Machine fault-oriented flexible job shop scheduling plan adaptive repair method

Abstract

The application discloses a kind of flexible job shop scheduling plan adaptive repair method for machine failure, it is related to dispatch automation technical field, it is difficult to effectively handle the operation input of variable length in prior art, and complex structure correlation information in scheduling diagram cannot be fully mined, leading to the problem that generalization ability and repair precision under dynamic disturbance are not ideal.The scheduling repair process is modeled as Markov decision process in the application, and the evaluation logic of whether the affected operation is rescheduled is introduced.The scheme can selectively rearrange part of the unaffected operation by actively intervening in the prescheduling plan, thereby releasing a continuous time window and physical space for the key operation affected by the failure on the available machine.This repair mode, which trades local adjustment for global space expansion, fundamentally solves the problem of limited operation space under dynamic disturbance and significantly improves the optimization upper limit of the repair scheme.

Description

Technical Field

[0001] This invention relates to the field of scheduling automation technology, and in particular to an adaptive repair method for flexible job shop scheduling plans in response to machine failures. Background Technology

[0002] Traditional research and practice of Flexible Job Shop Scheduling (FJSP) are mostly based on the idealized assumption that the production environment remains stable over a long period. However, in actual shop floor production, such pre-defined scheduling schemes are often very fragile and easily affected by various unforeseen disturbances, such as sudden machine failures, random arrival of new tasks, or product rework requirements due to quality issues. To cope with these disturbances, existing research often adopts a strategy of complete rescheduling, but this approach is usually too aggressive. It focuses more on how to quickly respond to changes to restore production, but often ignores the inherent requirement of maintaining the overall stability of the production system. The direct consequence is that the established production rhythm in the shop floor is suddenly disrupted. Pre-arranged personnel assignments and supporting preparatory work such as raw material delivery organized according to the original plan may be significantly interrupted or even restarted. This not only easily leads to chaos in shop floor management, but also causes the idleness and waste of valuable production resources.

[0003] While existing reactive recovery strategies attempt to reduce deviations from the original plan, they mostly employ a passive response model, lacking a deep integration of stability objectives and real-time response. When dealing with highly flexible production systems with diverse operations, traditional feedforward neural networks struggle to effectively handle variable-length operational inputs and fail to fully exploit the complex structural relationships within the scheduling graph, resulting in suboptimal generalization ability and repair accuracy under dynamic perturbations.

[0004] Therefore, an adaptive repair method for flexible job shop scheduling plans in response to machine failures is needed. Summary of the Invention

[0005] To address the shortcomings of existing technologies in effectively handling variable-length operation inputs and fully leveraging the complex structural relationships in scheduling graphs, resulting in unsatisfactory generalization ability and repair accuracy under dynamic disturbances, this invention provides an adaptive repair method for flexible job shop scheduling plans oriented towards machine failures. By modeling scheduling repair as a Markov decision process, a repair strategy integrating graph neural networks and deep reinforcement learning is proposed. This strategy aims to achieve selective operation relocation and resource reallocation without substantially interfering with ongoing scheduling, thereby significantly improving the plan recovery efficiency under dynamic disturbance conditions while ensuring the stability of the production system. The specific technical solution is as follows: An adaptive repair method for flexible job shop scheduling plans in response to machine failures includes the following steps: S1: Obtain the scheduling job and machine raw data, and model the relationship between the job and the machine to construct a weighted parsing graph. S2: Combine real-time operating conditions to dynamically filter and update scheduling-related data, and output a new node feature vector that integrates operating condition disturbance characteristics. S3: Further extract the time-series dynamic features such as operation processing and machine idle time during the scheduling process, transform them into feature vectors of the same dimension through the time-series feature gating unit, and concatenate them with the node feature vectors. The feature extraction and evolution are completed through the graph neural network to generate an updated node feature vector that integrates the dual features of time-series dynamics and working condition disturbances. S4: A perturbation-aware hierarchical collaborative agent MDP decision framework is used. This framework consists of a three-layer collaborative structure of condition-aware agent, job agent and machine agent. During the decision-making process, the condition-aware agent dynamically adjusts the operation selection probability of the job agent and the machine allocation probability of the machine agent according to the condition level. Based on the state transition rules of the Markov decision process, dynamic matching of machine decisions is realized. At each time step, a comprehensive decision action set consisting of operation selection action and machine allocation action is output, and the corresponding workshop scheduling status is recorded.

[0006] Preferably, in step S1, the weighted disjunctive graph includes at least a set of nodes, a set of conjunctive edges, and a set of disjunctive edges.

[0007] Preferably, step S1 outputs node feature vectors, edge weight data, and basic scheduling data consisting of operation sets, machine sets, and operation-machine pair sets of the scheduling relationship graph. The node feature vectors include at least operation and machine node features, and the node feature vectors involve at least conjunction and disjunction edge weights.

[0008] Preferably, the graph neural network is a dual attention graph neural network that integrates temporal features.

[0009] Preferably, it also includes implementing a multi-dimensional cumulative reward priority strategy, which at least takes into account the balance between production efficiency and resource utilization.

[0010] Preferably, the multi-dimensional cumulative reward priority strategy is as follows: Based on calculation rules that include at least several weighted dimensions such as maximum completion time, machine load balancing, and no operation delay, the system generates a single-step reward signal. By accumulating the reward signals at each step and using reward signal threshold filtering rules to determine the validity of the decision, once an invalid decision action is identified, the job agent and the machine agent are driven to re-match and make a decision.

[0011] Preferably, the weighting coefficients in the plurality of weighting dimensions are dynamically adjustable weighting coefficients.

[0012] A flexible job shop scheduling plan adaptive repair system, applied to the method described above, includes: Initial state modeling and disjunctive graph construction module: acquires the original data of scheduled jobs and machines, models the relationship between jobs and machines, and constructs a disjunctive graph with edge weights; The state update module that integrates operating condition disturbance features: dynamically filters and updates scheduling-related data based on real-time operating conditions, and outputs a new node feature vector that integrates operating condition disturbance features. The deep feature extraction module that integrates temporal features further extracts temporal dynamic features such as operation processing and machine idle time during the scheduling process. It transforms these features into feature vectors of the same dimension through a temporal feature gating unit and concatenates them with node feature vectors. The feature extraction and evolution are completed through a graph neural network to generate an updated node feature vector that integrates both temporal dynamics and operating condition disturbances. Dynamic Action Selection Module: The perturbation-aware hierarchical collaborative agent MDP decision framework is used. This framework consists of a three-layer collaborative structure of condition-aware agent, job agent and machine agent. During the decision-making process, the condition-aware agent dynamically adjusts the operation selection probability of the job agent and the machine allocation probability of the machine agent according to the condition level. Based on the state transition rules of the Markov decision process, dynamic matching of machine decisions is realized. At each time step, a comprehensive decision action set consisting of operation selection action and machine allocation action is output, and the corresponding workshop scheduling status is recorded. Dynamic optimization module: Based on calculation rules that include multiple weighted dimensions such as maximum completion time, machine load balancing, and no operation delay, the system generates a single-step reward signal. By accumulating the reward signals of each step and using reward signal threshold filtering rules to judge the validity of the decision, once an invalid decision action is identified, the system will drive the job agent and machine agent to re-match and make decisions.

[0013] A computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the flexible job shop scheduling plan adaptive repair method as described above.

[0014] A processor for running a program, wherein the program executes the flexible job shop scheduling adaptive repair method as described above.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention demonstrates significant improvements in scheduling recovery flexibility and global optimization capabilities under dynamic operating conditions. Unlike traditional recovery strategies that only passively adjust affected processes, this invention models the scheduling repair process as a Markov decision process, endowing the system with the ability to autonomously decide whether to reschedule operations not directly affected. Through this innovative strategy, the system can proactively release time and space resources on available machines for operations affected by faults, effectively expanding the operational space for scheduling optimization, thereby achieving efficient repair and optimization of pre-scheduled plans from a global perspective. Furthermore, in terms of efficient aggregation and representation of multi-dimensional heterogeneous features, this invention introduces a fully relaxed transport graph as input to a graph attention network (GAT), achieving deep aggregation of individual features of neighboring nodes, accurately capturing the complex spatial relationships between processes and machines. Combined with operation selection and machine selection networks designed based on cross-attention mechanisms and pointer networks, the system can quickly and accurately extract core state information in highly variable environments, providing high-quality data support and logical assurance for scheduling decisions in complex scenarios. In addition, this invention greatly reduces the execution deviation between the repair plan and the pre-plan. By optimizing for minimizing the sum of deviations in operation completion times and utilizing the PPO reinforcement learning algorithm to train agents to sequentially select the optimal candidate machine, this invention achieves end-to-end neural policy reasoning. This mechanism effectively avoids the pitfalls of traditional heuristic rules, which are prone to getting trapped in local optima. While ensuring that the repaired plan maintains high system stability, it maximizes the approximation of the initial preset plan's performance, achieving an optimal balance between production continuity and efficiency. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0017] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0020] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0021] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0022] In one embodiment of the present invention, such as Figure 1 As shown, this paper presents a flexible job shop scheduling method based on artificial neural networks, aiming to solve the scheduling decision-making challenges in dynamic production environments by deeply integrating condition awareness and reinforcement learning technologies. At the modeling level, this solution first performs a refined modeling of the relationship between operations and machines in the job shop scheduling, constructing a weighted disjunctive graph containing a set of nodes, a set of conjunctive edges, and a set of disjunctive edges. Through this model, the system can output the node feature vectors (including operation and machine node features), edge weight data (involving conjunctive and disjunctive edge weights), and basic scheduling data consisting of the operation set, machine set, and operation-machine pair set of the scheduling graph, laying a digital foundation for subsequent decision-making.

[0023] In the dynamic feature extraction stage, this solution utilizes a condition-aware agent to collect real-time workshop production conditions, accurately identifying various disturbance types such as machine malfunctions, emergency order insertions, and production material shortages, and outputting corresponding condition levels. After receiving initial modeling data, the system dynamically filters and updates scheduling-related data based on real-time conditions, outputting a novel node feature vector that integrates condition disturbance features. Subsequently, the system further extracts temporal dynamic features such as operational processing and machine idle time during the scheduling process, transforming them into feature vectors of the same dimension through a temporal feature gating unit, and concatenating them with the node feature vectors. By employing a dual-attention graph neural network that integrates temporal features, this solution can achieve deep feature extraction and evolution, generating updated node feature vectors that integrate both temporal dynamics and condition disturbance features, thereby accurately capturing the temporal correlation and adaptability relationships of workshop scheduling.

[0024] At the core decision-making and optimization level, this solution adopts a perturbation-aware hierarchical collaborative agent (MDP) decision-making framework. This framework consists of a three-layer collaborative structure comprising a condition-aware agent, a job agent, and a machine agent. During the decision-making process, the condition-aware agent dynamically adjusts the operation selection probability of the job agent and the machine allocation probability of the machine agent based on the condition level, achieving dynamic matching of machine decisions based on the state transition rules of a Markov decision process. At each time step, the system outputs a comprehensive set of decision actions consisting of operation selection actions and machine allocation actions, and records the corresponding shop floor scheduling status.

[0025] Finally, to ensure continuous optimization of the scheduling scheme, this solution implements a multi-dimensional cumulative reward optimization strategy. Based on calculation rules that include multiple weighted dimensions such as maximum completion time, machine load balancing, and zero operational delay, the system generates single-step reward signals, and the weighting coefficients of each reward can be dynamically adjusted according to production targets. By accumulating the reward signals at each step and using reward signal threshold filtering rules, the effectiveness of decisions is judged. Once an invalid decision action is identified, the system will drive the job agent and machine agent to re-match and make decisions. Through this adaptive update and feedback mechanism, this solution not only achieves millisecond-level response under dynamic operating conditions, significantly reducing decision bias, but also takes into account multiple balances of production efficiency and resource utilization, providing an efficient closed-loop scheduling solution for the intelligent upgrading of traditional manufacturing.

[0026] In one embodiment of the present invention, a flexible job shop scheduling plan adaptive repair system is provided, applied to the method described above, comprising: Initial state modeling and disjunctive graph construction module: acquires the original data of scheduled jobs and machines, models the relationship between jobs and machines, and constructs a disjunctive graph with edge weights; The state update module that integrates operating condition disturbance features: dynamically filters and updates scheduling-related data based on real-time operating conditions, and outputs a new node feature vector that integrates operating condition disturbance features. The deep feature extraction module that integrates temporal features further extracts temporal dynamic features such as operation processing and machine idle time during the scheduling process. It transforms these features into feature vectors of the same dimension through a temporal feature gating unit and concatenates them with node feature vectors. The feature extraction and evolution are completed through a graph neural network to generate an updated node feature vector that integrates both temporal dynamics and operating condition disturbances. Dynamic Action Selection Module: The perturbation-aware hierarchical collaborative agent MDP decision framework is used. This framework consists of a three-layer collaborative structure of condition-aware agent, job agent and machine agent. During the decision-making process, the condition-aware agent dynamically adjusts the operation selection probability of the job agent and the machine allocation probability of the machine agent according to the condition level. Based on the state transition rules of the Markov decision process, dynamic matching of machine decisions is realized. At each time step, a comprehensive decision action set consisting of operation selection action and machine allocation action is output, and the corresponding workshop scheduling status is recorded. Dynamic optimization module: Based on calculation rules that include multiple weighted dimensions such as maximum completion time, machine load balancing, and no operation delay, the system generates a single-step reward signal. By accumulating the reward signals of each step and using reward signal threshold filtering rules to judge the validity of the decision, once an invalid decision action is identified, the system will drive the job agent and machine agent to re-match and make decisions.

[0027] In summary, this invention demonstrates significant improvements in scheduling recovery flexibility and global optimization capabilities under dynamic operating conditions. Unlike traditional recovery strategies that only passively adjust affected processes, this invention models the scheduling repair process as a Markov decision process, empowering the system to autonomously decide whether to reschedule operations not directly affected. Through this innovative strategy, the system can proactively release time and space resources on available machines for operations affected by faults, effectively expanding the operational space for scheduling optimization, thereby achieving efficient repair and optimization of pre-scheduled plans from a global perspective. Furthermore, in terms of efficient aggregation and representation of multi-dimensional heterogeneous features, this invention introduces a fully relaxed transport graph as input to a graph attention network (GAT), achieving deep aggregation of individual features of neighboring nodes, accurately capturing the complex spatial relationships between processes and machines. Combined with operation selection and machine selection networks designed based on cross-attention mechanisms and pointer networks, the system can quickly and accurately extract core state information in highly variable environments, providing high-quality data support and logical assurance for scheduling decisions in complex scenarios. In addition, this invention significantly reduces the execution deviation between the repair plan and the pre-plan. By optimizing for minimizing the sum of deviations in operation completion times and utilizing the PPO reinforcement learning algorithm to train agents to sequentially select the optimal candidate machine, this invention achieves end-to-end neural policy reasoning. This mechanism effectively avoids the pitfalls of traditional heuristic rules, which are prone to getting trapped in local optima. While ensuring that the repaired plan maintains high system stability, it maximizes the approximation of the initial preset plan's performance, achieving an optimal balance between production continuity and efficiency.

[0028] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0029] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0030] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0031] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0032] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. An adaptive repair method for flexible job shop scheduling plans oriented towards machine failure, characterized in that, Includes the following steps: S1: Obtain the raw data of the scheduled jobs and machines, and model the relationship between jobs and machines to construct a weighted parsing graph. S2: Combine real-time operating conditions to dynamically filter and update scheduling-related data, and output a new node feature vector that integrates operating condition disturbance characteristics. S3: Further extract the time-series dynamic features such as operation processing and machine idle time during the scheduling process, transform them into feature vectors of the same dimension through the time-series feature gating unit, and concatenate them with the node feature vectors. The feature extraction and evolution are completed through the graph neural network to generate an updated node feature vector that integrates the dual features of time-series dynamics and working condition disturbances. S4: A perturbation-aware hierarchical collaborative agent MDP decision-making framework is used. This framework consists of a three-layer collaborative structure of condition-aware agent, job agent and machine agent. During the decision-making process, the condition-aware agent dynamically adjusts the operation selection probability of the job agent and the machine allocation probability of the machine agent according to the condition level. Based on the state transition rules of the Markov decision process, dynamic matching of machine decisions is realized. At each time step, a comprehensive decision action set consisting of operation selection action and machine allocation action is output, and the corresponding workshop scheduling status is recorded.

2. The adaptive repair method for flexible job shop scheduling plans oriented towards machine failures according to claim 1, characterized in that, In step S1, the weighted disjunctive graph includes at least a set of nodes, a set of conjunctive edges, and a set of disjunctive edges.

3. The adaptive repair method for flexible job shop scheduling plans oriented towards machine failures according to claim 2, characterized in that, Step S1 outputs the node feature vectors, edge weight data, and basic scheduling data consisting of operation sets, machine sets, and operation-machine pair sets of the scheduling relationship graph. The node feature vectors include at least operation and machine node features, and the node feature vectors involve at least conjunction and disjunction edge weights.

4. The adaptive repair method for flexible job shop scheduling plans oriented towards machine failures according to claim 1, characterized in that, The graph neural network is specifically a dual-attention graph neural network that integrates temporal features.

5. The adaptive repair method for flexible job shop scheduling plans oriented towards machine failures according to claim 1, characterized in that, It also includes implementing a multi-dimensional cumulative reward priority strategy, which at least balances production efficiency and resource utilization.

6. The adaptive repair method for flexible job shop scheduling plans oriented towards machine failures according to claim 5, characterized in that, The multi-dimensional cumulative reward priority strategy is as follows: Based on calculation rules that include at least several weighted dimensions such as maximum completion time, machine load balancing, and no operation delay, the system generates a single-step reward signal. By accumulating the reward signals at each step and using reward signal threshold filtering rules to determine the validity of the decision, once an invalid decision action is identified, the job agent and the machine agent are driven to re-match and make a decision.

7. The adaptive repair method for flexible job shop scheduling plans oriented towards machine failures according to claim 1, characterized in that, The weighting coefficients in the aforementioned weighting dimensions are dynamically adjustable.

8. A flexible job shop scheduling plan adaptive repair system, characterized in that, The method applied to any one of claims 1 to 7 includes: Initial state modeling and disjunctive graph construction module: acquires the original data of scheduled jobs and machines, models the relationship between jobs and machines, and constructs a disjunctive graph with edge weights; The state update module that integrates operating condition disturbance features: dynamically filters and updates scheduling-related data based on real-time operating conditions, and outputs a new node feature vector that integrates operating condition disturbance features. The deep feature extraction module that integrates temporal features further extracts temporal dynamic features such as operation processing and machine idle time during the scheduling process. It transforms these features into feature vectors of the same dimension through a temporal feature gating unit and concatenates them with node feature vectors. The feature extraction and evolution are completed through a graph neural network to generate an updated node feature vector that integrates both temporal dynamics and operating condition disturbances. Dynamic Action Selection Module: The perturbation-aware hierarchical collaborative agent MDP decision framework is used. This framework consists of a three-layer collaborative structure of condition-aware agent, job agent and machine agent. During the decision-making process, the condition-aware agent dynamically adjusts the operation selection probability of the job agent and the machine allocation probability of the machine agent according to the condition level. Based on the state transition rules of the Markov decision process, dynamic matching of machine decisions is realized. At each time step, a comprehensive decision action set consisting of operation selection action and machine allocation action is output, and the corresponding workshop scheduling status is recorded. Dynamic optimization module: Based on calculation rules that include multiple weighted dimensions such as maximum completion time, machine load balancing, and no operation delay, the system generates a single-step reward signal. By accumulating the reward signals of each step and using reward signal threshold filtering rules to judge the validity of the decision, once an invalid decision action is identified, the system will drive the job agent and machine agent to re-match and make decisions.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform the adaptive repair method for flexible job shop scheduling as described in any one of claims 1 to 7.

10. A processor, characterized in that, The processor is used to run a program, wherein the program executes the adaptive repair method for flexible job shop scheduling as described in any one of claims 1 to 7.

Images