Large model enabled workshop scheduling end-to-end self-decision engine, method and device

Abstract

The invention belongs to the field of intelligent workshop scheduling, and particularly discloses a large model enabled workshop scheduling end-to-end self-decision engine, method and device. According to the method, the large model is positioned as a core decision maker instead of a traditional auxiliary tool, so that the powerful capabilities of the large model in the aspects of natural language understanding, context learning, complex logical reasoning and content generation are fully utilized; according to workshop real-time state information, a dynamically changing workpiece pool, a scheduling target input by a user through a natural language and emergency description, an autonomous and end-to-end scheduling decision can be directly carried out, so that the rapid response capability and intelligent processing level of a scheduling system to dynamic events are improved, and the scheduling efficiency is improved. And the application threshold is reduced through natural language interaction, the dependence on explicit programming and rule customization is reduced, a low-code development paradigm is developed, and the method has a far-sighted decision potential exceeding a traditional heuristic rule.

Description

Technical Field

[0001] This application belongs to the field of intelligent manufacturing and industrial automation technology, and relates to the field of intelligent workshop scheduling technology. More specifically, it relates to a large-model-enabled end-to-end self-decision engine, method and device for workshop scheduling. Background Technology

[0002] Shop floor scheduling refers to the process of rationally arranging the time and allocating resources for production activities within a shop floor under limited resource conditions, based on production task requirements, in order to optimize production efficiency, reduce costs, shorten delivery times, and improve equipment utilization. It is a complex multi-objective optimization problem that requires consideration not only of the workpiece processing sequence but also of factors such as machine availability, personnel skill levels, and material supply. Shop floor scheduling is a core function of manufacturing execution systems, and its efficiency and quality directly impact a company's productivity, cost control, and market competitiveness. However, achieving efficient and intelligent scheduling in the face of complex and dynamic production environments still presents many challenges.

[0003] With the rise of large language models, the industry has begun to explore their application in the field of workshop scheduling. However, some workshop scheduling applications based on large models are still in the "indirect assistance" mode. Patent CN118468931A proposes a framework for solving optimization problems based on large models. Its essence is to use large models to "translate" natural language requirements into mathematical models or solver code, but the core of decision-making still relies on external, traditional optimization solvers. In this mode, the large model plays the role of a "translator" or "modeling assistant," rather than the subject of the decision itself. Its fundamental flaw is that the decision logic is not internalized into the large model, and the entire process is still a separation of "modeling-solving." When facing dynamic events, it is still necessary to remodel and solve time-consuming problems, which limits the response speed and the level of intelligence.

[0004] While the aforementioned "indirect assistance" model lowers the barrier to modeling to some extent, it suffers from fundamental flaws: the decision-making logic is not internalized into the large model, and the entire process remains a traditional paradigm of separating "modeling-solving." This leads to several insurmountable pain points for existing technologies when dealing with real, complex production environments: First, severely insufficient dynamic adaptability: Real workshop environments are full of uncertainties, such as sudden equipment failures, urgent order insertions, and material supply delays. Under the "modeling-solving" separation model, whenever a dynamic event occurs, the system must rebuild the mathematical model and initiate a time-consuming solution process, resulting in slow response, difficulty in adjustment, and difficulty in ensuring the continuity and stability of production plans. Second, the coupling of decisions is fragmented: Multiple decision variables in workshop scheduling, such as workpiece selection, machine assignment, and process sequencing, are actually mutually coupled and influential. Traditional methods often decouple these variables to simplify the problem, which destroys the overall integrity of the problem and makes it difficult to achieve true global optimization. Third, a lack of global perspective and foresight: Most methods make decisions based on local information or short-term goals, easily leading to suboptimal overall scheduling performance and lacking consideration for the long-term impact on the production process. In summary, existing large-model-based scheduling methods, unable to break free from the constraints of traditional paradigms, only optimize a certain stage of the process without fundamentally solving the problem. They have significant shortcomings in dynamic adaptability, coupled decision processing, and decision foresight, urgently requiring a novel technical solution that allows large models to directly participate in and lead decision-making to overcome these bottlenecks. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of this application is to provide a large model-enabled end-to-end self-decision engine, method and device for workshop scheduling, which aims to solve the problems of insufficient adaptability, high human-computer interaction threshold and difficulty in handling coupled decision-making in the face of dynamic production environment.

[0006] To achieve the above objectives, in a first aspect, this application provides a large-model-enabled end-to-end self-decision-making engine for shop floor scheduling, the engine comprising: The status awareness module is equipped with standard interfaces with the shop floor manufacturing execution system, equipment control system and sensor network to continuously acquire and update the global status information of the shop floor in real time. The workpiece pool management module, which is linked with the status perception module, is used to add or remove processes that meet the conditions into the workpiece pool in real time according to the process release rules, thereby dynamically constructing and maintaining a workpiece pool containing all processes that have been released and are waiting to be scheduled. The human-computer interaction module provides a user interface that supports natural language input and graphical result display. It is used to receive unstructured scheduling goals or dynamic event reports from users and to intuitively display the scheduling results. The large model decision engine module has a built-in large model that is used to obtain information from other modules when a scheduling decision is triggered. It integrates the current global status of the workshop, the process information in the workpiece pool, the scheduling target input by the user through natural language, and the dynamic event description to perform end-to-end autonomous reasoning and directly output scheduling instructions that include the coupling relationship between workpiece selection and machine assignment. The knowledge base module stores domain knowledge, hard constraint rules, and historical best practices related to workshop scheduling, and provides real-time support for the decision-making process of the large model decision engine module through retrieval-enhanced generation technology. The scheduling execution and feedback module is used to issue the scheduling instructions to the physical execution layer of the workshop, continuously monitor the execution status of the instructions, and feed back the actual completion status to the status perception module, thereby forming a closed-loop control loop. At the same time, the real-time information after scheduling execution will also drive the workpiece pool management module to update, forming an information closed loop to maintain the effectiveness of the decision candidate set.

[0007] Preferably, the global status information of the workshop includes at least: the current status of each machine, the current processing progress of each workpiece, and the availability status of related production resources.

[0008] Preferably, each process within the workpiece pool carries its detailed attributes, including the workpiece identifier, priority, standard processing time, list of available machines, and waiting time.

[0009] To achieve the above objectives, in a second aspect, this application provides a large model-enabled end-to-end self-decision-making method for shop floor scheduling, applied to the large model-enabled end-to-end self-decision-making engine for shop floor scheduling as described in the first aspect, the method comprising: S1. By connecting to the manufacturing execution system or sensor network, the global status information of the workshop can be sensed and collected in real time; S2. Based on global state information including the current processing progress of the workpiece, dynamically construct and maintain a workpiece pool containing all currently released processes that are waiting to be scheduled. S3. When the preset scheduling decision trigger event is met, input comprehensive information into a large model based on pre-training and fine-tuned by knowledge of the workshop scheduling domain; S4. Based on the inherent reasoning ability of the large model, perform end-to-end analysis and weighing of the comprehensive information, and directly generate the next executable scheduling decision instruction.

[0010] Preferably, at the initial moment of the scheduling process, the first process of all workpieces, or those processes without preceding processes or whose preceding conditions have been met, will be automatically placed into the workpiece pool.

[0011] Preferably, the dynamic maintenance of the workpiece pool in S2 specifically includes: when a certain process of a workpiece is completed, if the workpiece has a subsequent process and its start-up conditions are met, the subsequent process is automatically added to the workpiece pool; when a new workpiece or an urgent order is issued and confirmed, its first process or the process that meets the start-up conditions is added to the workpiece pool.

[0012] Preferably, the scheduling decision triggering event in S3 includes at least one of the following: the completion of the processing of at least one workpiece, thereby releasing the processing machine; a new high-priority process entering the workpiece pool; a change in the status of the workshop machine; or a dynamic interference event.

[0013] Preferably, the fine-tuning and decision-making process of the large model after pre-training includes: (1) Supervised fine-tuning of the large model using historical or simulation data containing workshop state-decision pairs; (2) Using expert preference feedback or a reward function based on scheduling performance indicators, perform direct preference optimization and / or policy optimization based on reinforcement learning on the large model; (3) In the process of reasoning and judgment, combined with the retrieval enhancement generation technology, the scheduling rules or historical cases are dynamically retrieved from the external knowledge base to assist decision-making.

[0014] Preferably, the method further includes: S5. The scheduling instruction is sent to the physical execution layer of the workshop, and the execution status of the instruction is continuously monitored. The actual completion status is fed back to the status perception module, thereby forming a closed-loop control loop. At the same time, the real-time information after scheduling execution will also drive the workpiece pool management module to update, forming an information closed loop to maintain the effectiveness of the decision candidate set.

[0015] To achieve the above objectives, in a third aspect, this application provides a large-model-enabled end-to-end self-decision-making device for workshop scheduling, the device being equipped with a large-model-enabled end-to-end self-decision-making engine for workshop scheduling as described in the first aspect.

[0016] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art: This application proposes a large-scale model-enabled end-to-end self-decision engine, method, and apparatus for shop floor scheduling, the core of which lies in changing the traditional scheduling paradigm of separating "modeling and solving". Specifically, it includes: real-time perception of shop floor status information and dynamic construction and maintenance of a "workpiece pool" containing all processes to be processed; when preset conditions such as process completion or dynamic event occurrence are triggered, a large-scale model finely tuned with knowledge of the shop floor scheduling domain is invoked; the large-scale model can integrate the current global state of the shop floor, process information in the workpiece pool, scheduling objectives input by the user through natural language, and dynamic event descriptions to perform end-to-end reasoning and judgment, and directly output executable scheduling instructions that simultaneously consider mutually coupled decision variables such as workpiece selection and machine assignment. This application positions the large model as the core decision-maker, rather than a traditional auxiliary tool. It fully leverages the large model's powerful capabilities in natural language understanding, context learning, complex logical reasoning, and content generation, enabling it to make autonomous, end-to-end scheduling decisions based on real-time workshop status information, a dynamically changing "workpiece pool," and user-inputted scheduling goals and descriptions of unexpected events via natural language. This improves the scheduling system's ability to respond quickly to dynamic events and enhances its intelligent processing capabilities. Furthermore, it lowers the application threshold through natural language interaction, reduces reliance on explicit programming and rule customization, moves towards a low-code development paradigm, and possesses the potential for far-sighted decision-making that surpasses traditional heuristic rules. Attached Figure Description

[0017] Figure 1 This is a flowchart of a large-model-enabled end-to-end self-decision-making method for workshop scheduling provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram illustrating the dynamic construction and maintenance of the "workpiece pool" provided in the embodiments of this application.

[0019] Figure 3 This is a schematic diagram of the processing flow for handling dynamic events in a large model provided in the embodiments of this application.

[0020] Figure 4 This is a schematic diagram of a large-model-enabled end-to-end self-decision engine structure for workshop scheduling provided in an embodiment of this application.

[0021] Figure 5 This is a schematic diagram of the hardware structure of a large-model-enabled end-to-end self-decision-making device for workshop scheduling provided in an embodiment of this application.

[0022] Figure 6 This is an example of a Gantt chart showing the scheduling results under dynamic events provided in this application embodiment. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0024] In this application, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A existing alone, A and B existing simultaneously, and B existing alone. In this application, the symbol " / " indicates that the related objects are in an "or" relationship, for example, A / B means A or B.

[0025] In this application, the terms "first" and "second," etc., are used to distinguish different objects, not to describe a specific order of objects. For example, "first response message" and "second response message," etc., are used to distinguish different response messages, not to describe a specific order of response messages.

[0026] In this application, the term "electrical connection" can refer to a direct circuit connection or a signal transmission via a communication protocol.

[0027] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0028] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple processing units means two or more processing units, multiple elements means two or more elements, etc.

[0029] For ease of understanding, the English abbreviations and related technical terms used in the embodiments of this application will be explained and described below.

[0030] Workpiece: The core object of scheduling, representing the product or component that needs to be processed. Each workpiece has unique technological requirements and consists of multiple processes in a specific sequence. The number of processes, processing time, and processing machine requirements vary for different workpieces.

[0031] Machines are the equipment resources that perform processing tasks. A workshop is equipped with various types of machines, each with different processing capabilities and functions. A single machine can only process one workpiece and one operation at a time, and different workpieces may require different machines for processing.

[0032] A process is the basic unit in the machining of a workpiece. Each workpiece consists of multiple processes, which are performed sequentially in a certain order. A process has a defined processing time and the required processing machine, and there are strict sequential constraints between processes; that is, the next process can only begin after the previous process is completed.

[0033] Processing time refers to the time required for each process to be performed on a designated machine. Processing time is an important parameter, usually determined by a variety of factors such as process requirements, machine performance, and operator skills.

[0034] The embodiments of this application are described below with reference to the accompanying drawings.

[0035] like Figure 1 As shown, this application proposes a large-model-enabled end-to-end self-decision-making method for workshop scheduling, which includes the following main steps: Step 1: Workshop status initialization and real-time sensing.

[0036] When the scheduling system starts, it first needs to load the static and quasi-static basic data of the workshop. This data includes: (1) Workpiece information: Detailed information of each workpiece to be processed, such as workpiece ID, product model, process route i.e. process sequence, standard processing time of each process, set of optional processing machines for each process and its specific processing time on the machine, workpiece priority (normal, urgent), promised delivery date, etc.

[0037] (2) Machine information: Detailed information of all production machines in the workshop, such as machine ID, machine type, processing capacity, current status (idle, processing, faulty, planned maintenance), etc.

[0038] (3) Other resource information: Other key production resources besides machines (such as manpower, tools, and materials), for example, available workers and their skill levels, availability of specific tooling fixtures, status and location of material handling equipment such as automated guided vehicles (AGVs).

[0039] During system operation, it needs to interface with the underlying information systems in the workshop, such as Manufacturing Execution System (MES), Supervisory Control System (SCADA), or sensor networks, to continuously acquire dynamic information about the workshop in real time.

[0040] The global status information of the workshop includes at least: the current status of each machine, such as idle, processing, faulty, or scheduled maintenance; the current processing progress of each workpiece, such as completed processes and its location; and the availability status of related production resources. For example, unexpected machine failures and their estimated repair time, the actual completion time of a certain process of a workpiece, the arrival of new orders, especially urgent orders, changes in the supply status of raw materials, such as material shortages or replenishment, temporary worker absences or reassignments, etc. This dynamic information is an important basis for scheduling decisions.

[0041] Step 2: Dynamic construction and maintenance of the "workpiece pool".

[0042] The "workpiece pool" is a core concept in the method of this application. It represents a dynamically changing set containing all currently released operations that are waiting to be scheduled. The workpiece pool provides a clear, real-time decision candidate space for the large model.

[0043] Definition of a workpiece pool: Each element in the workpiece pool corresponds to a specific operation on a workpiece. The workpiece pool is a dynamic collection, where each element represents an operation in a "released" state (i.e., meeting all prerequisites) and awaiting scheduling. It is called a "workpiece pool" to emphasize that each element (operation) in the pool must be associated with the global information of its associated workpiece (such as workpiece priority and delivery date), requiring comprehensive consideration during decision-making. Only when an operation meets all its prerequisites, such as the completion of its preceding operation, the availability of required raw materials, and the preparation of necessary tooling and fixtures, will that operation be "released" and added to the workpiece pool, awaiting scheduling for processing.

[0044] The dynamic construction and maintenance process of the "workpiece pool" is as follows: Figure 2 As shown, it includes the following steps: Initialization: At the initial moment of the scheduling process, the first operation of all workpieces, or those operations without a preceding operation or whose preceding conditions have been met, will be automatically placed into the workpiece pool.

[0045] Dynamic Update Mechanism: The contents of the workpiece pool change dynamically with the actual progress of production in the workshop. When a workpiece's process is completed and confirmed on a machine, if there are subsequent processes for that workpiece, the system checks the release conditions for those subsequent processes. Once the conditions are met, the subsequent process is automatically added to the workpiece pool. When a new production order, especially an urgent order, is received and confirmed, its first process or the process that meets the start-up conditions is added to the workpiece pool. If a process in the workpiece pool is temporarily suspended for some reason, such as the corresponding workpiece being canceled or unable to be processed after a long wait, and is no longer suitable for scheduling, it should be removed from the workpiece pool or marked as inactive. This dynamic workpiece pool mechanism allows scheduling decisions to closely revolve around currently available tasks, rather than pre-scheduling all future tasks, thus better adapting to the dynamics and uncertainties of the workshop environment. This mechanism can be seen as a focused attention approach designed for large models, limiting the decision space to the most relevant and urgent tasks through pre-screening and dynamic updates, thereby improving the efficiency and accuracy of large model decisions.

[0046] Information on operations in the workpiece pool: To support effective decision-making in large models, each operation in the workpiece pool should carry sufficiently rich information, such as the ID and priority of the workpiece to which it belongs, the ID of the operation itself and its sequence number in the process route, the standard processing time of the operation, a list of available processing machines and the processing time on each machine, the waiting time in the workpiece pool, and the urgency of the delivery date.

[0047] Step 3: Triggering the scheduling decision.

[0048] The scheduling decision-making process in this application adopts an event-driven mechanism, rather than the traditional time-driven fixed-periodic rescheduling. This means that the scheduling decision is triggered by specific critical events occurring within the workshop.

[0049] The decision-triggered event types in this application include, but are not limited to: (1) Process completion event: When a certain process of a workpiece is completed on a certain machine, the machine is released and becomes an available resource. At this time, it is necessary to select the next suitable processing task for this machine that has just become idle, or if there are higher priority tasks in the workpiece pool that require other machines, it will also trigger a wider scheduling adjustment.

[0050] (2) New workpiece / order arrival event: When a new workpiece, especially the related process of an order with high priority or urgent delivery time requirements, is released into the workpiece pool, an immediate decision needs to be made to determine whether it is necessary to interrupt some current low priority tasks or adjust the existing plan to arrange the new task as soon as possible.

[0051] (3) Machine status change events: For example, a running machine suddenly malfunctions and stops, or a previously malfunctioning machine is repaired and restored to a usable state. These events will change the availability of resources in the workshop, thereby triggering scheduling decisions to respond to the changes.

[0052] (4) Other dynamic disturbances: For example, changes in the expected arrival time of key raw materials (earlier or later), temporary absence of workers on a shift leading to labor shortages for certain skills, and customers making temporary changes to order requirements (such as quantity or delivery date). These all require reassessment and adjustment of the current scheduling plan.

[0053] When any of the above-mentioned decision-triggered events occur, the system will invoke the large model to make the next scheduling decision. This event-driven architecture, combined with the large model's ability to quickly understand and respond to dynamic information, forms an agile and efficient scheduling system that can better adapt to the randomness and complexity of real-world workshop environments.

[0054] Step 4: End-to-end decision-making for large models.

[0055] The large model plays the role of the decision-maker directly in this application.

[0056] Construction and organization of input information: Before calling the large model, the system needs to collect and organize all relevant information required for the current decision point, and construct it into an input format of text prompts suitable for the large model to process.

[0057] This comprehensive information mainly includes: 1) Real-time workshop status information obtained in step one: including the current status of all relevant machines, which machines are idle and available, which are processing specific processes for specific workpieces, which are in a fault state and their estimated recovery time, the current load of machines, such as the cumulative processing time of assigned tasks, and the occupancy of machine buffers. 2) Workpiece pool information obtained in step two: a detailed list of all candidate processes in the current workpiece pool and their respective attributes, such as workpiece ID, process ID, priority, processing time, available machines, waiting time, and delivery urgency, as mentioned above. This information needs to be presented to the large model in a clear and structured manner. 3) Preset or user-specified scheduling goals and constraints: clearly informing the large model what the current scheduling goal is, for example, whether to prioritize minimizing the maximum completion time (Makespan) of all workpieces, or to prioritize meeting the delivery time of urgent orders (minimizing total delays or maximizing on-time delivery rate), or to balance the load of each machine to improve overall utilization, etc. These objectives can be pre-configured in the system or dynamically proposed by the user through natural language during interaction. For example, the user could say, "At this stage, please focus on the timely delivery of orders A and B, while trying to balance the equipment load of the CNC group." In addition, the hard constraints that the model must adhere to must be communicated, such as the sequence of processes and specific machine capability limitations. 4) Description of Sudden Dynamic Events: Dynamic event information is provided by workshop managers in unstructured text form through a natural language interface. The large model has the ability to directly parse this natural language description, understand its inherent intent and key information, and integrate it as a core constraint into the decision-making process. Different types of dynamic events (such as machine failures and emergency order insertions) carry different information and inherent constraints. The large model can adaptively adjust its decision logic and output scheduling instructions based on the specific event description input, thereby achieving differentiated and intelligent responses to different events. If the decision is made due to a dynamic event such as machine failure or an urgent order, the description of the event entered by the user through the natural language interface, such as: "M2 milling machine hydraulic system alarm, machine stopped, estimated repair time 2 hours, please adjust the affected production schedule." or "Customer called, order JOB005 must be shipped before 3 pm tomorrow, please prioritize it!" should also be entered as important contextual information into the large model.

[0058] To make the input format of this application clearer, a complete input prompt example is provided below. This prompt is organized in a structured text format, facilitating parsing and understanding by large models: ###Workshop Scheduling Decision-Making Tips### Part 1: Real-time Status of the Workshop Timestamp: 2023-10-27 14:00:00 equipment: M01: Status = Fault, Estimated Repair Completion Time = 2 hours, Affected Part = J008 M02: Status = Idle M03: Status = In Progress, Current Workpiece = J005, Remaining Time = 35 minutes M04: Status = Idle resource: AGV_01: Status = Available Tool_T05: Status = In Use, Belonging Device = M03 Part Two: Workpiece Pool (Currently Pending Scheduling Processes) Processing tasks: OP11 (from workpiece J01): Priority = High, Estimated processing time = 60 minutes, Available equipment = [M02, M04], Deadline = 2023-10-27 18:00 OP31 (from workpiece J03): Priority = Normal, Estimated processing time = 90 minutes, Available equipment = [M02, M04], Deadline = 2023-10-28 12:00 OP22 (from workpiece J02): Priority = Normal, Estimated processing time = 45 minutes, Available equipment = [M02], Deadline = 2023-10-28 10:00 Part Three: Scheduling Objectives and Constraints Target: Main objective: Minimize the maximum completion time (Makespan) Secondary objective: Ensure high-priority task J01 is completed on time. Constraints: Workpiece J01 must be processed using equipment M02 or M04. Part Four: Description of Dynamic Events Event: The spindle bearing of equipment M01 is overheating, triggering an emergency shutdown for maintenance. The estimated downtime is 2 hours. This event just occurred and triggered this scheduling decision.

[0059] ###End of prompt### Selection and fine-tuning of the large model: In order for the large model to be able to handle complex workshop scheduling decision-making tasks, it is necessary to select a suitable base model and perform special fine-tuning.

[0060] 1) Base Model Selection: Considering the feasibility of practical deployment, especially for enterprises that wish to operate in a localized environment, and for reasons of data security and response speed, some open-source or commercial large-scale models with relatively small parameter sizes, such as 3B, 7B, or even smaller, such as the 1.7B level, will be selected as the base model, while still maintaining excellent performance. The model selection also needs to consider its ability to understand and generate Chinese. The base model can include, but is not limited to, mainstream open-source large language models in the industry, such as the Llama series models, ChatGLM series models, Qwen series models, DeepSeek series models, etc. When selecting a base model, its ability to understand Chinese, inference performance, and running efficiency in the target deployment environment will also be considered.

[0061] 2) Fine-tuning strategies and techniques: The purpose of fine-tuning is to combine general language skills with professional knowledge in the field of shop floor scheduling, decision-making logic, and specific production goals.

[0062] Commonly used fine-tuning techniques include: Dataset Construction: High-quality "state-decision" sample pairs for shop floor scheduling tasks need to be constructed. These samples can be obtained through various means: (a) utilizing historical production scheduling data and actual execution results from the enterprise; (b) generating a large amount of simulated data by running different scheduling strategies, including expert-designed rules or high-performance optimization algorithms, in a high-fidelity shop floor simulation environment; (c) using a larger-scale, more powerful "teacher" model (such as online GPT-4 or DeepSeek-R1) to generate high-quality "ideal" decisions for a given shop floor state, serving as a model for the "student" small model to learn from, i.e., knowledge distillation. Each sample should contain a detailed description of the shop floor state, including job pool information, machine state, etc., and a corresponding optimal scheduling decision, such as choosing which job and which process to process on which machine. Supervised Fine-tuning: Supervised learning is performed on the base model using the constructed "state-decision" dataset. The goal is to enable the model to learn patterns that map from a given shop floor state to the desired scheduling decision. Direct Preference Optimization (DPO): After supervised fine-tuning, preference optimization techniques can be used to make the model's decisions more consistent with human expert judgment or specific performance metrics. This method constructs a preference dataset, indicating which of two candidate scheduling schemes is better for the same shop floor state, allowing the model to directly learn this preference. This enables the model to output decisions that better meet expectations without relying on an explicit reward function. Embedding Domain Knowledge: Embedding specific domain knowledge in shop floor scheduling can be achieved through two main approaches: first, during the model training and fine-tuning phases, explicitly including this knowledge in the fine-tuning data allows the model to learn inherent patterns; second, during the model's decision-making and reasoning phases, combining it with Retrieval Augmentation Generation (RAG) techniques. RAG allows the model to dynamically retrieve relevant information from an external, structured shop floor scheduling knowledge base as context during decision-making, assisting it in making more accurate and domain-compliant reasoning. Reinforcement for Dynamics: To improve the model's rapid response and decision stability in the face of sudden dynamic events, reinforcement learning-based policy optimization (GRPO) can also be considered. By designing appropriate reward mechanisms, the model is encouraged to make robust and efficient decisions in dynamically changing environments. Decision-making reasoning process: After the large model, fine-tuned with knowledge from the shop floor scheduling domain, receives complete input information, it integrates the current global state of the shop floor, process information in the workpiece pool, the scheduling objectives input by the user via natural language, and dynamic event descriptions. Utilizing the knowledge and reasoning abilities learned during the fine-tuning phase, it makes decisions and directly generates executable scheduling instructions for the next step. For example, "Schedule process OP2 of workpiece J001 on machine M03" or, in response to dynamic events, generate "Pause all processes affected by the M02 failure and prioritize scheduling process OP1 of the urgent workpiece J005 in the workpiece pool on the idle machine M04."The instructions at least simultaneously determine the process to be processed and the optimal processing machine assigned to it, thereby achieving coupled processing of multiple decision variables.

[0063] This process is implicit; the neural network inside the model comprehensively evaluates various factors. 1) Job Selection: The model evaluates all candidate operations from the current "job pool." Evaluation criteria include the job's pre-defined priority, the urgency of its delivery date, the contribution of the selected operation to the overall scheduling objective, whether it's on the critical path, and whether it can free up bottleneck machines for more important subsequent operations. The model implicitly learns and applies traditional rules such as Shortest Processing Time (SPT) and Earliest Delivery Date (EDD), or more complex combinational logic.

[0064] 2) Machine Selection: After determining the process to be processed or several candidate processes, the model needs to select a suitable processing machine. The evaluation criteria include the processing time of the process on the machine, the current load of the machine (selecting a less busy machine to achieve load balancing), whether the machine has the specific capabilities required to process the process, whether selecting the machine can reduce unnecessary workpiece transportation or waiting time, whether the machine is about to undergo planned maintenance or whether there are potential failure risks, etc.

[0065] 3) Potential of coupled decision-making: Unlike traditional step-by-step decision-making (selecting the workpiece first, then the machine), large models have the potential to consider the coupling effect of workpiece selection and machine selection in a single inference, and can even take into account more relevant factors such as worker skill matching and the timeliness of material supply, thereby making a relatively better collaborative decision in multiple dimensions.

[0066] Output scheduling instructions: After reasoning and judgment, the large model will generate clear and explicit scheduling instructions. For example, the output text might be: "Schedule operation OP2 of workpiece J001 on machine M03, with an estimated start time [specific time] and an estimated completion time [specific time]." Or, when dealing with dynamic events, it may output a more complex sequence of instructions, such as: "Due to a malfunction in machine M5, operation OP3 of workpiece J008, which was originally scheduled to be processed on M5, is suspended. Please reschedule operation OP3 of workpiece J008 to machine M1, starting immediately after its current task is completed. At the same time, please schedule operation OP1 of the urgent order JOB_NEW on the now-idle machine M4."

[0067] Step 5: Scheduling and Status Feedback.

[0068] While generating the large model, executable scheduling instructions that consider the coupled decision variables such as workpiece selection and machine assignment need to be accurately communicated and executed. These instructions are then sent to the Manufacturing Execution System (MES), which coordinates and monitors the workpiece's processing on the designated machine. In highly automated workshops, instructions may also be directly passed to the underlying equipment control system or robot controller.

[0069] The actual execution status of each process, such as the actual start time, actual completion time, and whether any abnormalities occurred during processing, needs to be fed back to the scheduling system in real time. This feedback information is used to update the real-time status model of the workshop, including machine status and workpiece progress, and also to update the contents of the "workpiece pool," providing an accurate basis for the next round of scheduling decisions and forming a closed-loop control.

[0070] Step Six: Handle dynamic events.

[0071] One of the core advantages of this application's method lies in its flexibility and intelligence in handling dynamic events. When any dynamic event occurs in the workshop that may affect the existing scheduling plan, whether it is automatically detected by sensors or... Figure 3 The example shown is input by workshop managers through a natural language interface. For instance, a user might input: "Notification: The tool magazine of CNC machining center No. 2 is malfunctioning and cannot automatically change tools temporarily. Only single-tool machining is possible. Please adjust the plans for complex processes that rely on automatic tool changing." The large model can directly understand this natural language description of the event and its potential impact. The model will combine the current overall workshop status, the workpiece pool situation, and information about this newly occurring dynamic event to conduct a complete decision-making process again. The output decision may be a partial adjustment to the existing plan or a more global plan reordering. This method of directly understanding and responding to dynamic events by the large model is undoubtedly more direct, faster, and intelligent compared to the traditional method that requires manual analysis of event impact, modification of model parameters or rules, and rerunning of complex solution algorithms. This fully leverages the potential of natural language as a universal interface to handle complex and unexpected situations, simplifying the complexity of human-computer interaction.

[0072] To achieve the above method, such as Figure 4 As shown, this application proposes a large-model-enabled end-to-end self-decision engine for workshop scheduling. The system consists of the following core functional modules: Data sensing and interface module: It is equipped with a standard interface to the shop floor manufacturing execution system (MES), equipment control system (SCADA) or sensor network, which is used to collect and aggregate production data from various data sources in the shop floor in real time, and perform preliminary cleaning, verification and transformation. Status Management Module: Responsible for maintaining a comprehensive, accurate, and real-time digital model of the current status of the workshop; Workpiece Pool Management Module: In conjunction with the Data Sensing and Interface Module, it is used to dynamically create, update and manage the "workpiece pool" based on the workpiece processing progress information provided by the Status Management Module and the preset process release rules. Large Model Decision Engine Module: This is the "brain" of the entire system, responsible for executing the core scheduling decision logic. When a scheduling decision is triggered, this module obtains comprehensive information from other modules, performs end-to-end autonomous reasoning, and directly outputs specific scheduling instructions that include the coupling relationship between workpiece selection and machine assignment. Knowledge Base Module: As a structured information repository, it stores domain knowledge, hard constraint rules, and historical best practices related to workshop scheduling. It also provides real-time support for the decision-making process of the large model decision engine module through Retrieval Enhanced Generation (RAG) technology. The large model decision engine can dynamically retrieve relevant information from the knowledge base when needed to enhance the accuracy and robustness of its decisions. Human-Computer Interaction Module: This module provides a user interface that supports natural language input and graphical result display. It receives unstructured scheduling objectives or dynamic event reports from users and displays the scheduling results intuitively in the form of Gantt charts. It also supports users to input scheduling instructions, query the real-time status of the workshop, and view the current scheduling plan through natural language. The scheduling execution and monitoring module is responsible for accurately transmitting the scheduling instructions generated by the large model decision engine to the execution level in the workshop, continuously monitoring the actual execution of these instructions, and feeding back the actual completion status and other results to the status awareness module, thus forming a complete closed-loop control loop. Simultaneously, real-time information after scheduling execution also drives the workpiece pool management module to update, forming an information closed loop that maintains the validity of the decision candidate set, jointly ensuring the system's continuous adaptive adjustment.

[0073] To support the efficient operation of the above systems, such as Figure 5 As shown, this application also provides a large-model-enabled end-to-end self-decision-making device for workshop scheduling, which consists of the following key hardware components: Data storage and processing equipment: one or more high-performance servers, responsible for storing massive amounts of relevant data and running the various modules of the aforementioned scheduling system; High-performance computing units: Because the training and real-time inference of large models have extremely high requirements for computing resources, the device must include dedicated high-performance computing units. These units consist of one or more graphics processing units (GPUs) or ASIC chips designed specifically for artificial intelligence computing. Input / output devices: These devices are used to enable information exchange between people and the dispatching system, as well as between the dispatching system and other external systems, such as operator terminals, large data display screens, and network interface devices; Network communication equipment: Ensures high-speed, stable, and secure data communication between components within the device and between the device and external networks outside the workshop.

[0074] Compared with existing technologies, the beneficial effects of this application are summarized as follows: This application proposes a "direct reasoning-driven" paradigm in which the large model itself completes end-to-end reasoning and directly outputs decision instructions.

[0075] First, it enables end-to-end intelligent decision-making, enhancing the intelligence level of decision-making: Unlike previous approaches where large models merely served as auxiliary tools for modeling or code generation, the large model in this application undertakes the core scheduling decision-making task. It can comprehensively understand complex workshop conditions and diverse input information, autonomously reason and judge, and directly output scheduling instructions, achieving a closed loop from environmental perception to decision execution. This model is closer to the thinking of human scheduling experts, i.e., "human brain scheduling," making the scheduling decision-making process more intelligent and flexible.

[0076] Secondly, the system is endowed with natural language interaction capabilities and dynamic event handling capabilities: users can intuitively describe scheduling needs, adjust optimization goals, and report sudden dynamic events, such as equipment failures and order changes, through natural language. The large model, with its excellent natural language understanding capabilities, can capture user intent and key information about events and integrate them into the current decision-making process, dynamically adjusting scheduling strategies. This helps improve the system's usability and responsiveness, reduces the professional skills required of operators, and enables non-scheduling professionals to effectively manage and intervene in complex scheduling processes.

[0077] Third, it reduces the cost of customized development and the application threshold, enabling low-code / no-code development: The development of traditional scheduling systems often requires extensive hard-coded rule design and complex algorithm customization for specific workshop environments and production processes. This application leverages the powerful learning capabilities and generalization potential of large models to internalize much implicit scheduling knowledge and decision-making logic into the model parameters, thereby reducing explicit programming work. This allows workshop scheduling to move towards "low-code" or even "no-code" development, not only shortening the system development cycle and reducing initial investment costs, but also simplifying subsequent maintenance and upgrade work.

[0078] Fourth, it possesses the potential for better decision-making and "foresight": Large models can learn and capture deeper patterns, dependencies, and long-term effects in data during training. Therefore, when making scheduling decisions, it is expected to surpass traditional heuristic rules or DRL methods that only focus on the next step, considering the potential impact of the current decision on several future steps and even the entire scheduling cycle, thus making more "foresighted" judgments, avoiding getting trapped in local optima, and improving overall scheduling performance. For example, the model might choose a job with a slightly longer processing time if doing so can quickly release bottleneck resources for subsequent more critical jobs, thereby achieving better global results.

[0079] Fifth, it supports coupled decision-making for better resource coordination: Decisions in shop floor scheduling are often coupled, such as workpiece selection, machine selection, worker allocation, and machine speed selection, all of which require coordinated consideration. Traditional methods typically decouple these decisions, making it difficult to achieve global optimum. In this application, the large model can handle high-dimensional, complex inputs that mix structured data and natural language, and possesses powerful nonlinear mapping capabilities. This allows it to implicitly learn the complex dependencies between multiple coupled variables, such as workpiece selection and machine assignment, thereby generating a co-optimized decision in a single inference, which is difficult to achieve with traditional decoupled methods.

[0080] Sixth, clear optimization goals: This application aims to optimize a series of key production indicators through intelligent decision-making, including but not limited to minimizing maximum completion time, minimizing total delay, maximizing overall equipment efficiency, and balancing machine load, thereby potentially improving the production and operation level of the workshop.

[0081] Example This embodiment uses a typical Flexible Job Shop Scheduling (FJSP) scenario as an example to illustrate the specific application process of the method in this application. Assume a shop has three jobs (J1, J2, J3) to be processed. Each job has multiple operations, and each operation can be performed on multiple selectable machines, with processing times varying depending on the machine. There are three machines (M1, M2, M3) in the shop. The scheduling objective is to minimize the maximum completion time of all jobs.

[0082] Initialization: The system loads the process routes for workpieces J1, J2, and J3 (J1: OP11→OP12; J2: OP21→OP22; J3: OP31), the processing time for each operation on each selectable machine, and the workpiece priorities (initial priorities are the same). All machines M1, M2, and M3 are initially idle. The first operation OP11 of workpiece J1, the first operation OP21 of J2, and the first operation OP31 of J3 are released and added to the "workpiece pool". At this time, the workpiece pool contains {OP11, OP21, OP31}.

[0083] First Decision Trigger: Due to the availability of idle machines and pending processes, the system triggers its first scheduling decision. The input to the large model includes: the current shop floor status is M1 idle, M2 idle, and M3 idle; the workpiece pool is {OP11 (optional M1 / M2, J1), OP21 (optional M2 / M3, J2), OP31 (optional M1 / M3, J3)}, with example information in parentheses; the scheduling objective is to minimize Makespan. Large Model Decision: After reasoning, it is assumed that the large model outputs the instruction: "Schedule process OP21 of workpiece J2 on machine M2." This decision may be based on the shorter processing time of OP21, or that choosing M2 can create more favorable conditions for subsequent processes, reflecting the model's comprehensive consideration of coupling factors. Execution and Status Update: The instruction is issued, and OP21 begins processing on M2. The status of M2 changes to "Processing". OP21 is removed from the workpiece pool or marked as assigned.

[0084] Subsequent Decision-Making and Dynamic Event Handling: Assume that during the processing of OP21 on M2, the user inputs a dynamic event through the natural language interface: "Urgent Notice: The main bearing temperature of machine M1 is too high, requiring immediate shutdown for maintenance, with an estimated downtime of 2 hours. Please adjust the production plan." When OP21 completes processing on M2, M2 becomes idle, which constitutes a "process completion" decision trigger event. The system simultaneously checks the next process OP22 of J2, confirming that its release condition (i.e., the preceding process OP21 has been completed) is met, and therefore adds it to the workpiece pool. At this time, since M2 is idle and there are new tasks in the workpiece pool, a scheduling decision is triggered again. The information input to the large model (the second decision) includes: the current workshop status is M1 fault (expected to recover in 2 hours), M2 is idle, and M3 is idle (assuming no tasks were previously assigned or have been completed); the workpiece pool is {OP11 (J1), OP31 (J3), OP22 (J2)} and their respective details; the scheduling objective is to minimize Makespan; and the dynamic event information is the 2-hour M1 fault. Large-scale model decision-making: At this point, when making a decision, the large-scale model will explicitly know that M1 will be unavailable for the next 2 hours. It will comprehensively consider the constraints of M1 failure, the attributes of all tasks in the current task pool, and the overall scheduling goal to generate a set of coordinated scheduling instructions, rather than a single instruction. For example... Figure 6 As shown, an optimized decision might be to schedule the first operation OP11 of workpiece J1 on machine M3, which is also currently idle; simultaneously, for machine M2, which has just completed its task and been released, schedule the first operation OP31 of workpiece J3, followed by the second operation OP22 of workpiece J2. This fully utilizes all available resources and effectively responds to dynamic events. An example Gantt chart of the resulting scheduling is shown below. Figure 6 As shown.

[0085] This simplified example demonstrates how the method proposed in this application can make end-to-end, event-driven scheduling decisions using large models, and flexibly respond to and handle dynamic events described in natural language. In this process, users do not need to write code or modify complex model parameters, showcasing the advantages of low-code / no-code approaches.

[0086] It is understood that the detailed functional implementation of each of the above units / modules can be found in the description in the aforementioned method embodiments, and will not be repeated here.

[0087] It should be understood that the above-described device is used to execute the methods in the above embodiments. The implementation principle and technical effect of the corresponding program modules in the device are similar to those described in the above methods. The working process of the device can be referred to the corresponding process in the above methods, and will not be repeated here.

[0088] Based on the methods in the above embodiments, this application provides an electronic device that may include a processor, a communications interface, a memory, and a communication bus, wherein the processor, communications interface, and memory communicate with each other via the communication bus. The processor may invoke logical instructions stored in the memory to execute the methods in the above embodiments.

[0089] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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 this application.

[0090] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0091] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0092] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0093] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0094] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0095] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.

[0096] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A large-model-enabled end-to-end self-decision-making engine for workshop scheduling, characterized in that, The engine includes: The status awareness module is equipped with standard interfaces with the shop floor manufacturing execution system, equipment control system and sensor network to continuously acquire and update the global status information of the shop floor in real time. The workpiece pool management module, which is linked with the status perception module, is used to add or remove processes that meet the conditions into the workpiece pool in real time according to the process release rules, thereby dynamically constructing and maintaining a workpiece pool containing all processes that have been released and are waiting to be scheduled. The human-computer interaction module provides a user interface that supports natural language input and graphical result display. It is used to receive unstructured scheduling goals or dynamic event reports from users and to intuitively display the scheduling results. The large model decision engine module has a built-in large model that is used to obtain information from other modules when a scheduling decision is triggered. It integrates the current global status of the workshop, the process information in the workpiece pool, the scheduling target input by the user through natural language, and the dynamic event description to perform end-to-end autonomous reasoning and directly output scheduling instructions that include the coupling relationship between workpiece selection and machine assignment. The knowledge base module stores domain knowledge, hard constraint rules, and historical best practices related to workshop scheduling, and provides real-time support for the decision-making process of the large model decision engine module through retrieval-enhanced generation technology. The scheduling execution and feedback module is used to issue the scheduling instructions to the physical execution layer of the workshop, continuously monitor the execution status of the instructions, and feed back the actual completion status to the status perception module, thereby forming a closed-loop control loop. At the same time, the real-time information after scheduling execution will also drive the workpiece pool management module to update, forming an information closed loop to maintain the effectiveness of the decision candidate set.

2. The workshop scheduling end-to-end self-decision engine as described in claim 1, characterized in that, The overall status information of the workshop includes at least: the current status of each machine, the current processing progress of each workpiece, and the availability status of related production resources.

3. The workshop scheduling end-to-end self-decision engine as described in claim 1, characterized in that, Each process within the workpiece pool carries its detailed attributes, including the workpiece identifier, priority, standard processing time, list of available machines, and waiting time.

4. A large-model-enabled end-to-end self-decision-making method for workshop scheduling, characterized in that, Applied to the large model-enabled end-to-end self-decision engine for shop floor scheduling as described in any one of claims 1 to 3, the method comprises: S1. By connecting to the manufacturing execution system or sensor network, the global status information of the workshop can be sensed and collected in real time; S2. Based on global state information including the current processing progress of the workpiece, dynamically construct and maintain a workpiece pool containing all currently released processes that are waiting to be scheduled. S3. When the preset scheduling decision trigger event is met, input comprehensive information into a large model based on pre-training and fine-tuned by knowledge of the workshop scheduling domain; S4. Based on the inherent reasoning ability of the large model, perform end-to-end analysis and weighing of the comprehensive information, and directly generate the next executable scheduling decision instruction.

5. The decision-making method as described in claim 4, characterized in that, At the initial moment of the scheduling process, the first operation of all workpieces, or those operations without preceding operations or whose preceding conditions have been met, will be automatically placed into the workpiece pool.

6. The decision-making method as described in claim 4, characterized in that, The dynamic maintenance of the workpiece pool in S2 specifically includes: when a certain process of a workpiece is completed, if there is a subsequent process of the workpiece and its start-up conditions are met, the subsequent process is automatically added to the workpiece pool; when a new workpiece or an urgent order is issued and confirmed, its first process or the process that meets the start-up conditions is added to the workpiece pool.

7. The decision-making method as described in claim 4, characterized in that, The scheduling decision triggering events in S3 include at least one of the following: the completion of at least one workpiece's process, thereby releasing the processing machine; a new high-priority process entering the workpiece pool; a change in the status of the workshop machines; or a dynamic disturbance event.

8. The decision-making method as described in claim 4, characterized in that, The fine-tuning and decision-making process of large models after pre-training includes: (1) Supervised fine-tuning of the large model using historical or simulation data containing workshop state-decision pairs; (2) Using expert preference feedback or a reward function based on scheduling performance indicators, perform direct preference optimization and / or policy optimization based on reinforcement learning on the large model; (3) In the process of reasoning and judgment, combined with the retrieval enhancement generation technology, the scheduling rules or historical cases are dynamically retrieved from the external knowledge base to assist decision-making.

9. The decision-making method according to any one of claims 4 to 8, characterized in that, The method also includes: S5. The scheduling instruction is sent to the physical execution layer of the workshop, and the execution status of the instruction is continuously monitored. The actual completion status is fed back to the status perception module, thereby forming a closed-loop control loop. At the same time, the real-time information after scheduling execution will also drive the workpiece pool management module to update, forming an information closed loop to maintain the effectiveness of the decision candidate set.

10. A large-scale model-enabled end-to-end self-decision-making device for workshop scheduling, characterized in that, The device is equipped with a large-model-enabled end-to-end self-decision engine for workshop scheduling as described in any one of claims 1 to 3.

Images