Intelligent scheduling method for collaborative manufacturing of tunnel boring machines

Abstract

The invention belongs to the field of intelligent manufacturing, and particularly relates to an intelligent scheduling method for collaborative manufacturing of tunnel boring machines, in the method, multi-source heterogeneous data including design data, process data, production data, material data and the like are acquired, and a rich data basis is provided for intelligent scheduling. The data fusion method based on the ontology is adopted, and the problems of semantic heterogeneity and structural heterogeneity of multi-source heterogeneous data are effectively solved. By constructing the tunnel boring machine manufacturing field ontology model, semantic association and sharing of data from different sources are realized, and the availability and understandability of the data are improved. An algorithm combining mixed integer programming and deep reinforcement learning is provided for intelligent scheduling of tunnel boring machine collaborative manufacturing. A robust optimization method is introduced to enhance the robustness of the scheduling scheme, and the influence of uncertain factors such as raw material supply and equipment maintenance in the manufacturing process of the tunnel boring machine is considered.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel boring machine manufacturing, and in particular to an intelligent scheduling method for collaborative manufacturing of tunnel boring machines, aiming to solve the problem of collaborative scheduling of multiple processes and multiple resources in the manufacturing process of tunnel boring machines through innovative algorithms and data processing flows, thereby improving manufacturing efficiency, reducing costs and ensuring product quality. Background Art

[0002] With the large-scale development of tunnel engineering construction, the demand for tunnel boring machines is growing. Its manufacturing process involves the processing, assembly and complex process flow of many parts. Traditional tunnel boring machine manufacturing scheduling mainly relies on manual experience and simple planning and scheduling tools. Production planners manually formulate manufacturing plans and scheduling plans based on factors such as order requirements, equipment capacity, and personnel skills. This method has many limitations: First, faced with massive manufacturing data and complex process associations, it is difficult for humans to fully consider various constraints and optimization goals, which can easily lead to unreasonable plans, such as the coexistence of idle and overloaded equipment and long production cycles; secondly, in the manufacturing process, once there are emergencies such as equipment failure, delayed raw material supply or order changes, manual scheduling is difficult to respond and adjust quickly, often causing production interruptions or delays, increasing manufacturing costs and delivery risks; thirdly, traditional scheduling methods lack real-time monitoring and data analysis capabilities for the manufacturing process, and cannot promptly discover potential production bottlenecks and quality risks, making it difficult to achieve continuous optimization of the manufacturing process.

[0003] In recent years, intelligent scheduling technology has been gradually applied in the manufacturing industry, but it still faces some challenges in the field of tunnel boring machine manufacturing. Most of the existing intelligent scheduling algorithms are designed for common problems in general manufacturing, and fail to fully consider the particularity of tunnel boring machine manufacturing, such as the high-precision requirements for large-scale parts processing, the coordination of multi-system assembly, and the impact of complex geological adaptability design on manufacturing processes; moreover, the existing scheduling system has deficiencies in processing the fusion of multi-source heterogeneous data (such as design drawings, process documents, equipment operation data, quality inspection data, etc.), making it difficult to achieve data-driven accurate scheduling decisions; in addition, the current intelligent scheduling method needs to improve its robustness and adaptability when dealing with uncertain factors, and cannot effectively guarantee the stable operation of the tunnel boring machine manufacturing process in a complex and changing environment. Therefore, there is an urgent need for an innovative intelligent scheduling system and method for collaborative manufacturing of tunnel boring machines to meet the needs of industry development. Summary of the invention

[0004] The present invention provides an intelligent scheduling method for collaborative manufacturing of tunnel boring machines, comprising a data collection and preprocessing step, an intelligent scheduling modeling and solving step, and a monitoring and feedback optimization step; the data collection and preprocessing step collects multi-source heterogeneous data from various links of tunnel boring machine manufacturing and performs preprocessing; the intelligent scheduling modeling and solving step uses a mixed integer programming model to solve static scheduling problems and introduces a deep reinforcement learning algorithm for dynamic optimization, while using a robust optimization method to enhance the robustness of the scheduling solution; the monitoring and feedback optimization step monitors the manufacturing process in real time and optimizes the scheduling solution according to feedback information.

[0005] Preferably: the data collection and preprocessing steps include collecting data from design, process planning, production and manufacturing, and material management, using an ontology-based data fusion method for preprocessing, and using an outlier detection algorithm to process data to ensure reliability.

[0006] Preferably: the intelligent scheduling modeling and solving step utilizes a mixed integer programming model to decompose the manufacturing task into subtasks, establishes an allocation matrix between tasks and resources, solves the initial scheduling plan with a multi-objective function, and the deep reinforcement learning algorithm performs dynamic optimization based on real-time state information. The robust optimization method considers uncertain parameters to enhance the stability of the scheduling plan.

[0007] Preferably: the monitoring and feedback optimization step establishes a real-time monitoring system to display key information in a visual interface, collects feedback information including worker operation feedback, quality problem feedback, equipment maintenance records, and optimizes the scheduling plan and updates the scheduling algorithm and model based on the feedback information.

[0008] Preferably: it also includes training an echo state network, taking historical manufacturing data as input and outputting a prediction of future manufacturing status; then, in the ant colony optimization algorithm, ants select the next task and resource for scheduling decisions based on the pheromone concentration and the prediction results of ESN; after each iteration, the pheromone concentration is updated according to the quality of the scheduling plan, and the ESN is used to update the prediction of the new manufacturing status, guiding the ant colony to search for a better scheduling plan.

[0009] Preferably: it also includes encoding the equipment failure state into a binary vector, expressing the production progress deviation as a percentage value, selecting actions through a deep Q network, and the actions include adjusting the process sequence, replacing processing equipment, and delaying or advancing the task start time; according to the execution results of the actions, the environment gives the agent corresponding reward feedback, and the reward function comprehensively considers the changes in the production cycle, the improvement of equipment utilization, and the reduction of quality risks.

[0010] Preferably: an intelligent scheduling system is used, including a data acquisition and integration module, an intelligent scheduling decision module and a monitoring and feedback module; the data acquisition and integration module is used to collect multi-source heterogeneous data from various links of tunnel boring machine manufacturing, and integrate and preprocess them; the intelligent scheduling decision module uses an algorithm combining mixed integer programming and deep reinforcement learning to make scheduling decisions, and considers uncertainty factors to enhance robustness; the monitoring and feedback module monitors the manufacturing process in real time and collects feedback information to optimize the scheduling plan.

[0011] Preferably: the manufacturing process is regarded as a dynamic system, and the ESN is used to model it, so as to learn the changing rules of manufacturing tasks and resources in time series.

[0012] Preferably: the mixed integer programming model of the intelligent scheduling decision module takes minimizing the production cycle, maximizing equipment utilization and reducing production costs as the objective functions, considers the order delivery period, equipment capacity, manpower hours and material supply constraints to solve the initial scheduling plan, and the deep reinforcement learning algorithm dynamically optimizes the initial plan, and the robust optimization method enhances the scheduling plan's ability to resist interference from uncertain factors.

[0013] Preferably: the monitoring and feedback module displays key information of production progress, equipment operation status, and quality inspection results in a visual interface, collects feedback information during the manufacturing process and feeds it back to the intelligent scheduling decision module to optimize the scheduling algorithm and model.

[0014] Beneficial technical effects: Multi-source heterogeneous data, including design data, process data, production data, and material data, are obtained in each link of tunnel boring machine manufacturing, providing a rich data foundation for intelligent scheduling. The semantic heterogeneity and structural heterogeneity of multi-source heterogeneous data are effectively solved by adopting an ontology-based data fusion method. By constructing an ontology model in the field of tunnel boring machine manufacturing, the semantic association and sharing of data from different sources are realized, and the availability and comprehensibility of the data are improved. An algorithm combining mixed integer programming with deep reinforcement learning is proposed for intelligent scheduling of collaborative manufacturing of tunnel boring machines. The robust optimization method is introduced to enhance the robustness of the scheduling scheme, taking into account the influence of uncertain factors such as raw material supply and equipment maintenance in the manufacturing process of tunnel boring machines. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Schematic diagram of the process. DETAILED DESCRIPTION

[0016] Example 1

[0017] Collect multi-source heterogeneous data from all aspects of tunnel boring machine manufacturing. In the design stage, obtain CAD design drawings, design change records, and product structure tree information, and convert them into a processable data format through a specific data interface; in the process planning stage, collect CAPP process files, including process flow, processing parameters, fixture requirements and other data; in the production and manufacturing stage, install various sensors, such as machine tool power sensors, vibration sensors, and temperature sensors to monitor the equipment operation status, high-precision measuring instruments for parts processing size detection, and visual sensors installed on the assembly line for assembly process monitoring, etc., to collect equipment operation data, parts quality data, and assembly progress data; in the material management stage, track the storage, outbound storage, inventory location and quantity of raw materials and parts through barcodes, RFID and other technologies.

[0018] Integrate and preprocess the collected multi-source heterogeneous data. Use an ontology-based data fusion method to build an ontology model for tunnel boring machine manufacturing, map data from different sources and formats to a unified ontology framework, and achieve semantic association and sharing of data. For example, associate and match the names and numbers of parts in the design drawings with the process objects in the process documents and the actual parts in production and manufacturing to solve the problem of data heterogeneity. At the same time, clean, denoise and process outliers for the data to ensure its accuracy and completeness. For example, use an outlier detection method based on statistical analysis to identify and eliminate abnormal fluctuation points in equipment operation data, providing a reliable data basis for subsequent analysis and decision-making.

[0019] Based on the integrated data, an innovative algorithm combining mixed integer programming and deep reinforcement learning is used to make scheduling decisions. First, the static scheduling problem of tunnel boring machine manufacturing is modeled using a mixed integer programming model, taking into account constraints such as order delivery period, equipment capacity, labor hours, and material supply, and taking minimizing the production cycle, maximizing equipment utilization, and reducing production costs as the objective function to solve the initial scheduling solution. For example, the assembly process and parts processing process of the tunnel boring machine are regarded as tasks, and machine tools and assembly stations are regarded as resources. A distribution relationship model between tasks and resources is established and solved through an optimization algorithm.

[0020] Then, a deep reinforcement learning algorithm is introduced to dynamically optimize and adjust the initial scheduling plan. An intelligent agent based on a deep Q network (DQN) is constructed, and the real-time status information of the manufacturing process (such as equipment failure information, order change information, production progress deviation, etc.) is used as the input of the intelligent agent. The intelligent agent learns the optimal scheduling strategy through interaction with the manufacturing environment. For example, when a key processing equipment fails, the intelligent agent decides whether to adjust the process sequence, transfer production tasks to other equipment, or wait for equipment repair based on the current production task status, the idleness of other equipment, and the inventory status of materials. During the training process, a technology based on priority experience playback is used to prioritize the storage and learning of experience data that is of great value to the improvement of the intelligent agent strategy, thereby improving learning efficiency and convergence speed.

[0021] In addition, considering the uncertainty factors in the manufacturing process of tunnel boring machines, the robust optimization method is used to enhance the robustness of the scheduling scheme. The interval representation of uncertain parameters is introduced into the mixed integer programming model, such as the fluctuation range of raw material supply time and the uncertainty of equipment maintenance time. By solving the robust optimization problem, a scheduling scheme that can still maintain good performance within a certain uncertainty range is obtained, thereby improving the scheduling scheme's ability to resist interference from uncertain factors.

[0022] Real-time monitoring of the tunnel boring machine manufacturing process is carried out, and key information such as production progress, equipment operation status, quality inspection results, etc. is displayed through a visual interface. For example, the planned progress and actual progress of each process are compared in the form of a Gantt chart, and indicators such as equipment utilization and failure rate are displayed in the form of a dashboard, so that managers can intuitively understand the production situation.

[0023] Collect feedback information from the manufacturing process, such as workers' operational feedback, quality problem feedback, and equipment maintenance records, and promptly feed this information back to the intelligent scheduling decision module. For example, when a worker finds that a certain process has quality risks, the information is reported through the feedback system. The intelligent scheduling decision module re-evaluates the scheduling plan based on the feedback information, and may add a quality inspection process or adjust the processing parameters of subsequent processes to ensure product quality. At the same time, based on monitoring and feedback information, the scheduling algorithm and model are continuously optimized and improved to continuously improve the performance and adaptability of the scheduling system.

[0024] 2. Methodology

[0025] Comprehensively collect multi-source heterogeneous data from the design, process planning, production and manufacturing, and material management of tunnel boring machine manufacturing. In the design stage, the geometric information and attribute information in the CAD design drawings are extracted through data conversion tools and integrated with the design change records; in the process planning stage, the CAPP process files are parsed to extract process steps, processing methods, process parameters and other data; in the production and manufacturing stage, the sensor network is used to collect equipment operation data and quality inspection data in real time, such as the operation status data of the machine tool is collected every 1 second, and the dimensional accuracy data of the parts after processing is collected immediately after each process is completed; in the material management stage, the barcode or RFID reader is used to scan the material information regularly to update the material inventory and flow data.

[0026] The collected data is preprocessed using an ontology-based data fusion method. A tunnel boring machine manufacturing ontology model is constructed to define various entities (such as parts, equipment, processes, etc.) and their attributes and relationships (such as assembly relationships, processing sequence relationships, etc.). Data from different sources are mapped and converted according to the ontology model to achieve semantic unification of data. For example, the part entity in the design drawing is associated with the process entity in the process file through the "processing object" relationship. In the data cleaning process, an outlier detection algorithm based on data distribution characteristics, such as the box plot method, is used to identify and process outliers in equipment operation data. Data that exceeds the 1.5 times interval of the upper and lower quartiles of the box plot is regarded as an outlier, and is corrected or replaced according to the historical trend of the data and the values ​​of adjacent data points to ensure the reliability of the data.

[0027] The static scheduling problem of tunnel boring machine manufacturing is modeled using a mixed integer programming model. The manufacturing task of the tunnel boring machine is decomposed into multiple subtasks, such as parts processing tasks and assembly tasks, each of which has different processing time, resource requirements and process constraints. Manufacturing resources, such as machine tools, assembly workers, and fixtures, are classified and quantified, and an allocation matrix between tasks and resources is established. Taking minimizing the production cycle, maximizing equipment utilization, and minimizing material inventory costs as objective functions, while considering order delivery period constraints, equipment capacity constraints, and material supply constraints, a mixed integer programming model is constructed. For example, for the cutterhead processing task of a certain model of tunnel boring machine, it is necessary to determine which machine tools to process, the processing order, and the processing time arrangement, while meeting the constraints such as the processing accuracy requirements of the machine tools, the service life limit of the tool, and the supply time requirements of the raw materials.

[0028] An optimization algorithm is used to solve the mixed integer programming model to obtain an initial scheduling solution. For example, the model is solved using a branch-and-bound algorithm, which searches for the optimal task allocation solution by continuously branching and pruning. In the solution process, the algorithm parameters, such as branching strategies and pruning conditions, are reasonably set according to the scale and complexity of the problem to improve the solution efficiency.

[0029] The deep reinforcement learning algorithm is introduced to dynamically optimize the initial scheduling scheme. The real-time state information in the manufacturing process is digitized and characterized to construct the state space. For example, the equipment failure state is encoded as a binary vector, and the production progress deviation is expressed as a percentage value. According to the current state, the agent selects actions through the deep Q network. The actions include adjusting the process sequence, replacing processing equipment, delaying or advancing the task start time, etc. According to the execution results of the action, the environment gives the agent corresponding reward feedback. The reward function comprehensively considers factors such as changes in the production cycle, improved equipment utilization, and reduced quality risks. For example, if the agent's action shortens the production cycle and improves equipment utilization, a higher reward is given; conversely, if it causes quality problems or production delays, a penalty is given. Through a large number of training iterations, the agent learns the optimal scheduling strategy and realizes dynamic optimization of the manufacturing process.

[0030] Robust optimization methods are used to enhance the robustness of scheduling schemes. In the mixed integer programming model, uncertain parameters, such as raw material supply time and equipment maintenance time, are represented as interval numbers. For example, the supply time of a certain raw material is expected to be between 3-5 days, which is represented as an interval [3,5]. By solving the robust optimization problem, a scheduling scheme that can still meet certain performance requirements within the fluctuation range of uncertain parameters is obtained. For example, even if the raw material supply time is delayed to 5 days, the scheduling scheme can still ensure the continuity of production and will not cause delays in key processes.

[0031] A real-time monitoring system for the manufacturing process of tunnel boring machines is established. The real-time data obtained by the data acquisition and integration module is used to display information such as production progress, equipment operation status, and quality inspection results on a visual interface. For example, the big data visualization technology is used to draw a production progress curve to show how the completion ratio of each process changes over time; the internal structure and operation status of the equipment are displayed through a three-dimensional model, such as using different colors to indicate the normal, faulty, and maintenance status of the equipment.

[0032] Collect feedback information during the manufacturing process, including workers' operational experience feedback, quality problem feedback, equipment maintenance feedback, etc. For example, a worker feedback terminal is set up on the assembly line, and workers can input problems found during the assembly process at any time, such as loose fit of parts and components, unreasonable assembly sequence, etc. In the quality inspection link, the inspection results and quality problems found are fed back to the system in a timely manner.

[0033] Based on monitoring and feedback information, the intelligent scheduling decision module optimizes and adjusts the scheduling plan. For example, if a certain device is found to have frequent failures and the fault repair time is long, the scheduling decision module will re-evaluate the task allocation of the device, transfer some tasks to other similar devices, and adjust the planned time of subsequent processes. At the same time, use this feedback information to update and improve the scheduling algorithms and models. For example, based on feedback from quality issues, adjust the weight of the quality control strategy in the scheduling model, strengthen the monitoring and resource allocation of key quality processes; based on equipment maintenance feedback, optimize the arrangement of equipment maintenance plans in the scheduling plan to improve equipment reliability and availability. Specifically:

[0034] 1. Task decomposition and resource quantification

[0035] 1. Task decomposition:

[0036] The manufacturing task of the tunnel boring machine is decomposed into a series of subtasks, such as parts processing tasks (such as cutter head processing, shield processing, etc.), assembly tasks (such as assembling the processed parts into a complete machine), etc. Let the set (C = {C1, C2, ..., C n} represents the set of all subtasks, where n is the total number of subtasks.

[0037] 2. Resource quantification:

[0038] Classify and quantify the various resources involved in the manufacturing process. For example, manufacturing resources include the machine tool set M = {M1, M2, ..., M m} (where m is the number of machine tools), the set of assembly workers W = {W1, W2, ..., W w} (w is the number of assembly workers), fixture set F = {F1, F2, ..., F f}(f is the number of types of fixtures), etc.

[0039] 3. Objective function:

[0040] Minimize production cycle:

[0041] Assume t i For the i-th subtask C i The processing time on the jth machine tool M (if the subtask is not processed on this machine tool, then t i =0),s i is the start time of the ith subtask. The production cycle T can be expressed as the maximum completion time of all subtasks, that is:

[0042] T=max i∈C {s i +∑ j∈M t ij}

[0043] The goal is to minimize the production cycle, denoted as minT.

[0044] Maximize equipment utilization:

[0045] For each machine tool M, let the total time available for processing be T, and the total time actually used to process each subtask be ∑ i∈C t ij .Then the utilization rate U of machine tool M can be expressed as:

[0046]

[0047] We hope to maximize the average utilization of all machine tools, and the objective function is:

[0048]

[0049] The goal is to maximize Z1, i.e. maxz1.

[0050] Minimize material inventory costs:

[0051] Assume I is the inventory level of the k-th material, c is the unit inventory cost of the k-th material, and the material type set is K = {K1, K2, ..., K k}. Then the material inventory cost C I It can be expressed as:

[0052] C I =∑ k∈K c k ×I k

[0053] Material supply constraints:

[0054] Assume r i is the unit time demand of the kth material for the i-th subtask, and R is the supply rate of the kth material. Then:∑ i∈C r ik t ij x ij ≤R k

[0055] For all material types k∈K and all machine tools j∈M.

[0056] Human resource constraints:

[0057] For tasks that require manual operation, such as assembly tasks, h iw The number of working hours required by the wth assembler for the ith subtask. If the total number of assemblers is finite, let it be H, then:

[0058] ∑ i∈C h iw≤H

[0059] For all assemblers w∈W.

[0060] (II) Solution of mixed integer programming model

[0061] For the constructed mixed integer programming model, commonly used solution methods include branch and bound algorithm, cutting plane algorithm, etc. Take the branch and bound algorithm as an example:

[0062] 1. Initialization:

[0063] Take the original problem as the root node and set the initial upper and lower bounds. The upper bound can quickly get a feasible solution through some heuristic algorithms, and calculate its objective function value as the initial upper bound; the lower bound is usually set to negative infinity (because we want to minimize the objective function).

[0064] 2. Branches:

[0065] At the current node (starting from the root node), select an integer variable (e.g., a decision variable; x i ), divides the current node into two child nodes according to its possible value (0 or 1), corresponding to different value situations.

[0066] 3. Delimitation:

[0067] For each newly generated child node, a lower bound is obtained by solving the corresponding linear programming relaxation problem (i.e., allowing integer variables to take real numbers). If this lower bound is greater than the current upper bound, the child node can be pruned (i.e., no more branches are added to it) because it is impossible for it to produce a better solution.

[0068] 4. Iteration:

[0069] Repeat the above branch and bound operations, and continuously update the upper and lower bounds until the optimal solution is found or it is determined that there is no solution.

[0070] By continuously branching and pruning, the optimal task allocation plan is searched. During the solution process, according to the scale and complexity of the problem, the algorithm parameters are reasonably set, such as branching strategy (which integer variable to choose for branching), pruning conditions (such as deciding whether to prune based on the comparison between the lower bound and the upper bound), etc., to improve the solution efficiency.

[0071] After obtaining the initial scheduling plan, a deep reinforcement learning algorithm is introduced to dynamically optimize it.

[0072] 1. State space definition:

[0073] Equipment status: Let F be the fault status of the jth machine tool, normal is 0, and fault is 1. Encode the fault status of all machine tools into a binary vector F = [F1, F2, ..., F m].

[0074] Production progress status: Let P be the production progress, expressed as the percentage of the number of completed subtasks to the total number of subtasks, that is, Where ncompleted is the number of completed subtasks. target is the predetermined production progress target, then the production progress deviation can be expressed as d P =PP target .

[0075] Material inventory status: Let I be the inventory level of the kth material, and the inventory levels of all materials form a vector I = [I1, I2, ..., I k ].

[0076] In summary, the state space S can be expressed as S = {F, d P ,I).

[0077] 2. Action space definition:

[0078] Adjust the process sequence: Let (a1 be the action of adjusting the process sequence. For the two subtasks (C i and C, and the process sequence can be adjusted by exchanging their execution order in the scheduling scheme. An action vector a1 = [C i , C].

[0079] Replace processing equipment: Let (a2 be the action of replacing processing equipment. For subtask C i , transfer it from the current processing equipment M to another equipment M. An action vector a2 can be defined as [C i , M, M]. Delay or advance the task start time: Let a3 be the action of delaying or advancing the task start time. For subtask a3 C i , delaying or advancing its start time by Δt△t time units. An action vector a3=[C i △t].

[0080] The action space A can be expressed as A=a1,a2,a3.

[0081] 3. Reward function definition:

[0082] Production cycle reward: Set T rev is the production cycle before executing the action, T new is the production cycle after executing the action. new <T prev , then a reward for shortening the production cycle is given, and the reward value can be set to r T =α1(T prev -T new), where α1 is the reward coefficient for shortening the production cycle. new ≥T prev , then a corresponding penalty is given, and the penalty value can be set to r T = -α1(T new -T prev ).

[0083] Equipment utilization rate reward: Set U rev is the average utilization rate of the equipment before the action is executed, U new is the average utilization rate of the equipment after the action is executed. new >U prev , then a reward for improving equipment utilization will be given, and the reward value can be set to r U =α2(U new -U prev ), where α2 is the bonus coefficient for improving equipment utilization. new ≤U prev , then a corresponding penalty is given, and the penalty value can be set to r U = -α2(U new -U prev ).

[0084] Quality Risk Reward: Set Q rev Qnew is the quality risk indicator before the action is executed (it can be comprehensively evaluated by the frequency and severity of quality problems), and Qnew is the quality risk indicator after the action is executed. new prev , then a reward for reducing quality risk is given, and the reward value can be set to r Q =α3(Q prev -Q new ), where α3 is the reward coefficient for quality risk reduction. new ≥Q prev , then a corresponding penalty is given, and the penalty value can be set to r Q = -α3(Q new -Q prev ).

[0085] In summary, the reward function can be expressed as r = r T +r U +r Q .

[0086] Training process:

[0087] The agent selects action a∈A through a deep Q network (DQN) based on the current state S. After executing the action, the environment gives the agent corresponding reward feedback based on the reward function. Through a large number of training iterations, the agent learns the optimal scheduling strategy and realizes dynamic optimization of the manufacturing process.

[0088] ​In order to enhance the robustness of the scheduling scheme, a robust optimization method is adopted.

[0089] 1. Uncertain parameter representation:

[0090] Uncertain parameters, such as raw material supply time, equipment maintenance time, etc., are expressed as interval numbers. For example, if the supply time of a certain raw material is expected to be between a and b, it is expressed as the interval [a, b]. Let the uncertain parameter vector be ξ = [ξ1, ξ2, ...], where ξ; is a different uncertain parameter.

[0091] 2. Robust optimization model construction:

[0092] On the basis of the mixed integer programming model, the influence of uncertain parameters is considered. For example, for the objective function of minimizing the production cycle, within the fluctuation range of uncertain parameters, the production cycle is required to meet certain robustness requirements. Let T(ξ) be the production cycle after considering the uncertain parameters, and require that for all possible uncertain parameter values ​​(ξ∈Ξ (where Ξ is the set of uncertain parameter value ranges), the production cycle does not exceed a given robustness upper limit T R ,Right now:

[0093]

[0094] For other objective functions and constraints, they also need to be adjusted and redefined accordingly according to the influence of uncertain parameters to ensure the robustness of the scheduling scheme in an uncertain environment.

[0095] Advantageous Effects of the Invention

[0096] 1. Innovative data collection and fusion methods:

[0097] A comprehensive data collection system has been built, which can obtain multi-source heterogeneous data from all aspects of tunnel boring machine manufacturing, including design data, process data, production data, and material data, providing a rich data foundation for intelligent scheduling. Compared with traditional data collection methods, the data collection scope of the present invention is wider and the data types are more complete, which can more comprehensively reflect the actual situation of the manufacturing process.

[0098] The ontology-based data fusion method effectively solves the semantic heterogeneity and structural heterogeneity of multi-source heterogeneous data. By building an ontology model for tunnel boring machine manufacturing, the semantic association and sharing of data from different sources are realized, which improves the availability and comprehensibility of data. This data fusion method can better explore the intrinsic connections between data and provide more accurate and comprehensive information support for intelligent scheduling decisions.

[0099] 2. Innovation of intelligent scheduling algorithm based on mixed integer programming and deep reinforcement learning:

[0100] An algorithm combining mixed integer programming and deep reinforcement learning is proposed for intelligent scheduling of collaborative manufacturing of tunnel boring machines. Mixed integer programming can accurately model and solve the static scheduling problem of the manufacturing process, consider various constraints and optimization goals, and obtain the initial scheduling plan; deep reinforcement learning can optimize and adjust the initial plan according to the real-time dynamic information of the manufacturing process, and achieve effective response and dynamic decision-making to uncertain factors. This hybrid algorithm gives full play to the advantages of the two methods, overcomes the limitations of traditional scheduling methods in dealing with static and dynamic problems, and improves the accuracy and flexibility of scheduling.

[0101] The robust optimization method is introduced to enhance the robustness of the scheduling scheme, taking into account the influence of uncertain factors such as raw material supply and equipment maintenance in the manufacturing process of tunnel boring machines. By solving the robust optimization problem, a scheduling scheme that can still maintain good performance within a certain uncertainty range is obtained, which effectively reduces the interference of uncertain factors on the manufacturing process and improves the stability and reliability of production.

[0102] 3. Real-time monitoring and feedback optimization mechanism innovation:

[0103] A comprehensive monitoring and feedback system has been established to monitor the tunnel boring machine manufacturing process in real time and intuitively display key information such as production progress, equipment operating status, and quality inspection results. Through the visual interface, managers can promptly identify problems and abnormal situations in the production process, providing timely information support for decision-making.

[0104] Make full use of feedback information from the manufacturing process, including workers' operating experience, quality issues, and equipment maintenance information, to optimize and adjust the scheduling plan. This feedback optimization mechanism enables the scheduling system to continuously learn and improve, adapt to various changes and challenges in the manufacturing process, and improve the adaptability and intelligence of the scheduling system.

[0105] Example 2

[0106] Various types of sensors are deployed at various stages of the manufacturing site, including equipment status sensors (monitoring the operating parameters and fault information of processing equipment and assembly equipment), material sensors (inventory quantity and location information of raw materials and parts), personnel positioning sensors (worker's working position and status), and environmental sensors (temperature, humidity and other environmental parameters that affect manufacturing). At the same time, data such as order information, production plans, and process documents are obtained from information systems such as the enterprise resource planning (ERP) system and the manufacturing execution system (MES).

[0107] Data fusion technology is used to integrate data from different sources to eliminate data redundancy and inconsistency. The collected data is pre-processed, including data cleaning (removing outliers and noise data), data standardization (unifying the scale of data of different magnitudes) and data encoding (converting non-numeric data into a form suitable for calculation), providing a high-quality data foundation for subsequent scheduling decisions.

[0108] According to the design drawings and manufacturing process requirements of the tunnel boring machine, a manufacturing task model is established. The model includes the manufacturing tasks, assembly tasks, and quality inspection tasks of each component, and clarifies the input and output, process parameters, quality standards, and time limits of each task.

[0109] Decompose the overall manufacturing task into multiple subtasks, and analyze the sequence, parallel relationship and resource dependency between subtasks. Build a task relationship network to clearly express the logical structure of the entire manufacturing process and provide task-level constraint information for the scheduling algorithm.

[0110] Classify and model all types of resources involved in the manufacturing process, including production equipment (machine tools of different types and specifications, welding equipment, etc.), human resources (workers of different skill levels), material resources (raw materials, parts and their supplier information), and auxiliary resources (such as jigs and fixtures, transportation tools, etc.). Define the attributes of each resource, such as the processing capacity of the equipment, the skill range of the personnel, the supply cycle of the material, etc.

[0111] Track resource status changes in real time, and update resource availability, remaining workload, location, and other information in a timely manner through sensor data and information system feedback. Establish a resource status prediction model to predict resource status changes in the future based on historical data and current trends, and provide forward-looking information for scheduling decisions.

[0112] The intelligent scheduling model is constructed by comprehensively considering manufacturing tasks and resource information. The model takes minimizing the manufacturing cycle, maximizing resource utilization, and ensuring product quality as optimization goals, while considering complex constraints such as equipment maintenance plans, material supply restrictions, and personnel working hours.

[0113] The following three unreported algorithm models are used to implement intelligent scheduling decisions and generate detailed scheduling plans, including the start time, execution resources, completion time, and material distribution plan of each subtask. The generated scheduling plan is converted into specific scheduling instructions, and the instructions are issued to the corresponding execution units through the interface with the equipment control system, personnel management system, and material distribution system of the manufacturing site. Ensure that each link works in an orderly manner according to the scheduling plan. Monitor the execution of the manufacturing process in real time, and timely discover abnormal situations such as production delays, equipment failures, and material shortages by comparing the actual progress with the scheduling plan. When an abnormality occurs, the emergency response mechanism is activated, and the dynamic scheduling algorithm is used to adjust the original scheduling plan, reallocate resources, and adjust the task order to ensure the continuity of the manufacturing process.

[0114] (I) Solution 1: Scheduling model based on hyper-heuristic evolutionary fuzzy system The hyper-heuristic evolutionary fuzzy system combines the advantages of hyper-heuristic algorithm and evolutionary fuzzy system. Hyper-heuristic algorithm is used to select and combine low-level heuristic algorithms at a high level to adapt to different scheduling scenarios. Evolutionary fuzzy system optimizes fuzzy rules and membership functions through evolutionary algorithm, so as to deal with ambiguity and uncertainty in manufacturing scheduling, such as fuzzy evaluation of worker skill level and fuzzy estimation of equipment failure risk.

[0115] First, a set of multiple low-level heuristic scheduling algorithms is constructed. The hyper-heuristic algorithm selects appropriate heuristic algorithms from the set for combination according to the characteristics of the current manufacturing tasks and resources. At the same time, the evolutionary algorithm is used to evolve the rule base and membership function of the fuzzy system. During the scheduling process, the information of tasks and resources is input into the evolutionary fuzzy system, and scheduling decisions such as task allocation, resource selection and time scheduling are obtained based on fuzzy reasoning.

[0116] The model can adaptively select appropriate scheduling strategies in complex scheduling environments and effectively deal with various fuzzy and uncertain factors in tunnel boring machine manufacturing. Compared with traditional scheduling algorithms, it has stronger robustness and versatility, and can improve the quality and efficiency of scheduling decisions.

[0117] (II) Solution 2: Dynamic scheduling model integrating echo state network and ant colony optimization Echo state network (ESN) is a special recursive neural network that has the ability to quickly model dynamic systems. The manufacturing process is regarded as a dynamic system, and it is modeled using ESN to learn the changing patterns of manufacturing tasks and resources in time series. The ant colony optimization algorithm searches for the optimal solution by simulating the pheromone propagation mechanism of ants in the process of searching for food. In this model, ants search for scheduling solutions in the solution space composed of tasks and resources, and the update of pheromones is affected by the output of ESN, so that the ant colony optimization algorithm can better adapt to the dynamic changes of the manufacturing process.

[0118] First, the echo state network is trained, taking historical manufacturing data (including task progress, resource status, etc.) as input and outputting predictions of future manufacturing status. Then, in the ant colony optimization algorithm, ants select the next task and resource for scheduling decisions based on the pheromone concentration and the prediction results of ESN. After each iteration, the pheromone concentration is updated according to the quality of the scheduling plan, and the new manufacturing status is updated and predicted using ESN, guiding the ant colony to search for a better scheduling plan.

[0119] This fusion model can make full use of the modeling ability of the echo state network for dynamic systems and the search ability of the ant colony optimization algorithm to achieve dynamic and accurate scheduling of the tunnel boring machine manufacturing process. It can effectively respond to real-time changes in the manufacturing process, such as urgent order insertion, sudden equipment failure, etc., and improve the flexibility and response speed of scheduling.

[0120] (III) Solution 3: Hybrid Scheduling Model Based on Quantum Annealing and Particle Swarm Optimization The quantum annealing algorithm uses quantum fluctuation characteristics to search for the optimal solution in the solution space, and has the potential to quickly find the global optimal solution in complex energy terrain. The particle swarm optimization algorithm simulates the foraging behavior of bird flocks, allowing particles to fly in the solution space and adjust the flight direction based on their own and the group's experience to find the optimal solution. Combining the two, the quantum annealing algorithm is used to explore the high-quality solution area globally, and the particle swarm optimization algorithm performs a detailed search in the area found by the quantum annealing algorithm to improve the accuracy of the solution.

[0121] In the quantum annealing part, the scheduling problem is mapped to the spin state space of the quantum bit. By setting the appropriate Hamiltonian and annealing plan, the system gradually converges to a low-energy state under the action of quantum fluctuations, that is, a better scheduling solution. Then, the solution obtained by quantum annealing is used as the initial population of the particle swarm optimization algorithm, and the speed and position update rules of the particles are set to make the particles evolve towards a better scheduling solution in the solution space. In the whole process, the quality of the solution is evaluated according to the scheduling objective function (such as manufacturing cycle, resource utilization, etc.) to guide the search direction of the algorithm.

[0122] The hybrid model combines the global search capability of the quantum annealing algorithm with the local optimization capability of the particle swarm optimization algorithm, and can more effectively search for the optimal scheduling solution in the complex scheduling problem space. For scheduling problems such as tunnel boring machine collaborative manufacturing with a large number of constraints and complex objectives, this hybrid algorithm can improve the quality of the scheduling solution and reduce the calculation time.

[0123] Example 3

[0124] 1. Sensor data collection and transmission

[0125] According to the structure and operation characteristics of the shield machine, the installation positions of various sensors are determined. For example, displacement sensors are installed at key parts of the shield machine, such as the cutter head, propulsion cylinder, and shield body, to accurately measure the displacement of different positions; angle sensors are installed near the rotating parts of the shield machine to monitor the rotation angle; pressure sensors are installed on the hydraulic system pipeline to obtain pressure data; speed sensors are installed at the travel mechanism of the shield machine to monitor the travel speed.

[0126] Configure parameters for each sensor, including sampling frequency, measurement range, accuracy, etc. Set the sampling frequency reasonably according to the speed of change of shield machine operation data and the accuracy requirements. For example, for data with relatively slow changes in displacement and angle, the sampling frequency can be set to 1-10 times per second; for data that may change quickly, such as speed and pressure, the sampling frequency can be set to 10-100 times per second. At the same time, according to the range of the actual operating parameters of the shield machine, determine the measurement range of each sensor, and select sensors with appropriate accuracy to ensure that the collected data can accurately reflect the operating status of the shield machine.

[0127] Establish a transmission mode that combines wired and wireless networks. Inside the shield machine, for sensors with short distances and high data transmission requirements and local data acquisition nodes, a wired network (such as Ethernet) is used to connect to ensure the stability and high speed of data transmission. For long-distance transmission between the shield machine and the central server, considering the complexity of the construction environment, a wireless network (such as 4G / 5G network) is used for data transmission. Configure the data transmission protocol to ensure that the collected data can be accurately and completely transmitted to the central server. For example, the TCP / IP protocol is used for network communication, the collected data is encapsulated at the data sending end, and a data packet header (including source address, destination address, data length, etc.) and a check code (used to detect the integrity of the data during transmission) are added. The received data is decapsulated and checked at the receiving end. If the check fails, the data is requested to be resent.

[0128] (II) Data cleaning and feature extraction

[0129] Invalid data removal: invalid data is identified by setting a reasonable range and logical rules for the data. For example, for the data collected by the displacement sensor, if its value exceeds the actual possible displacement range of the shield machine (determined according to the design size and construction conditions of the shield machine), the data is determined to be invalid and removed. For pressure data, if its value is negative or exceeds the normal working pressure range of the hydraulic system, it is also treated as invalid data and processed.

[0130] Redundant data processing: Analyze the collected data sequence and find data segments where the data of multiple consecutive sampling points are almost unchanged or change very little. These data may be redundant data caused by the accuracy limitation of the sensor or the relatively stable collection environment. For such redundant data, you can choose to retain some key data points (such as retaining one data point at a certain time interval) or delete them directly according to the specific situation to reduce the burden of data storage and processing.

[0131] Missing value processing: When missing values ​​are detected in the data, different processing methods are used according to the characteristics of the data and the missing situation. If there are few missing values ​​and the data changes are relatively smooth, linear interpolation can be used to fill them, that is, the estimated value of the missing value is calculated through a linear relationship based on the data points before and after the missing value. If there are many missing values ​​and the data has a certain periodicity or correlation, methods based on time series analysis (such as moving average method, seasonal decomposition method, etc.) can be used to fill them. If the data where the missing value is located has little impact on subsequent analysis, you can also choose to directly delete the record corresponding to the missing value.

[0132] Feature extraction based on physical meaning: Directly extract physical features directly related to the operating status of the shield machine, such as the displacement, angle, speed, pressure, etc. mentioned above. These features can intuitively reflect the working status of the shield machine under different working conditions and are an important basis for subsequent model analysis.

[0133] Derived feature extraction: In addition to basic physical features, some derived features can be extracted by performing mathematical operations and combinations on the original data. For example, the rate of change of displacement (i.e., the approximate value of speed) can be calculated by dividing the displacement difference between adjacent time points by the time interval, which can reflect the acceleration of the shield machine's advancement; the ratio of pressure to flow (assuming that flow data can be obtained through other sensors or known conditions) can be calculated, which can reflect the working efficiency of the hydraulic system to a certain extent. These derived features can provide richer information and help to gain a deeper understanding of the operating characteristics of the shield machine.

[0134] 3. Model construction and training

[0135] 1. Deep learning model based on variational autoencoder (VAE):

[0136] Determine the dimension of the input layer of VAE, which is consistent with the dimension of the feature vector after data cleaning and feature extraction. For example, if 10 features including displacement, angle, speed, pressure, etc. are extracted, the number of input layer nodes is set to 10. Design the encoder and decoder structure of VAE. The encoder is usually composed of multiple hidden layers, each hidden layer uses a different number of neurons (such as the first hidden layer is set to 128 neurons, the second hidden layer is set to 64 neurons, etc.), and the input data is gradually encoded through the fully connected layer and activation function (such as ReLU activation function), compressing the high-dimensional input data into a low-dimensional latent variable representation (i.e., mean and variance vectors). The decoder is the inverse process of the encoder, which gradually decodes the latent variables through multiple hidden layers and restores them to output data with the same dimension as the input data.

[0137] The loss function of VAE consists of two parts: reconstruction loss and KL divergence loss. Reconstruction loss is used to measure the difference between the decoder output data and the original input data. The mean square error (MSE) is usually used as the reconstruction loss function, that is, the average of the sum of the squares of the differences between the output data and the corresponding elements of the input data is calculated. KL divergence loss is used to measure the difference between the distribution of latent variables and the prior distribution (usually assumed to be a standard normal distribution). It is obtained by calculating the KL divergence between the mean and variance of the latent variables and the mean and variance of the standard normal distribution. The total loss function is the weighted sum of the reconstruction loss and the KL divergence loss. By adjusting the weight parameters, the influence of the two during the training process can be balanced.

[0138] Use labeled data (i.e., data with changes in the TBM posture marked) for training. Divide the data into a training set and a validation set according to a certain ratio (e.g., 70% for training and 30% for validation). During the training process, the feature vectors in the training set are input into the VAE model, and the gradient of the loss function to the model parameters (e.g., the weights and biases of each hidden layer neuron) is calculated by the back propagation algorithm. Then, the model parameters are updated according to the gradient descent algorithm (e.g., the stochastic gradient descent algorithm and its variants, such as Adagrad, Adadelta, etc.) to gradually reduce the loss function. After each training epoch (a complete round of training data set traversal), the training effect of the model is judged based on the loss function value on the validation set and other evaluation indicators (e.g., reconstruction error, rationality of the distribution of latent variables, etc.), and the training parameters (e.g., learning rate, weight decay, etc.) are adjusted until the model achieves better performance on the validation set (e.g., the loss function value converges to a smaller value and no longer decreases significantly).

[0139] Graph Neural Network Model Based on Graph Attention Network (GAT):

[0140] The graph structure is constructed based on the relationship between the component structure and operation data of the shield machine. The various components of the shield machine (such as the cutter head, propulsion cylinder, shield body, etc.) are regarded as nodes of the graph, and the connection relationship between the components (such as hydraulic transmission relationship, mechanical transmission relationship, etc.) and the correlation between the data (such as the relationship between the displacement data at different positions and the overall posture) are regarded as edges of the graph. A corresponding weight is assigned to each edge, and the weight can be determined based on factors such as the importance between the components and the strength of the data correlation. For example, a higher weight is assigned to the connection relationship between components that have a greater impact on the posture of the shield machine. The node feature vector of the graph is determined, and its dimension is consistent with the dimension of the feature vector after data cleaning and feature extraction, that is, the extracted displacement, angle, speed, pressure and other features are used as the feature vector of the node.

[0141] The core of GAT is the attention mechanism, which calculates the attention of each node to other nodes. In the model architecture, multiple attention layers are set, and each attention layer contains multiple attention mechanisms (such as an attention mechanism set to 8 heads). The attention mechanism of each head calculates the attention weights between nodes, performs weighted summation on the feature vectors of the nodes, and obtains a new feature vector. After processing through multiple attention layers, the information interaction and feature extraction capabilities between nodes are gradually enhanced.

[0142] After the last attention layer, a fully connected layer is set to map the processed node feature vector to the output layer. The number of nodes in the output layer is determined by the dimension of the shield machine posture change that needs to be predicted. For example, if the posture change of the shield machine in three-dimensional space is to be predicted, the number of output layer nodes can be set to 3. The cross entropy loss function is used as the loss function of the GAT model. When predicting the posture change of the shield machine, the predicted result is compared with the annotated actual posture change, and the cross entropy between the predicted result and the actual result is calculated. The smaller the cross entropy, the closer the predicted result is to the actual result, and the better the performance of the model.

[0143] Similarly, the labeled data is divided into a training set and a validation set according to a certain ratio (such as 70% for training and 30% for validation).

[0144] During the training process, the constructed graph structure and the corresponding node feature vectors are input into the GAT model, and the gradient of the loss function to the model parameters (such as the weight of the attention layer, the weight and bias of the fully connected layer, etc.) is calculated through the back-propagation algorithm, and then the model parameters are updated according to the gradient descent algorithm to gradually reduce the loss function. After each training epoch, the training effect of the model is judged according to the loss function value and other evaluation indicators (such as prediction accuracy, recall rate, etc.) on the validation set, and the training parameters (such as learning rate, number of attention layers, etc.) are adjusted until the model achieves good performance on the validation set (such as the prediction accuracy reaches a certain threshold and no longer decreases significantly).

[0145] 2. Integrated learning model based on hybrid expert system (MoE):

[0146] Construct multiple expert networks of different types, each of which can be based on different machine learning algorithms or deep learning architectures. For example, you can build an expert network based on a decision tree, an expert network based on a support vector machine, an expert network based on a convolutional neural network, etc. Each expert network learns and predicts different aspects or features of the shield machine operation data.

[0147] Determine the input feature vector of each expert network, whose dimension is consistent with the feature vector dimension after data cleaning and feature extraction. Each expert network processes and learns the input feature vector differently according to its own characteristics and data processing methods.

[0148] The function of building a gating network is to determine the degree of participation of each expert network in the prediction process based on the characteristics of the input data and the current operation status. The gating network usually consists of multiple hidden layers, processes the input data through a fully connected layer and an activation function (such as the ReLU activation function), and outputs the weight coefficient of each expert network. These weight coefficients represent the importance of each expert network in the ensemble prediction.

[0149] When predicting, first input the feature vector after data cleaning and feature extraction into the gated network, and the gated network calculates the weight coefficient of each expert network based on the input data. Then input the same feature vector into each expert network respectively, and each expert network outputs its own prediction results according to its own learning and prediction capabilities. Finally, the prediction results of each expert network are weighted and summed according to the weight coefficients given by the gated network to obtain the final integrated prediction result. The mean square error (MSE) is used as the loss function of the MoE model. The final integrated prediction result is compared with the annotated actual result, and the mean square error between the two is calculated. The smaller the MSE, the closer the integrated prediction result is to the actual result, and the better the performance of the model. The annotated data is divided into a training set and a validation set according to a certain ratio (such as 70% for training and 30% for validation). During the training process, the feature vectors in the training set are input into the gated network and each expert network in turn, and the gradient of the loss function to the model parameters (such as the weights and biases of each hidden layer of the gated network, the weights and biases of the expert network, etc.) is calculated through the back propagation algorithm, and then the model parameters are updated according to the gradient descent algorithm to gradually reduce the loss function. After each training epoch, the training effect of the model is judged according to the loss function value and other evaluation indicators (such as the accuracy and stability of the integrated prediction) on the validation set, and the training parameters (such as the learning rate, the structure of the gated network, etc.) are adjusted until the model achieves good performance on the validation set (such as the accuracy of the integrated prediction reaches a certain threshold and no longer decreases significantly).

[0150] (IV) Posture Control

[0151] During the operation of the shield machine, sensors continuously collect data and transmit the data to the central server. On the central server side, the real-time data received is quickly processed, key features (such as displacement, angle, speed, pressure, etc.) are extracted, and compared with the preset normal operating range. For example, for displacement data, a reasonable displacement range is set. If the displacement value monitored in real time exceeds this range, it is determined that the posture of the shield machine may have changed and further analysis is required.

[0152] By using data visualization technology, the key operating data and status information of the shield machine are displayed in real time on the visualization interface of the human-computer interaction module, so that on-site engineers can intuitively understand the real-time operation of the shield machine. For example, the real-time changes of displacement, angle, speed, pressure and other data can be displayed in the form of dashboards, line graphs, etc., and the normal and abnormal states of the shield machine can be marked with different colors or icons.

[0153] The key feature data extracted from real-time monitoring is input into the trained model (such as the model based on VAE, GAT, MoE), and the model predicts the posture change of the shield machine in the future (such as the next 10 minutes, 30 minutes, etc.) based on the input data and its own learning ability. For example, the future rotation angle change of the shield machine's cutter head, the propulsion speed change of the propulsion cylinder, the displacement change of the shield body, etc., and the overall posture change that may be caused by these changes are predicted.

[0154] In order to improve the accuracy and reliability of the prediction, the prediction results of the model are fused. If multiple different models are used for prediction at the same time, the prediction results of each model can be integrated by weighted average, Bayesian fusion and other methods to obtain a comprehensive prediction result. For example, let the prediction result based on the VAE model be A, the prediction result based on the GAT model be B, and the prediction result based on the MoE model be C. Using the weighted average method, let the weights be w1, w2, and w3 (and w1+w2+w3=1), then the comprehensive prediction result D=w1A+w2B+w3C.

[0155] According to the results of posture prediction and the pre-set optimization strategy, specific control measures are formulated to adjust the posture of the shield machine. For example, if it is predicted that the rotation angle of the shield machine's cutterhead will exceed the normal range, which may cause uneven excavation, the drive motor speed of the cutterhead is adjusted to return its rotation angle to the normal range. If it is predicted that the propulsion cylinder is too fast, which may cause the shield machine's posture to be unstable, the oil supply pressure of the propulsion cylinder is reduced to slow down the propulsion speed.

[0156] These control measures are implemented through the shield machine's control system (such as hydraulic control system, electrical control system, etc.) to achieve real-time adjustment of the shield machine's posture. In the process of implementing control measures, the shield machine's operating status is continuously monitored to observe whether the posture is adjusted as expected. If the expected effect is not achieved, the prediction results and control strategies are re-evaluated and further adjustments are made.

[0157] A large amount of operating data is collected from multiple different shield machine construction projects, covering data under different geological conditions (such as soft soil formations, hard rock formations, composite formations, etc.), different construction stages (such as starting, excavation, receiving, etc.) and different equipment conditions (such as different cutter head speeds, propulsion speeds, etc.). Ensure that the collected data can fully reflect the various conditions of the shield machine in actual construction. Each set of collected data is divided into 80% for training and testing and 20% for final verification. Among them, the training set is used for the initial training of the model, the test set is used for preliminary evaluation and adjustment of the model performance during the training process, and the verification set is used to finally compare the performance of different models.

[0158] Model comparison selection: Model 1: Traditional empirical model: A model built based on traditional engineering experience and simple mathematical formulas. For example, through simple statistical analysis of historical construction data, a linear regression model is established to predict the change of the shield machine posture, and a fixed control threshold and strategy are set according to experience for posture control. This model does not involve complex deep learning or integrated learning mechanisms. Model 2: Single deep learning model (taking VAE as an example): Only a deep learning model based on variational autoencoder (VAE) is used for posture prediction and control. Operate according to the VAE model construction and training methods described above, and do not fuse or integrate with other models. Model 3: Integrated learning model (MoE+GAT+VAE fusion): An integrated learning model based on a hybrid expert system (MoE), combined with a graph neural network model based on a graph attention network (GAT) and a deep learning model based on a variational autoencoder (VAE) for fusion prediction. For example, the weighted average method is used to fuse the prediction results of the three models to obtain the final posture prediction result, and posture control is performed based on this result. The specific fusion weight can be adjusted and optimized according to the performance on the test set.

[0159] Evaluation index, accurately calculate the proportion of the number of samples with correctly predicted shield machine posture changes to the total number of samples. For each sample, the posture change (including changes in displacement, angle, speed, etc.) of the shield machine predicted by the model at a specific time in the future (such as the next 10 minutes) is compared with the actual posture change. If the predicted value and the actual value are within the preset error range (such as the displacement error does not exceed ±5 cm, the angle error does not exceed ±2 degrees, the speed error does not exceed ±0.1 m / s, etc.), the prediction is considered correct. By making such a comparison for all validation set samples, the prediction accuracy index is obtained. The higher the index, the stronger the model's ability to predict the posture change of the shield machine. For each sample, calculate the square difference between the shield machine posture change value predicted by the model (such as the predicted displacement value, angle value, speed value, etc.) and the actual value. Then average the square differences of all validation set samples to obtain the mean square error (MSE). The smaller the MSE value, the higher the prediction accuracy of the model, that is, the closer the predicted value is to the actual value. Posture stability: observe the fluctuation of the shield machine's posture during the entire construction process after it is controlled according to the model prediction results. The stability of the posture is quantified by calculating the standard deviation of posture parameters (such as displacement, angle, speed, etc.). The smaller the standard deviation, the more stable the shield machine posture and the better the control effect. For example, during a period of continuous construction time, the standard deviation of the displacement data is calculated. If the value is less than a certain threshold (such as 5 cm), it means that the displacement posture is relatively stable. Construction quality indicators: evaluate the construction quality based on the actual effect after construction. For example, for tunnel excavation accuracy, it can be evaluated by measuring the deviation between the actual tunnel profile and the designed profile. The smaller the deviation, the higher the excavation accuracy; for tunnel flatness, a laser flatness meter can be used to measure the flatness of the inner surface of the tunnel. The smaller the flatness value, the better the construction quality. Combining these construction quality indicators, a quantitative construction quality score is given, with a full score of 10 points, and points are deducted according to the actual deviation. Safety risk assessment: monitor whether there are situations that may affect safety during the construction process, such as tunnel collapse, abnormal wear of shield machine components, groundwater leakage, etc. If no safety issues occur, the safety risk assessment score is 10 points; if there are minor safety hazards (such as local groundwater leakage in a small area), the score is 6 points; if there are serious safety issues (such as tunnel collapse risk or serious wear of key components), the score is 2 points. The comprehensive score of the control effect evaluation is obtained by combining the posture stability, construction quality indicators and safety risk assessment scores. The higher the score, the better the control effect. Record the computing resources consumed by each model during the training and prediction process, including CPU usage, memory usage, and the time required for training or prediction. Evaluate the operating efficiency of the model by comparing the computing resource consumption of different models under the same data volume and hardware environment.Lower computing resource consumption means that the model can run on more common hardware devices, or training and prediction tasks can be completed faster on the same hardware, which has better practicality.

[0160] The test results show that after comparing the test data of multiple groups of different projects, the prediction accuracy of model 1 (traditional empirical model) is about 55% on average. This is because the traditional empirical model is based on simple linear relationships and fixed thresholds, and it is difficult to accurately capture the complex nonlinear relationships in the shield machine operation data and the changing laws under different working conditions. The prediction accuracy of model 2 (single deep learning model-VAE) is about 70% on average. Although the VAE model can learn the nonlinear characteristics in some data, the limitation of a single model is that it may not be able to fully consider the various factors involved in the change of the shield machine posture, such as the mutual influence between different components and the comprehensive effect of complex factors such as geological conditions. The prediction accuracy of model 3 (integrated learning model-MoE+GAT+VAE fusion) is about 85% on average. By integrating different types of models and performing fusion prediction, the advantages of each model can be fully utilized. For example, the expert system of MoE can effectively process data from different aspects, GAT can better capture the relationship in graph structure data, and VAE can learn the potential characteristics of the data, thereby more accurately predicting the posture change of the shield machine. The mean square error (MSE) of model 1 is about 0.25 on average. Due to its relatively low prediction accuracy and large deviation from the actual value, the MSE value is high. The MSE of model 2 is about 0.15 on average. Although it is lower than the traditional empirical model, there is still a certain prediction error, indicating that the single deep learning model has room for improvement in accurately fitting data. The MSE of model 3 is about 0.08 on average. By integrating the advantages of multiple models, its predicted value is closer to the actual value, which effectively reduces the mean square error, reflecting the advantages of the integrated learning model in improving prediction accuracy. The attitude stability of the shield machine under the control of model 1 is poor, and the standard deviation of attitude parameters such as displacement, angle and speed is about 8 cm, 3 degrees and 0.2 m / s on average. This is because it is difficult for traditional empirical models to dynamically adjust the control strategy according to real-time data, resulting in large attitude fluctuations. The attitude stability of the shield machine under the control of model 2 has improved, and the standard deviation of attitude parameters is about 5 cm, 2 degrees and 0.15 m / s on average. Although the VAE model can provide certain prediction information for control, the limitations of a single model still make the control effect room for improvement. The attitude stability of the shield machine under the control of model three is good, and the standard deviation of the attitude parameters is about 3 cm, 1 degree and 0.1 m / s on average. The integrated learning model can formulate a more effective control strategy through more accurate prediction and comprehensive consideration of various factors, so as to make the attitude of the shield machine more stable. Construction quality index: The average score of the construction quality index corresponding to model one is about 6 points. Due to the inaccurate control of the attitude of the shield machine, there are certain deviations in the excavation accuracy and tunnel flatness, which affects the construction quality. The average score of the construction quality index corresponding to model two is about 7 points. The application of the VAE model has improved the construction quality to a certain extent, but it has not yet achieved the best effect.The average score of the construction quality index corresponding to model three is about 9 points. The integrated learning model effectively improves the construction quality indicators such as excavation accuracy and tunnel flatness by better controlling the posture of the shield machine, which significantly improves the construction quality. The average safety risk assessment score of model one is about 6 points. The limitations of traditional empirical models may lead to safety hazards in some complex working conditions. For example, improper posture control may cause tunnel collapse risks or abnormal wear of components. The average safety risk assessment score of model two is about 7 points. Although the VAE model helps to reduce safety risks, a single model still has shortcomings in dealing with complex situations. The average safety risk assessment score of model three is about 9 points. The integrated learning model can minimize safety risks and ensure construction safety by accurately predicting and effectively controlling the posture of the shield machine. Comprehensive control effect evaluation score: The average comprehensive control effect evaluation score of model one is about 6 points. Considering the posture stability, construction quality indicators and safety risk evaluation scores, the traditional empirical model performs poorly in overall control effect. The average comprehensive control effect evaluation score of model two is about 7 points. Although the single deep learning model has improved in some aspects, the overall control effect still needs to be improved. The average comprehensive control effect evaluation score of model three is about 9 points. By integrating the advantages of multiple models, the integrated learning model performs well in terms of prediction accuracy and control effect, achieving effective control of the shield machine posture, improving construction quality and ensuring construction safety. Model one consumes relatively few computing resources during training and prediction, with an average CPU usage of about 20%, an average memory usage of about 2GB, and a short training or prediction time of about 10 minutes on average. This is because the computational complexity of traditional empirical models is low and does not require a large amount of data learning and complex calculations. Model two (VAE) has an average CPU usage of about 60% and an average memory usage of about 8GB during training. The training time is long, about 60 minutes on average. Because deep learning models require a large amount of data learning and complex neural network calculations, their computing resource consumption is large. Model three (MoE+GAT+VAE fusion) has an average CPU usage of about 80% and an average memory usage of about 12GB during training. The training time is the longest, about 90 minutes on average. Since the ensemble learning model involves the fusion and training of multiple models, its computational complexity is higher, so it consumes more computing resources. However, despite the large consumption of computing resources, this resource consumption is worthwhile from the perspective of its significant advantages in prediction accuracy and control effect evaluation, because it can provide more accurate posture prediction and more effective control for shield machine construction, thereby improving construction quality and safety.

[0161] Through the above comparative efficiency improvement test, it can be clearly seen that the scheme of fusion prediction based on the integrated learning model of hybrid expert system (MoE) combined with the graph neural network model based on graph attention network (GAT) and the deep learning model based on variational autoencoder (VAE) has obvious advantages in shield machine posture prediction, control effect, construction quality and safety assurance. Although its computing resource consumption is relatively high, the comprehensive benefits are significantly better than traditional empirical models and single deep learning models.

[0162] The present invention has been described in detail above in conjunction with the embodiments. However, those skilled in the art will appreciate that, without departing from the spirit of the present invention, the specific parameters in the above embodiments may be changed to form a plurality of specific embodiments, which are all within the common variation range of the present invention and will not be described in detail herein.

Claims

1. An intelligent scheduling method for collaborative manufacturing of tunnel boring machines, characterized in that: The method comprises a data collection and preprocessing step, an intelligent scheduling modeling and solving step, and a monitoring and feedback optimization step; the data collection and preprocessing step collects multi-source heterogeneous data from various links of tunnel boring machine manufacturing and performs preprocessing; the intelligent scheduling modeling and solving step uses a mixed integer programming model to solve the static scheduling problem and introduces a deep reinforcement learning algorithm for dynamic optimization, while using a robust optimization method to enhance the robustness of the scheduling scheme; the monitoring and feedback optimization step monitors the manufacturing process in real time and optimizes the scheduling scheme according to the feedback information.

2. The intelligent scheduling method for collaborative manufacturing of tunnel boring machines according to claim 1, characterized in that: The data collection and preprocessing steps include collecting data from design, process planning, production and manufacturing, and material management, using an ontology-based data fusion method for preprocessing, and using an outlier detection algorithm to process data to ensure reliability.

3. The intelligent scheduling method for collaborative manufacturing of tunnel boring machines according to claim 1, characterized in that: The intelligent scheduling modeling and solving steps utilize a mixed integer programming model to decompose the manufacturing task into subtasks, establish an allocation matrix between tasks and resources, solve the initial scheduling plan with a multi-objective function, perform dynamic optimization based on real-time status information using a deep reinforcement learning algorithm, and enhance the stability of the scheduling plan by considering uncertain parameters using a robust optimization method.

4. The intelligent scheduling method for collaborative manufacturing of tunnel boring machines according to claim 1, characterized in that: The monitoring and feedback optimization step establishes a real-time monitoring system to display key information in a visual interface, collects feedback information including worker operation feedback, quality problem feedback, and equipment maintenance records, and optimizes scheduling plans and updates scheduling algorithms and models based on the feedback information.

5. The intelligent scheduling method for collaborative manufacturing of tunnel boring machines according to claim 1, characterized in that: It also includes training an echo state network, which takes historical manufacturing data as input and outputs a prediction of future manufacturing states; then, in the ant colony optimization algorithm, ants select the next task and resource for scheduling decisions based on the pheromone concentration and the ESN prediction results; After each iteration, the pheromone concentration is updated according to the quality of the scheduling plan, and the ESN is used to update the prediction of the new manufacturing status to guide the ant colony to search for a better scheduling plan.

6. The intelligent scheduling method for collaborative manufacturing of tunnel boring machines according to claim 1, characterized in that: It also includes encoding equipment failure status into binary vectors, expressing production progress deviations as percentage values, and selecting actions through a deep Q network. The actions include adjusting the process sequence, replacing processing equipment, and delaying or advancing the task start time. Based on the execution results of the actions, the environment gives the agent corresponding reward feedback. The reward function comprehensively considers changes in the production cycle, improved equipment utilization, and reduced quality risks.

7. The intelligent scheduling method for collaborative manufacturing of tunnel boring machines according to any one of claims 1 to 6, characterized in that: An intelligent scheduling system is used, including a data acquisition and integration module, an intelligent scheduling decision module and a monitoring and feedback module; the data acquisition and integration module is used to collect multi-source heterogeneous data from various links of tunnel boring machine manufacturing, and integrate and pre-process them; the intelligent scheduling decision module uses an algorithm combining mixed integer programming and deep reinforcement learning to make scheduling decisions, and considers uncertainty factors to enhance robustness; the monitoring and feedback module monitors the manufacturing process in real time and collects feedback information to optimize the scheduling plan.

8. The intelligent scheduling system for collaborative manufacturing of tunnel boring machines according to claim 7, characterized in that: The manufacturing process is regarded as a dynamic system, and ESN is used to model it and learn the changing rules of manufacturing tasks and resources in time series.

9. The intelligent scheduling system for collaborative manufacturing of tunnel boring machines according to claim 7, characterized in that: The mixed integer programming model of the intelligent scheduling decision module takes minimizing the production cycle, maximizing equipment utilization, and reducing production costs as the objective functions, and considers the order delivery period, equipment capacity, labor hours, and material supply constraints to solve the initial scheduling plan. The deep reinforcement learning algorithm dynamically optimizes the initial plan, and the robust optimization method enhances the scheduling plan's ability to resist interference from uncertain factors.

10. The intelligent scheduling system for collaborative manufacturing of tunnel boring machines according to claim 7, characterized in that: The monitoring and feedback module displays key information such as production progress, equipment operating status, and quality inspection results in a visual interface, collects feedback information from the manufacturing process, and feeds it back to the intelligent scheduling decision module to optimize the scheduling algorithm and model.

Images