δt ≤ t2 (Δt is the minimum time interval between task A and task B) can be obtained.

[0152] A complete maintenance task allocation optimization function is constructed in combination with the preset maintenance resource constraints, such as the number of maintenance personnel limit p ≤ P (P is the preset number of personnel) and the number of tool use limit t ≤ T (T is the preset number of tool use). This function can be in the form of a target function (such as minimizing maintenance time or cost) plus a series of constraint conditions, for example, min (t) s.t. p ≤ P, t ≤ T, t1 + Δt ≤ t2, etc.

[0153] The above implementation based on mathematical programming converts various relationships and constraints into explicit mathematical forms, facilitating the use of optimization algorithms for solving.

[0154] Step S403: According to the maintenance task allocation optimization function, a particle swarm optimization algorithm is used for iterative calculation to obtain a set of maintenance task allocation solutions that satisfy the maintenance resource constraints.

[0155] Particle swarm optimization algorithm is a kind of optimization algorithm based on swarm intelligence, simulating the group behavior of bird swarm or fish swarm. In the algorithm, each particle represents a possible solution (in this step, a maintenance task allocation scheme), and the particle flies in the search space by constantly updating its position to find the optimal solution. The position of the particle represents the specific scheme of the maintenance task allocation (such as personnel allocation to each task, tool allocation, etc.), and the speed of the particle determines its moving direction and speed in the search space.

[0156] The set of maintenance task allocation solutions is a set of maintenance task allocation schemes that satisfy the maintenance resource constraints. Through the iterative calculation of the particle swarm optimization algorithm, multiple possible maintenance task allocation schemes are obtained, and these schemes are all within the allowable range of the maintenance resource constraints.

[0157] In an embodiment, the particle swarm optimization algorithm can start iterative calculation based on the maintenance task allocation optimization function, specifically:

[0158] First, a swarm of particles is initialized, each representing an initial maintenance task assignment scheme. These initial schemes can be randomly generated but must satisfy the maintenance resource constraints. For example, for each particle, randomly assign maintenance personnel to each maintenance task, ensuring that the number of assigned personnel does not exceed the pre-set personnel constraint, and the assignment of tools also complies with the corresponding constraints.

[0159] Then, the fitness value of each particle is calculated. The fitness value is calculated according to the maintenance task assignment optimization function, which reflects the pros and cons of each maintenance task assignment scheme. For example, if the goal of the maintenance task assignment optimization function is to minimize the maintenance time, then the fitness value can be the inverse of the maintenance time, and the shorter the maintenance time, the higher the fitness value.

[0160] In each iteration, the particle updates its speed and position according to its own experience (individual optimal position) and the experience of the group (global optimal position). The individual optimal position refers to the best position (i.e. maintenance task assignment scheme) found by each particle during the search process, and the global optimal position refers to the best position found in the entire particle swarm. The particle adjusts its speed and moves towards a better position, thereby exploring different maintenance task assignment schemes.

[0161] As the iteration proceeds, the particle swarm gradually converges to a region that satisfies the maintenance resource constraints, and eventually obtains a set of maintenance task assignment schemes, i.e. the maintenance task assignment solution set. These schemes are within the scope of the maintenance resource constraints and are the better solutions found by the particle swarm optimization algorithm.

[0162] In an embodiment, the size of the particle swarm can be determined, for example, by setting the number of particles to N.

[0163] For each particle i (i = 1, 2, …, N), randomly initialize its position vector Xi and velocity vector Vi. The position vector Xi represents the maintenance task assignment scheme, for example Xi = [x1, x2, …, xn], where xj represents the resource (such as personnel, tools, etc.) assigned to the jth maintenance task; the velocity vector Vi represents the movement speed of the particle, for example Vi = [v1, v2, …, vn].

[0164] According to the maintenance task assignment optimization function, the fitness value Fi of each particle is calculated.

[0165] Initialize the individual optimal position Pbest_i of each particle as its initial position Xi, and the global optimal position Gbest as the position of the particle with the highest fitness value.

[0166] In each iteration k, for each particle i, its velocity and position are updated according to the following formulas:

[0167] Vi(k + 1)= w * Vi(k)+ c1 * r1 * (Pbest_i - Xi(k))+ c2 * r2 * (Gbest -Xi(k))

[0168] Xi(k + 1)= Xi(k)+ Vi(k + 1)

[0169] where w is the inertia weight, used to balance the global search and local search capabilities of particles; c1 and c2 are learning factors, respectively controlling the degree of particles approaching individual optimal position and global optimal position; r1 and r2 are random numbers between 0 and 1.

[0170] According to the updated position Xi (k + 1), the fitness value Fi (k + 1) of the particle is recalculated.

[0171] If Fi (k + 1) is better than the fitness value of Pbest_i, update Pbest_i = Xi (k + 1); if Fi (k + 1) is better than the fitness value of Gbest, update Gbest = Xi (k + 1).

[0172] Repeat the above steps until the preset number of iterations is reached or the convergence condition is met (such as the global optimal position no longer changes in multiple iterations).

[0173] Finally, all Pbest_i form a set of maintenance task allocation solutions that meet the maintenance resource constraints.

[0174] This standard particle swarm optimization algorithm implementation is an effective solution method, which can improve efficiency and accuracy by continuously adjusting the velocity and position of particles to search for the optimal solution.

[0175] Step S404: According to the maintenance task allocation solution set and the maintenance resource nonlinear relationship model, analyze the balance of maintenance task allocation, and determine the solution with the best balance in the maintenance task allocation solution set as the optimal maintenance task allocation scheme

[0176] In the context of maintenance task allocation, balance refers to the uniformity of maintenance tasks allocated among different maintenance resources (such as personnel, tools, etc.). For example, in terms of personnel allocation, if some maintenance personnel are allocated too many tasks while others have few tasks, it indicates that the task allocation is unbalanced; in terms of tool usage, if some tools are overused while others are idle, it is also an unbalanced performance. Balanced task allocation helps to improve maintenance efficiency, reduce resource waste, and reduce personnel fatigue, etc.

[0177] In the embodiments of the present application, the maintenance task allocation solution set contains multiple maintenance task allocation schemes that meet the maintenance resource constraints, and the optimal scheme needs to be selected from these schemes. Specifically:

[0178] First, the balance of each maintenance task allocation scheme is analyzed according to the maintenance resource nonlinear relationship model. For the balance of personnel allocation, the difference degree of the task workload allocated to each maintenance personnel can be calculated. For example, by calculating the standard deviation of the workload allocated to each personnel, if the standard deviation is small, it indicates that the personnel task allocation is relatively balanced. For the balance of tool usage, the difference in usage frequency of tools can be analyzed. If a certain tool is used very frequently in a maintenance task allocation scheme, while other tools are rarely used, it indicates that the tool usage is unbalanced.

[0179] In addition to considering the balance of personnel and tools, the balance among maintenance tasks also needs to be considered. For example, some maintenance tasks may be executed for too long due to unreasonable resource allocation, affecting the efficiency of the entire maintenance process. If in a maintenance task allocation scheme, the execution time of each maintenance task is relatively close, then this scheme performs better in terms of maintenance task balance.

[0180] By considering the balance of maintenance personnel, tools, and maintenance tasks, etc. from multiple aspects, each scheme in the maintenance task allocation solution set is evaluated. The scheme that is best in these aspects is determined, which is the optimal maintenance task allocation scheme. For example, if a scheme has smaller differences in personnel task allocation, tool usage, and maintenance task execution time, etc. compared to other schemes, it is the optimal maintenance task allocation scheme that can achieve effective use of resources and efficient maintenance process.

[0181] In an embodiment, based on the evaluation of the balance index, specifically:

[0182] For each solution in the repair task allocation solution set, a personnel task allocation balance index is defined. For example, let the set of repair personnel be {P1, P2, …, Pm}, and the workload of each personnel Pi be Wi. Calculate the personnel task allocation balance index E1 = √(∑(Wi - W_avg)^2 / m), where W_avg is the average workload. The smaller the E1 value, the more balanced the personnel task allocation.

[0183] A tool usage balance index is defined. Let the set of tools be {T1, T2, …, Tn}, and the usage frequency of each tool Tj be Fj. Calculate the tool usage balance index E2 = √(∑(Fj - F_avg)^2 / n), where F_avg is the average usage frequency. The smaller the E2 value, the more balanced the tool usage.

[0184] A repair task execution time balance index is defined. Let the set of repair tasks be {M1, M2, …, Mk}, and the execution time of each task Mj be Tj. Calculate the repair task execution time balance index E3 = √(∑(Tj - T_avg)^2 / k), where T_avg is the average execution time. The smaller the E3 value, the more balanced the repair task execution time.

[0185] For each repair task allocation solution, calculate the comprehensive balance index E = w1 × E1 + w2 × E2 + w3 × E3, where w1, w2, w3 are weights that can be set according to the importance of personnel, tools, and task balance in the enterprise. Select the repair task allocation solution with the smallest comprehensive balance index E as the optimal repair task allocation solution.

[0186] The above evaluation technique implementation based on balance index quantitatively evaluates the balance of repair task allocation solutions, which can objectively select the optimal solution.

[0187] In an embodiment, with reference to Figure 6 , step S50 can specifically include the following steps:

[0188] Step S501: According to the association relationship model, dynamic coupling analysis of the repair time window and the equipment running state is performed to obtain the equipment running state change characteristics.

[0189] The maintenance time window refers to the time period during which maintenance operations can be performed on the equipment during its operation. Determining this time period requires considering various factors, such as the production plan of the equipment, the stability of the equipment's operation, the availability of maintenance resources, and so on. For example, for some continuously producing equipment, the maintenance time window can only be arranged during the regular maintenance shutdown period of the equipment or during the short gap of production task switching.

[0190] Dynamic coupling analysis is a method of analyzing the dynamic relationship between two or more variables. In this application, the mutual influence relationship between the maintenance time window and the equipment operating state over time is analyzed. For example, changes in the equipment operating state can affect the selection of the maintenance time window, and the selection of the maintenance time window can in turn affect the operating state of the equipment.

[0191] The equipment operating state change feature is a description of the state change pattern of the equipment during operation. This feature can include the change trend of the equipment operating parameters (such as temperature, pressure, speed, etc.), the frequency and pattern of equipment failure, the degradation of equipment performance, and other aspects.

[0192] In an embodiment, information related to the equipment operating state and the maintenance time window can be extracted from the association relationship model. For example, the model can contain information about the time suitable for maintenance under different operating states of the equipment, and information about the impact of different maintenance time windows on the operating state of the equipment. If the equipment is in a high-temperature operating state, the association relationship model may indicate that performing maintenance in this state can increase the risk of equipment failure, so the maintenance time window should be avoided as much as possible; or if the equipment operates relatively stably within a certain time period, the model may suggest that this time period is a suitable maintenance time window. Then, analyze how the dynamic changes in the equipment operating state affect the selection of the maintenance time window. For example, as the equipment operating time increases, some components of the equipment may gradually wear out, and the operating temperature of the equipment may rise, which will affect the selection of the maintenance time window. If the key components of the equipment wear out to a certain extent, maintenance may need to be arranged as soon as possible, which may shorten the originally set maintenance time window; and if the equipment operating state is relatively stable within a certain period of time, the maintenance time window can be appropriately extended.

[0193] By analyzing a large amount of equipment operating data and maintenance time window historical data (which can be obtained from equipment monitoring systems, maintenance records, etc.), the equipment operating state change feature is determined. For example, by analyzing the equipment temperature data, it is found that the temperature fluctuates greatly within a certain time period of each day, which is a kind of equipment operating state change feature; or by analyzing the equipment failure records, it is found that the probability of equipment failure significantly increases after continuous operation for a certain period of time, which is also a manifestation of the equipment operating state change feature.

[0194] In an embodiment, features can be mined based on dynamic coupling analysis of data mining. Specifically, historical data of equipment operating state parameters (such as temperature, pressure, speed, etc.) and maintenance time window are collected as analysis samples. These data are mined using data mining techniques, such as association rule mining algorithm. The goal of mining is to find the association rules between equipment operating state parameters and maintenance time window. For example, a rule such as "when the equipment temperature exceeds threshold A, the maintenance time window should be shortened to [specific time period]" is mined. According to the mined association rules, the trend of changes in equipment operating state and the impact of these changes on the maintenance time window are analyzed. For example, if it is found that the equipment temperature has shown an upward trend in recent times, according to the association rules, it can be known that the maintenance time window may need to be adjusted accordingly. Through multi-dimensional analysis of equipment operating state data (such as considering the changes of temperature, pressure, speed, etc. at the same time), the equipment operating state change characteristics are summarized. For example, if it is found that the equipment temperature and pressure rise at the same time and the speed decreases, the risk of equipment failure increases significantly, which is a comprehensive equipment operating state change characteristic. This implementation based on data mining can automatically discover the hidden relationship between equipment operating state and maintenance time window from a large amount of data, which helps to accurately analyze the equipment operating state change characteristics.

[0195] Step S502: According to the equipment operating state change characteristics, a time series prediction algorithm is used to perform maintenance time window fluctuation analysis to obtain a maintenance time window adjustment interval.

[0196] The time series prediction algorithm is an algorithm that predicts future data trends based on the time sequence of historical data. In this step, it uses historical data of equipment operating state change characteristics to predict the fluctuation of maintenance time window. For example, commonly used time series prediction algorithms include autoregressive moving average model (ARMA), difference autoregressive moving average model (ARIMA), etc.

[0197] Maintenance time window fluctuation analysis is an analysis of the fluctuations of maintenance time window with changes in equipment operating state. Since the equipment operating state is not fixed, the maintenance time window also needs to be adjusted according to the changes in the equipment operating state. This analysis aims to determine the possible adjustment range of the maintenance time window to adapt to the changes in the equipment operating state.

[0198] The maintenance time window adjustment interval is an interval representing the possible adjustment range of the maintenance time window. For example, if the original maintenance time window is [9 am - 5 pm], after fluctuation analysis, the maintenance time window adjustment interval may be [8 am - 6 pm], indicating that the maintenance time window may be adjusted within this range.

[0199] In an embodiment, the device operating state change feature data is sorted in chronological order to form time series data. For example, if the device operating state change feature includes changes in device temperature, pressure, and other parameters, the values of these parameters at different time points are arranged in chronological order to form temperature time series, pressure time series, and the like. Then, a suitable time series prediction algorithm is selected to analyze these time series data. Taking the ARIMA model as an example, first, the data needs to be tested for stationarity. If the data is not stationary, it needs to be differenced to make it stationary. Then, the order of the model (p, d, q) is determined, where p is the autoregressive order, d is the differencing order, and q is the moving average order. Through fitting of historical data, the parameters of the ARIMA model are obtained.

[0200] The obtained time series prediction model is used to predict the device operating state in the future period of time. For example, the changes in temperature, pressure, and other parameters of the device in the future week are predicted. According to these prediction results, the possible impact on the maintenance time window is analyzed. If it is predicted that the temperature of the device will rise to a level that may affect the normal operation of the device in a certain period of time, the maintenance time window may need to be advanced or shortened; if it is predicted that the device will run stably in a certain period of time, the maintenance time window may be appropriately extended.

[0201] By comprehensively considering the prediction results of various parameters of the device operating state, the maintenance time window adjustment interval is determined. This interval takes into account not only the changes in the device operating state, but also actual factors such as the production plan of the enterprise and the availability of maintenance resources. The technical implementation based on the ARIMA model can accurately predict the changes in the device operating state, thereby providing a basis for determining the maintenance time window adjustment interval.

[0202] Step S503: According to the maintenance time window adjustment interval, a priority sorting function of the maintenance plan is constructed to obtain a maintenance time window adjustment strategy.

[0203] The priority sorting function of the maintenance plan is a mathematical function used to determine the priority of the maintenance plan. It determines the priority order of different maintenance tasks according to the maintenance time window adjustment interval and other related factors (such as the severity of device failure, availability of maintenance resources, etc.). For example, if the maintenance time window adjustment interval of a maintenance task is short and the device failure is severe, the priority of this maintenance task in the maintenance plan is high.

[0204] The maintenance time window adjustment strategy is a strategy for adjusting the maintenance plan according to the maintenance time window adjustment interval. It includes determining which maintenance tasks need to be arranged in priority, how to adjust the order of maintenance tasks, whether to adjust the allocation of maintenance resources, etc.

[0205] In an embodiment, first, consider the length of the maintenance time window adjustment interval. If the maintenance time window adjustment interval of a certain maintenance task is short, it means that this maintenance task needs to be arranged as soon as possible, otherwise the optimal maintenance opportunity may be missed. For example, for a failure of a critical component of a device, its maintenance time window adjustment interval may be only one day, then this maintenance task should be in a higher position in the priority order. Second, consider the severity of the device failure. Even if the maintenance time window adjustment intervals of two maintenance tasks are the same, but if one of the failures will cause the device to stop immediately, and the other failure will only affect the partial performance of the device, then the maintenance task that causes the device to stop has a higher priority. This can be reflected by assigning different weights to the severity of the device failure. For example, the weight of the device stop failure is 0.8, and the weight of the partial performance impact failure is 0.2.

[0206] At the same time, the availability of maintenance resources also needs to be considered. If the maintenance resources required by a maintenance task cannot be obtained within a certain period of time, even if its maintenance time window adjustment interval is short and the severity of the failure is high, its priority may need to be adjusted. For example, if a maintenance task requires a specific maintenance tool, and this tool cannot be used for the next three days, then the priority of this maintenance task may need to be lowered, or the allocation of maintenance resources needs to be adjusted to meet the needs of this maintenance task.

[0207] According to these factors, a priority ordering function for the maintenance plan is constructed. For example, let the maintenance time window adjustment interval length of maintenance task i be Ti, the severity of the device failure be Si, and the availability of maintenance resources be Ri, then the priority ordering function of the maintenance plan can be represented as Pi = w1 * (1 / Ti) + w2 * Si + w3 * Ri, where w1, w2, w3 are weights, which can be set according to the actual situation of the enterprise. By calculating the Pi value of each maintenance task, the priority of the maintenance task is sorted, and the maintenance time window adjustment strategy is obtained.

[0208] In an embodiment, the priority ordering function can be constructed based on multi-factor weighting, for example:

[0209] Determine the factors that affect the priority of the maintenance plan, such as the maintenance time window adjustment interval, the severity of the device failure, the availability of maintenance resources, etc.

[0210] Each factor is quantified. For the maintenance time window adjustment interval, the reciprocal of the interval length can be used to represent its urgency; for the equipment failure severity, a score (such as 0-10) can be given according to the impact of the failure on the operation of the equipment; for the maintenance resource availability, quantification can be performed according to the available time of the resource (such as the closer the available time, the higher the quantification value).

[0211] According to the importance of each factor to the enterprise, weights w1, w2, w3, etc. are determined. For example, if the enterprise pays more attention to the urgency of the maintenance time window, w1 can be set larger; if more attention is paid to the severity of the equipment failure, w2 can be set larger.

[0212] A priority ranking function Pi = w1 * (1 / Ti) + w2 * Si + w3 * Ri is constructed for the maintenance plan, where Ti is the maintenance time window adjustment interval length, Si is the equipment failure severity quantification value, and Ri is the maintenance resource availability quantification value.

[0213] According to the calculated Pi values, the maintenance tasks are prioritized, and the maintenance time window adjustment strategy is determined. For example, the maintenance tasks are arranged in order of Pi values from high to low, and the maintenance tasks with high Pi values are prioritized. For maintenance tasks with low Pi values, the maintenance time window or maintenance resource allocation can be adjusted according to the actual situation.

[0214] The above implementation based on multi-factor weighting can comprehensively consider multiple factors, reasonably construct a priority ranking function for the maintenance plan, and thus obtain an effective maintenance time window adjustment strategy.

[0215] Step S504: According to the maintenance time window adjustment strategy, the optimal maintenance path, and the optimal maintenance task allocation scheme, the maintenance plan is overall optimized to obtain an optimal maintenance strategy.

[0216] Maintenance plan overall optimization is a process of comprehensively considering the maintenance time window adjustment strategy, the optimal maintenance path, and the optimal maintenance task allocation scheme, and overall optimizing the entire maintenance plan. The purpose is to develop an optimal maintenance plan under the conditions of meeting equipment maintenance needs, maintenance resource constraints, etc., to improve maintenance efficiency, reduce maintenance cost, and reduce equipment downtime, etc.

[0217] The optimal maintenance strategy is a comprehensive equipment maintenance scheme obtained after overall optimization of the maintenance plan. It includes the best maintenance time arrangement, maintenance task execution sequence, maintenance resource allocation, etc., and is the most beneficial to equipment maintenance after considering various factors.

[0218] In the embodiments of the present application, the execution order of the maintenance tasks can be determined according to the maintenance time window adjustment strategy. The maintenance tasks are arranged in order of priority from high to low to ensure that urgent and important maintenance tasks are processed first within the maintenance time window. For example, if the core component failure maintenance task of a device has the highest priority, this task will be arranged first in the maintenance plan.

[0219] Then, the specific operations of each maintenance task are optimized in combination with the optimal maintenance path. The optimal maintenance path specifies the order of operations in the maintenance task, and during the overall optimization of the maintenance plan, it is ensured that each maintenance task is operated according to the optimal maintenance path. For example, if the optimal maintenance path indicates that the maintenance of a certain fault needs to be detected first and then repaired, the corresponding maintenance operations should be arranged in this order in the maintenance plan.

[0220] At the same time, the maintenance resources are allocated according to the optimal maintenance task allocation scheme. It is ensured that each maintenance task has sufficient maintenance resources during execution, and the maintenance resources are reasonably utilized. For example, if the optimal maintenance task allocation scheme specifies that a certain maintenance task requires a maintenance personnel with a specific skill and a specific maintenance tool, the corresponding personnel and tool should be arranged to participate in the maintenance task at the appropriate time in the maintenance plan.

[0221] Through the comprehensive optimization of the execution order of the maintenance tasks, the operation order of the maintenance tasks, and the allocation of the maintenance resources, the optimal maintenance strategy is obtained. This strategy can maximize the maintenance efficiency, reduce the maintenance cost, and reduce the equipment downtime under the conditions of meeting the equipment maintenance requirements and maintenance resource constraints, etc.

[0222] In an embodiment, the maintenance plan overall optimization can be based on integer programming, for example:

[0223] Let the set of maintenance tasks be {T1, T2, …, Tn}, and define the decision variable xij. If the maintenance task Ti is executed in the jth time unit, then xij = 1, otherwise xij = 0.

[0224] According to the maintenance time window adjustment strategy, the earliest start time Ei and the latest end time Li of each maintenance task Ti are determined. This can be obtained by analyzing the priority of the maintenance task, the maintenance time window adjustment interval, etc.

[0225] According to the optimal maintenance path, the order constraints of each operation in each maintenance task Ti are determined. For example, if the operation O1 in the maintenance task T1 must be executed before the operation O2, the corresponding constraint condition can be obtained.

[0226] According to the optimal maintenance task allocation scheme, the number of maintenance resources required for each maintenance task Ti is determined, and combined with the availability of maintenance resources, the maintenance resource constraint condition is obtained. For example, if maintenance task T1 requires 2 maintenance personnel with specific skills, and the enterprise has only 3 such maintenance personnel available, then the personnel resource constraint condition can be obtained.

[0227] The objective function is constructed, such as minimizing the total maintenance time, minimizing the maintenance cost, or maximizing the equipment availability, etc. For example, if the goal is to minimize the total maintenance time, the objective function Z = ∑∑xij * tij can be constructed, where tij is the maintenance time of maintenance task Ti when executed in the jth time unit.

[0228] According to the above decision variables, constraint conditions and objective function, an integer programming model is established.

[0229] The integer programming solution algorithm (such as branch and bound method, cutting plane method, etc.) is used to solve this integer programming model, and the optimal solution obtained is the optimal maintenance strategy. This optimal solution will determine the execution time of each maintenance task, the maintenance operation sequence, and the maintenance resource allocation method, etc. The technical implementation method based on integer programming can obtain the optimal maintenance strategy through optimization algorithm on the basis of meeting various constraint conditions, and is suitable for complex maintenance plan overall optimization problem.

[0230] In an embodiment, step S504 can be implemented by the following steps:

[0231] According to the maintenance time window adjustment strategy, a dynamic programming algorithm is used for global optimization analysis to obtain a global optimization result; and according to the global optimization result, the maintenance time window adjustment strategy, the optimal maintenance path and the optimal maintenance task allocation scheme, a maintenance plan overall optimization is performed to obtain an optimal maintenance plan and an optimal maintenance strategy.

[0232] The maintenance time window adjustment strategy is a strategy constructed based on maintenance time window adjustment interval, equipment failure severity, maintenance resource availability, etc., which is used to determine the priority order of maintenance tasks and the adjustment method of maintenance time window. It provides basic guidelines for the optimization of maintenance plan, such as which maintenance tasks need to be executed first, and in what time range to perform maintenance, etc.

[0233] Among them, dynamic programming is an algorithm for solving the optimization problem of multi-stage decision-making process. It decomposes a complex problem into a series of interrelated sub-problems and obtains the global optimal solution by solving these sub-problems. In this step, it is based on the maintenance time window adjustment strategy, considering various stages in the maintenance process (such as the sequence of different maintenance tasks, the change of resource allocation in different stages, etc.), and achieving global optimization through step-by-step analysis and calculation.

[0234] Among them, the global optimization result is obtained by dynamic programming algorithm to analyze and optimize the whole maintenance plan. This result comprehensively considers various constraints and objectives of maintenance tasks (such as the shortest maintenance time, the lowest cost, the highest equipment availability, etc.), which contains the optimal decision information about the execution order, time arrangement, resource allocation, etc.

[0235] In an embodiment, first, the stage division of the problem is determined according to the maintenance time window adjustment strategy. For example, maintenance tasks are sorted by priority from high to low, and each maintenance task can be regarded as a stage. In each stage, the execution conditions of maintenance tasks need to be considered, such as whether the required maintenance resources are available, whether it is within the maintenance time window, etc.

[0236] Then, the state of each stage is determined. The state can include the current running state of the equipment, the situation of completed maintenance tasks, the remaining maintenance resources, etc. For example, in a certain stage, the equipment may be in the running test state after partial failure repair, some key components have been completed, and a certain number of maintenance personnel and tools are available.

[0237] Next, the state transition equation is defined. This equation describes how to transfer from one state of a stage to another state of the next stage. For example, if a maintenance task is completed in the current stage, the running state of the equipment may be improved, and the remaining maintenance resources will be reduced accordingly, which is a state transition. The state transition equation needs to consider various rules in the maintenance time window adjustment strategy, such as high-priority maintenance tasks are executed first, reasonable allocation of maintenance resources, etc.

[0238] Through the dynamic programming algorithm, starting from the initial state, the optimal decision in each state is calculated step by step according to the stage order. For example, in each stage, the cost, time, etc. of different maintenance tasks in the current state are calculated, and the optimal maintenance task execution order is selected to make the whole maintenance process achieve global optimization under the condition of meeting various constraints. The final result is the global optimization result, which contains the optimal decision of each stage, such as the best execution time of each maintenance task, the optimal allocation of maintenance resources, etc.

[0239] Since the global optimization result provides preliminary optimization information on maintenance task execution order, time arrangement, and resource allocation, etc., on this basis, further overall optimization is carried out in combination with the maintenance time window adjustment strategy, the optimal maintenance path, and the optimal maintenance task allocation scheme. Specifically:

[0240] According to the maintenance time window adjustment strategy, the execution time of the maintenance task in the global optimization result is fine-tuned. For example, if the global optimization result determines the execution time of a certain maintenance task, but according to the maintenance time window adjustment strategy, considering the change of the running state of the equipment or the temporary adjustment of the maintenance resources, this execution time may need to be advanced or delayed.

[0241] According to the optimal maintenance path, the operation order of the maintenance task is optimized. Although the global optimization result has determined the general execution order of the maintenance task, the optimal maintenance path gives a more detailed operation order within each maintenance task. For example, for the maintenance task of a certain equipment, the global optimization result determines that it is executed in a certain time period, while the optimal maintenance path indicates that the detection operation in this maintenance task should be performed before the repair operation, so the operation order should be clearly defined in the maintenance plan.

[0242] At the same time, according to the optimal maintenance task allocation scheme, the maintenance resources are allocated. Ensure that each maintenance task has enough and appropriate maintenance resources support during execution. For example, if the optimal maintenance task allocation scheme specifies that a certain maintenance task requires maintenance personnel with specific skills and specific maintenance tools, then in the maintenance plan, it is necessary to clearly define the configuration of these resources at the corresponding time and place.

[0243] Through the comprehensive adjustment and optimization of these aspects, the optimal maintenance plan is obtained. This plan realizes the optimal combination of maintenance task execution time, operation order, and resource allocation on the basis of meeting various constraint conditions.

[0244] Further, the optimal maintenance plan is combined with other equipment maintenance related strategies (such as equipment running state monitoring strategy, preventive maintenance strategy, etc.) to form the optimal maintenance strategy. For example, in the optimal maintenance strategy, it can be specified that specific running state monitoring is performed before and after the execution of the maintenance task, or a preventive maintenance plan is formulated according to the running history and maintenance of the equipment, such as regular replacement of vulnerable parts, etc., to ensure the long-term stable operation of the equipment.

[0245] In an embodiment, with reference to Figure 7 , step S60 can specifically include the following steps:

[0246] Step S601: According to the optimal maintenance strategy, multi-objective parameter optimization is carried out to construct a maintenance decision optimization space containing maintenance path, maintenance task allocation, and maintenance time window.

[0247] Multi-objective parameter optimization refers to the process of optimizing adjustments on multiple target-related parameters. In the context of industrial equipment maintenance, factors such as maintenance paths, maintenance task allocation, and maintenance time windows all involve multiple parameters. For example, each operation step in a maintenance path may involve different operation times, resource requirements, and other parameters; maintenance task allocation involves skill level parameters of different maintenance personnel, priority parameters of tasks, and other parameters; maintenance time windows involve start time, duration, and other parameters. Through multi-objective parameter optimization, the goal is to comprehensively consider these parameters to achieve optimization of overall maintenance decisions.

[0248] The maintenance decision optimization space is an abstract conceptual space that encompasses all possible combinations of maintenance paths, maintenance task allocation, and maintenance time windows, as well as their related parameters. For example, for a complex automated production equipment, there may be multiple choices for maintenance paths, each containing different operation sequences and operation methods, corresponding to different maintenance task allocation schemes (such as different maintenance personnel undertaking different tasks) and maintenance time windows (different time periods for maintenance have different impacts on production). All these possible combinations constitute the maintenance decision optimization space.

[0249] In an embodiment, first, the maintenance path, maintenance task allocation, and maintenance time window in the optimal maintenance strategy are analyzed in detail. Taking the maintenance path as an example, it is decomposed into multiple operation steps, and the operation time, required tools, personnel skill requirements, and other parameters of each step are determined. For maintenance task allocation, the number of personnel required, skill level, work duration, and other parameters of each task are specified. For the maintenance time window, its adjustable range, impact coefficient on production, and other parameters are determined. Then, a mathematical modeling tool (such as a linear programming model or a nonlinear programming model, selected according to actual circumstances) is used to construct the maintenance decision optimization space. These parameters are used as variables, the objective function is set to optimize these parameters comprehensively (such as minimizing total cost, maximizing equipment availability, etc.), and the constraints include actual equipment operation requirements, personnel resource limitations, and other constraints. In this way, a maintenance decision optimization space containing all possible decision schemes is constructed. This approach can comprehensively consider various factors related to maintenance decisions, providing a complete framework for subsequent optimization constraints, allowing decisions to be optimized within a reasonable range, and avoiding omission of important factors.

[0250] Step S602: According to the maintenance decision optimization space, combined with the preset maintenance cost threshold and the preset maintenance efficiency threshold, a first maintenance decision combination is obtained.

[0251] The maintenance cost threshold is a pre-set numerical limit used to limit the range of maintenance costs. For example, for a specific industrial equipment, if the enterprise determines that the cost of each maintenance cannot exceed a certain amount based on past maintenance experience and budget planning, this amount is the maintenance cost threshold. It can be determined according to the value of the equipment, the financial situation of the enterprise, the price level of the maintenance market, and other factors.

[0252] The maintenance efficiency threshold refers to a pre-set minimum standard for measuring maintenance efficiency. Maintenance efficiency can be measured in various ways, such as maintenance time, equipment downtime ratio to normal operation time, etc. For example, if the enterprise requires that the maintenance time after equipment failure cannot exceed a certain length of time, the maintenance efficiency indicator corresponding to this length of time is the maintenance efficiency threshold.

[0253] The coordination constraint is the process of using the maintenance cost threshold and the maintenance efficiency threshold to limit various possible decision-making schemes in the maintenance decision optimization space. It indicates that while considering that the maintenance cost cannot exceed the set threshold, the maintenance efficiency must also reach or exceed the pre-set threshold. This constraint method does not consider cost or efficiency in isolation, but combines both to ensure that the maintenance decision obtained is both economical and efficient. For example, when evaluating a maintenance scheme, it cannot be chosen simply because its maintenance cost is low but the maintenance time is too long (below the maintenance efficiency threshold), nor can it be accepted simply because the maintenance speed is fast but the cost is too high (exceeding the maintenance cost threshold). It must find a decision-making scheme that meets the requirements in both cost and efficiency.

[0254] In an embodiment, in the maintenance decision optimization space, the maintenance cost and maintenance efficiency of each possible decision-making scheme (i.e. the combination of maintenance path, maintenance task allocation and maintenance time window) can be calculated. The calculation of maintenance cost can be obtained by accumulating the material cost, labor cost and indirect cost caused by equipment downtime of each operation step. Maintenance efficiency can be evaluated comprehensively according to equipment downtime time, maintenance time and performance improvement after the equipment returns to normal operation. Then, compare the calculated maintenance cost and maintenance efficiency with the pre-set maintenance cost threshold and maintenance efficiency threshold. Filter out those decision-making schemes whose maintenance cost does not exceed the cost threshold and whose maintenance efficiency reaches or exceeds the efficiency threshold. The set of these schemes constitutes the first maintenance decision combination. The technical effect of this technical implementation is that by setting clear cost and efficiency limits, it can quickly filter out maintenance decision combinations that initially meet the requirements of the enterprise, reducing the workload of subsequent optimization calculations, while ensuring that the maintenance decision meets the basic requirements of the enterprise in terms of cost and efficiency.

[0255] Step S603: Initialize a multi-objective optimization algorithm according to the first maintenance decision combination; perform maintenance decision optimization iteration according to the multi-objective optimization algorithm to obtain a second maintenance decision combination, and record the number of iterations.

[0256] A multi-objective optimization algorithm is an algorithm specially designed for handling the simultaneous optimization of multiple objective functions. In industrial equipment maintenance decision-making, since multiple objectives such as maintenance cost, maintenance efficiency, and equipment reliability need to be considered simultaneously, a multi-objective optimization algorithm can balance and optimize among these objectives. For example, a common multi-objective evolutionary algorithm such as the NSGA-II algorithm simulates the biological evolution process to search for better decision-making schemes under the constraints of multiple objective functions.

[0257] Take the NSGA-II algorithm as an example for illustration.

[0258] First, initialize the algorithm according to the first maintenance decision combination. Each maintenance decision scheme in the first maintenance decision combination is regarded as an individual, and these individuals collectively constitute the initial population of the NSGA-II algorithm. For example, if there are 5 different maintenance decision schemes in the first maintenance decision combination, these 5 schemes are taken as 5 individuals in the initial population.

[0259] Next, define the fitness function. The fitness function is a mathematical expression used to evaluate the goodness of each individual (i.e., maintenance decision scheme). For the three objectives of maintenance cost, equipment reliability, and production efficiency, fitness functions are defined respectively. For the maintenance cost objective, the fitness function can be defined as the inverse of the maintenance cost, i.e., the lower the maintenance cost, the higher the fitness; for the equipment reliability objective, the fitness function can be defined according to the inverse of the equipment failure probability or the equipment failure-free operating time, i.e., the higher the equipment reliability, the higher the fitness; for the production efficiency objective, the fitness function can be defined according to the ratio of equipment operating time to total time or the production output per unit time, i.e., the higher the production efficiency, the higher the fitness.

[0260] Then, perform iterative calculations according to the basic flow of the NSGA-II algorithm. In each iteration process, new maintenance decision schemes (new individuals) are generated through selection operations (selecting better individuals according to their fitness), crossover operations (exchanging partial information of selected individuals to generate new individuals), and mutation operations (randomly changing some parameters of individuals to generate new individuals). The fitness function values of new individuals are calculated, and the population is updated according to non-dominated sorting and crowding calculation. After each iteration, the number of iterations is incremented by 1.

[0261] After multiple iterations, a new set of maintenance decision schemes is obtained, which constitutes a second maintenance decision combination.

[0262] This method can utilize the mature multi-objective optimization algorithm NSGA-II to effectively balance and search among multiple objectives. By defining the fitness function reasonably, the optimization direction of each objective is accurately reflected, and better maintenance decision schemes are continuously generated and screened in the iteration process, thereby improving the comprehensive performance of the decision and gradually achieving a better balance among multiple objectives such as maintenance cost, equipment reliability, and production efficiency.

[0263] Step S604: If the second maintenance decision combination is better than the first maintenance decision combination, the second maintenance decision combination is taken as the first maintenance decision combination; if not, the second maintenance decision combination is discarded.

[0264] The second maintenance decision combination being better than the first maintenance decision combination means that, under multi-objective comprehensive evaluation, the second maintenance decision combination performs better overall in terms of maintenance cost, maintenance efficiency, equipment reliability, etc. For example, if the second maintenance decision combination improves maintenance efficiency and equipment reliability without increasing maintenance cost, or significantly improves maintenance efficiency and equipment reliability with a slight increase in maintenance cost, then the second maintenance decision combination is considered better than the first maintenance decision combination.

[0265] In an embodiment, the values of maintenance cost, maintenance efficiency, equipment reliability, etc. of each decision scheme in the second maintenance decision combination and the first maintenance decision combination can be calculated again. Then, a multi-attribute decision method such as the TOPSIS method is used for comprehensive evaluation. The TOPSIS method calculates the distance of each decision scheme from the ideal solution (the optimal value of each objective) and the negative ideal solution (the worst value of each objective) to obtain a comprehensive evaluation index. The comprehensive evaluation indexes of each decision scheme in the second maintenance decision combination and the first maintenance decision combination are compared. If the comprehensive evaluation indexes of most decision schemes in the second maintenance decision combination are better than those of the corresponding decision schemes in the first maintenance decision combination, then the second maintenance decision combination is considered better than the first maintenance decision combination, and the second maintenance decision combination replaces the first maintenance decision combination; otherwise, the second maintenance decision combination is discarded. This method can accurately determine the superiority of two decision combinations through a scientific evaluation method, ensuring that the maintenance decision develops in a better direction.

[0266] Step S605: Continue the next iteration, and when the number of iterations is greater than or equal to the preset number of iterations, output the current first maintenance decision combination as the final maintenance decision.

[0267] The preset iteration number is a value preset before the algorithm starts, which is used to determine the stopping condition of the multi-objective optimization algorithm. The setting of this value can be determined according to factors such as the complexity of the problem, the limitation of the computing resources, and the requirement for the decision accuracy. For example, for a complex industrial equipment maintenance decision problem, if a more accurate result is expected, a larger preset iteration number can be set; if the problem is relatively simple or the computing resources are limited, a smaller preset iteration number can be set.

[0268] In an embodiment: In the multi-objective optimization algorithm (such as the NSGA-II algorithm), a counter can be set to record the iteration number. The counter is incremented by 1 each time an iteration is completed. After each iteration, it is checked whether the value of the counter is greater than or equal to the preset iteration number. If the preset iteration number has not been reached, the next iteration is continued, and the maintenance decision combination is updated according to the methods of steps S603 and S604; if the iteration number has reached or exceeded the preset iteration number, the iteration is stopped, and the current first maintenance decision combination is output as the final maintenance decision. The technical effect of this technical implementation is that by setting a reasonable preset iteration number, the computing cost can be controlled to avoid excessive calculation under the premise of ensuring the quality of the decision, and a final maintenance decision that achieves a good balance between multiple objectives is ultimately obtained.

[0269] Correspondingly, in order to better implement the above method, the embodiments of the present application also provide an intelligent maintenance decision system based on a knowledge graph. As shown in Figure 8 The intelligent maintenance decision system based on a knowledge graph 80 includes an acquisition module 801, a graph construction module 802, a relationship analysis module 803, an optimization module 804, a window adjustment module 805, and a decision module 806, as follows:

[0270] The acquisition module 801 is configured to acquire the equipment running state data and the historical maintenance data of the industrial equipment.

[0271] The atlas construction module 802 is configured to construct an industrial knowledge graph according to the equipment operation state data and the historical maintenance data, and generate a correlation relationship model of equipment operation state and maintenance behavior according to the industrial knowledge graph; the relationship analysis module 803 is configured to analyze the relationship between fault types and maintenance paths according to the correlation relationship model and the equipment operation state data, and obtain an optimal maintenance path; the optimization module 804 is configured to perform maintenance task allocation optimization according to the optimal maintenance path, the correlation relationship model and a preset maintenance resource constraint condition, and obtain an optimal maintenance task allocation scheme; the window adjustment module 805 is configured to perform dynamic adjustment analysis of a maintenance time window according to the optimal maintenance path, the optimal maintenance task allocation scheme and the correlation relationship model, and obtain an optimal maintenance strategy; and the decision module 806 is configured to perform multi-objective optimization iterative analysis according to the optimal maintenance strategy, and obtain a final maintenance decision of the industrial equipment.

[0272] In an embodiment, the atlas construction module 802 is configured to:

[0273] construct nodes and edges of an initial knowledge graph according to the equipment operation state data and the historical maintenance data, and obtain a preliminary knowledge graph structure;

[0274] perform relationship weight calculation between nodes according to the preliminary knowledge graph structure, and obtain a correlation strength distribution of the initial knowledge graph;

[0275] perform enhancement processing on the preliminary knowledge graph structure according to the correlation strength distribution, and obtain an industrial knowledge graph.

[0276] In an embodiment, the atlas construction module 802 is configured to: map to a low-dimensional space by using a graph embedding algorithm according to the correlation strength distribution and the preliminary knowledge graph structure, and obtain an enhanced knowledge graph representation; and optimize node grouping and edge weight by using a hierarchical clustering method according to the enhanced knowledge graph representation, and obtain a correlation relationship model of equipment operation state and maintenance behavior.

[0277] In an embodiment, the relationship analysis module 803 is configured to:

[0278] analyze the dynamic response relationship between fault type occurrence probability and maintenance path execution order according to the correlation relationship model and the equipment operation state data, and obtain an influence weight of the fault type on the maintenance path; and analyze a historical matching degree matrix of the fault type and the maintenance path according to the influence weight and the correlation relationship model, and obtain a path selection constraint condition;

[0279] obtain a candidate maintenance path set by obtaining a candidate maintenance path according to the path selection constraint condition;

[0280] determine an optimal maintenance path according to the candidate maintenance path set.

[0281] In an embodiment, the relationship analysis module 803 is configured to: calculate an execution efficiency index of each path according to the candidate maintenance path set, to obtain a path execution efficiency ranking; when the path execution efficiency ranking meets a preset efficiency threshold, determine the candidate maintenance path with the highest ranking as the optimal maintenance path; and when the path execution efficiency ranking does not meet the preset efficiency threshold, perform local search optimization on the candidate maintenance path set to obtain the optimal maintenance path.

[0282] In an embodiment, the optimization module 804 is configured to:

[0283] perform nonlinear relationship analysis of maintenance task allocation and maintenance resource utilization according to the correlation relationship model, to obtain a maintenance resource nonlinear relationship model; perform function decomposition on the maintenance resource nonlinear relationship model and the optimal maintenance path in combination with a preset maintenance resource constraint condition, to obtain a maintenance task allocation optimization function; perform iterative calculation on the maintenance task allocation optimization function by using a particle swarm optimization algorithm, to obtain a maintenance task allocation solution set that meets the maintenance resource constraint; analyze the balance of maintenance task allocation according to the maintenance task allocation solution set and the maintenance resource nonlinear relationship model, and determine a solution with the best balance in the maintenance task allocation solution set as an optimal maintenance task allocation scheme.

[0284] In an embodiment, the window adjustment module 805 is configured to:

[0285] perform dynamic coupling analysis of a maintenance time window and a device operating state according to the correlation relationship model, to obtain a device operating state change characteristic; perform maintenance time window fluctuation analysis by using a time series prediction algorithm according to the device operating state change characteristic, to obtain a maintenance time window adjustment interval; and perform priority ranking function construction of a maintenance plan according to the maintenance time window adjustment interval, to obtain a maintenance time window adjustment strategy.

[0286] perform maintenance plan overall optimization according to the maintenance time window adjustment strategy, the optimal maintenance path, and the optimal maintenance task allocation scheme, to obtain an optimal maintenance strategy.

[0287] In an embodiment, the window adjustment module 805 is configured to:

[0288] perform global optimization analysis by using a dynamic programming algorithm according to the maintenance time window adjustment strategy, to obtain a global optimization result; and perform maintenance plan overall optimization according to the global optimization result, the maintenance time window adjustment strategy, the optimal maintenance path, and the optimal maintenance task allocation scheme, to obtain an optimal maintenance plan and an optimal maintenance strategy.

[0289] In one embodiment, the decision module 806 is used for

[0290] Based on the optimal maintenance strategy, multi-objective parameter optimization is performed to construct a maintenance decision optimization space that includes maintenance paths, maintenance task allocation, and maintenance time windows. Based on the maintenance decision optimization space, a first maintenance decision combination is obtained by combining preset maintenance cost thresholds and preset maintenance efficiency thresholds with collaborative constraints. Based on the first maintenance decision combination, a multi-objective optimization algorithm is initialized. Based on the multi-objective optimization algorithm, maintenance decision optimization iterations are performed to obtain a second maintenance decision combination, and the iteration count is recorded. If the second maintenance decision combination is better than the first maintenance decision combination, it is adopted as the first maintenance decision combination; otherwise, it is discarded. The next iteration continues, and when the iteration count is greater than or equal to the preset iteration count, the current first maintenance decision combination is output as the final maintenance decision.

[0291] The implementation details of each module are provided in the preceding method embodiments and will not be repeated here. The technical effects achieved by each module and device are described in the foregoing method embodiments.

[0292] It should be noted that, in practical implementation, the above modules can be arbitrarily combined and integrated into one or more modules, or implemented as independent entities. Furthermore, the above modules can be implemented in hardware or as software functional modules. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The aforementioned storage medium can be a read-only memory, a hard disk, or an optical disk, etc.

[0293] like Figure 9 As shown, this application embodiment also provides a computer device 90, characterized in that it includes a processor 901 and a memory 902, wherein the memory 902 stores a computer program, and when the computer program is executed by the processor 901, the processor 901 performs the steps of any of the methods described above.

[0294] In several embodiments provided in the present application, it should be understood that the disclosed system and method can be implemented in other manners. For example, the division of the above-described apparatus embodiments is only a logical function division, and there can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, or between elements, can be indirect couplings or communication connections through some interfaces, devices, or units, and can be in electrical, mechanical, or other forms.

[0295] The above disclosures are merely preferred embodiments of the present application, and of course cannot limit the scope of the present application. Therefore, any equivalent changes made according to the claims of the present application shall still fall within the scope of the present application.

Claims

1. A knowledge graph-based intelligent maintenance decision method, characterized in that, The method comprises the following steps: obtaining equipment running state data and historical maintenance data of an industrial equipment; constructing nodes and edges of an initial knowledge graph according to the equipment running state data and the historical maintenance data, and obtaining a preliminary knowledge graph structure; calculating the relationship weight between nodes according to the preliminary knowledge graph structure, and obtaining the correlation strength distribution of the initial knowledge graph; mapping to a low-dimensional space by using a graph embedding algorithm according to the correlation strength distribution and the preliminary knowledge graph structure, and obtaining an enhanced knowledge graph representation; optimizing node grouping and edge weight by using a hierarchical clustering method according to the enhanced knowledge graph representation, and obtaining a correlation relationship model of equipment running state and maintenance behavior; analyzing the relationship between fault types and maintenance paths according to the correlation relationship model and the equipment running state data, and obtaining an optimal maintenance path; performing maintenance task allocation optimization according to the optimal maintenance path, the correlation relationship model and a preset maintenance resource constraint condition, and obtaining an optimal maintenance task allocation scheme; performing dynamic adjustment analysis of a maintenance time window according to the optimal maintenance path, the optimal maintenance task allocation scheme and the correlation relationship model, and obtaining an optimal maintenance strategy; performing multi-objective optimization iteration analysis according to the optimal maintenance strategy, and obtaining a final maintenance decision of the industrial equipment.

2. The knowledge graph based intelligent maintenance decision making method according to claim 1, characterized in that, The method comprises the following steps: analyzing the dynamic response relationship between fault type occurrence probability and maintenance path execution order according to the correlation relationship model and the equipment running state data, and obtaining an influence weight of fault type on maintenance path; analyzing the historical matching degree matrix of fault type and maintenance path according to the influence weight and the correlation relationship model, and obtaining a path selection constraint condition; obtaining a candidate maintenance path set according to the path selection constraint condition; determining an optimal maintenance path according to the candidate maintenance path set. 3.The knowledge graph based intelligent maintenance decision method according to claim 2, characterized in that, The method comprises the following steps: calculating the execution efficiency index of each path according to the candidate maintenance path set, and obtaining a path execution efficiency ranking; when the path execution efficiency ranking meets a preset efficiency threshold, determining the candidate maintenance path with the highest ranking as the optimal maintenance path; when the path execution efficiency ranking does not meet the preset efficiency threshold, performing local search optimization on the candidate maintenance path set to obtain the optimal maintenance path. 4.The knowledge graph based intelligent maintenance decision method according to claim 1, characterized in that, The method comprises the following steps: performing nonlinear relationship analysis of maintenance task allocation and maintenance resource utilization according to the correlation relationship model, and obtaining a maintenance resource nonlinear relationship model; performing function decomposition on the maintenance resource nonlinear relationship model and the optimal maintenance path in combination with a preset maintenance resource constraint condition, and obtaining a maintenance task allocation optimization function. According to the maintenance task allocation optimization function, a particle swarm optimization algorithm is used for iterative calculation to obtain a maintenance task allocation solution set satisfying maintenance resource constraints; According to the maintenance task allocation solution set and the maintenance resource nonlinear relationship model, the balance of the maintenance task allocation is analyzed, and a solution with the best balance in the maintenance task allocation solution set is determined as an optimal maintenance task allocation scheme. 5.The knowledge graph based intelligent maintenance decision method according to claim 1, characterized in that, According to the optimal maintenance path, the optimal maintenance task allocation scheme and the correlation relationship model, dynamic adjustment analysis of the maintenance time window is performed to obtain an optimal maintenance strategy, including: According to the correlation relationship model, dynamic coupling analysis of the maintenance time window and the equipment operating state is performed to obtain equipment operating state change characteristics; According to the equipment operating state change characteristics, a time series prediction algorithm is used for maintenance time window fluctuation analysis to obtain a maintenance time window adjustment interval; According to the maintenance time window adjustment interval, a priority sorting function of the maintenance plan is constructed to obtain a maintenance time window adjustment strategy; According to the maintenance time window adjustment strategy, the optimal maintenance path and the optimal maintenance task allocation scheme, the maintenance plan is overall optimized to obtain an optimal maintenance strategy.

6. The knowledge graph based intelligent maintenance decision making method of claim 5, wherein, According to the maintenance time window adjustment strategy, the optimal maintenance path and the optimal maintenance task allocation scheme, the maintenance plan is overall optimized to obtain an optimal maintenance strategy, including: According to the maintenance time window adjustment strategy, a dynamic programming algorithm is used for global optimization analysis to obtain a global optimization result; According to the global optimization result, the maintenance time window adjustment strategy, the optimal maintenance path and the optimal maintenance task allocation scheme, the maintenance plan is overall optimized to obtain an optimal maintenance plan and an optimal maintenance strategy.

7. The knowledge graph based intelligent maintenance decision making method of claim 1, wherein, According to the optimal maintenance strategy, multi-objective optimization iterative analysis is performed to obtain a final maintenance decision of the industrial equipment, including: According to the optimal maintenance strategy, multi-objective parameter optimization is performed to construct a maintenance decision optimization space including a maintenance path, a maintenance task allocation and a maintenance time window; According to the maintenance decision optimization space, a first maintenance decision combination is obtained by combining a preset maintenance cost threshold and a preset maintenance efficiency threshold for collaborative constraint; According to the first maintenance decision combination, a multi-objective optimization algorithm is initialized; According to the multi-objective optimization algorithm, maintenance decision optimization iteration is performed to obtain a second maintenance decision combination and record the number of iterations; If the second maintenance decision combination is better than the first maintenance decision combination, the second maintenance decision combination is taken as the first maintenance decision combination; if not, the second maintenance decision combination is discarded; The next iteration is continued, and when the number of iterations is greater than or equal to a preset number of iterations, the current first maintenance decision combination is output as a final maintenance decision.

8. An intelligent maintenance decision system based on a knowledge graph, characterized in that, The method comprises the following steps: An acquisition module is configured to acquire equipment operating state data and historical maintenance data of an industrial equipment; The atlas construction module is configured to construct nodes and edges of an initial knowledge graph according to the equipment operation state data and the historical maintenance data, to obtain a preliminary knowledge graph structure; perform inter-node relationship weight calculation according to the preliminary knowledge graph structure, to obtain an initial knowledge graph correlation strength distribution; map to a low-dimensional space by using a graph embedding algorithm according to the correlation strength distribution and the preliminary knowledge graph structure, to obtain an enhanced knowledge graph representation; and optimize node grouping and edge weight by using a hierarchical clustering method according to the enhanced knowledge graph representation, to obtain an equipment operation state and maintenance behavior correlation relationship model. The relationship analysis module is configured to analyze the relationship between a fault type and a maintenance path according to the correlation relationship model and the equipment operation state data, to obtain an optimal maintenance path. The optimization module is configured to perform maintenance task allocation optimization according to the optimal maintenance path, the correlation relationship model, and a preset maintenance resource constraint condition, to obtain an optimal maintenance task allocation scheme. The window adjustment module is configured to perform dynamic adjustment analysis of a maintenance time window according to the optimal maintenance path, the optimal maintenance task allocation scheme, and the correlation relationship model, to obtain an optimal maintenance strategy. The decision module is configured to perform multi-objective optimization iterative analysis according to the optimal maintenance strategy, to obtain a final maintenance decision of the industrial equipment.

Images