A casting scheduling method and system based on digital twinning

Abstract

The application relates to the technical field of foundry production management, in particular to a foundry production scheduling method and system based on digital twinning. According to the comprehensive quality feature vector, a dynamic disturbance event is determined first, and reinforcement learning adjustment is adopted for the disturbance event. If there is no dynamic disturbance or the disturbance is eliminated, a multi-objective optimization algorithm is directly adopted to accurately obtain a global optimal production scheduling method. The global optimal production scheduling method is accurately obtained through the multi-objective optimization algorithm, and the dynamic disturbance in the foundry production scheduling is actively and flexibly responded to through constraint condition determination and reinforcement learning adjustment, so that the accuracy and robustness of the global optimal foundry production scheduling method are further improved, thereby effectively improving the foundry production efficiency and reducing the production cost.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of foundry production management, in particular to a foundry production scheduling method and system based on digital twinning. BACKGROUND

[0002] As a core link of production management, the essence of foundry production scheduling is to analyze the multi-dimensional coupling of customer orders, production capacity and resource constraints, and to construct a time sequence planning and resource allocation scheme of production tasks. Under the current background of the surge of personalized customization demand and the improvement of equipment intelligence level, enterprises are facing higher production efficiency requirements and shorter production delivery cycles. Therefore, how to scientifically and reasonably arrange production plans and complete production requirements with quality and quantity has become a key technical breakthrough for improving the core competitiveness of foundry enterprises.

[0003] At present, some technical solutions for optimizing foundry production scheduling have been proposed. For example, Chinese patent CN202411681338.8 discloses an intelligent production scheduling and dispatching system and method based on digital twinning. The invention patent simulates the process of different production scheduling methods, generates corresponding simulation logs, and then determines the optimal scheme through simulation log comparison. However, this method is difficult to deal with sudden equipment failures and urgent orders in foundry production, and it is difficult to accurately obtain the optimal scheduling method under dynamic disturbance.

[0004] Chinese patent CN202411496857.7 discloses a production scheduling method, production scheduling device and electronic equipment for foundry production. First, a multi-dimensional array of sand box array, site array, bottleneck process array and material array is established, and then a genetic algorithm is used to integrate production resources throughout the foundry process, and then each process is optimized. However, the scheduling method obtained by this method is prone to local optimal solution, and cannot accurately obtain the global optimal scheduling method. SUMMARY

[0005] To solve the above problems existing in the prior art, the application provides a foundry production scheduling method and system based on digital twinning, which accurately obtains the global optimal scheduling method through a multi-objective optimization algorithm, actively and flexibly responds to dynamic disturbances in foundry production scheduling through constraint condition determination and reinforcement learning adjustment, further improves the accuracy and robustness of the global optimal foundry production scheduling method, and effectively improves the foundry production efficiency and reduces the production cost.

[0006] In a first aspect, the application provides a foundry production scheduling method based on digital twinning, comprising the following steps:

[0007] Parameter information of the foundry workshop is collected by workshop Internet of Things equipment, the foundry workshop is mapped to a digital twin layer according to the parameter information, a three-dimensional visualization model of the digital twin layer is constructed to map device states, process parameters and resource data in real time, and the mapped parameter information is reduced in dimension, and a comprehensive quality feature vector after dimension reduction is imported into a production scheduling integrated database;

[0008] Dynamic disturbance of production scheduling behavior is determined according to the production scheduling integrated database, if there is disturbance of production scheduling behavior, reinforcement learning adjustment is started, production scheduling strategy is dynamically adjusted in real time according to the type of disturbance, if there is no dynamic disturbance or the disturbance is eliminated, a multi-objective optimization algorithm is directly used to accurately obtain a global optimal production scheduling method.

[0009] As a preferred technical solution of the application, the parameter information dimension reduction includes the following steps:

[0010] The parameter information includes device parameters, casting finished product parameters, order delivery cycle and material logistics information;

[0011] The data exceeding the physical range or invalid in the parameter information is removed;

[0012] The parameter information is normalized, and the random forest algorithm is used to retain parameters with importance scores greater than an importance score threshold;

[0013] N1 groups of original space-time snapshots are extracted from the parameter information, a covariance matrix is calculated, eigenvalue decomposition is performed according to the covariance matrix to obtain a feature vector matrix, N2 groups of dominant modes with energy proportion exceeding a threshold are screened and retained, the original space-time snapshots are projected into a low-dimensional space formed by the dominant modes, so that the residual norm precision error is less than a precision error threshold, and a modal coefficient matrix after dimension reduction is obtained;

[0014] The modal coefficient matrix Y obtained after dimension reduction is extracted according to the column to obtain the original coefficient of the kth dominant mode, the weighted coefficient is calculated, the weighted modal coefficient matrix is superimposed along the row direction to obtain a comprehensive quality feature vector;

[0015] The comprehensive quality feature vector includes device parameter index, casting finished product parameter index, order delivery cycle index and material logistics information index;

[0016] The projection formula is:

[0017]

[0018] The covariance matrix is represented as:

[0019] ;

[0020] The weighted coefficient calculation formula is:

[0021] ;

[0022] Where X is the original spatiotemporal snapshot, Y represents the dominant mode, and Y is the dimensionality-reduced modal coefficient matrix. For the first A snapshot, Let C be the mean field and C be the covariance matrix. These are modal weights.

[0023] As a preferred embodiment of the present invention, the determination of dynamic disturbances in production scheduling includes:

[0024] Set constraints: Constraints include the range of equipment index when the equipment is operating normally, the range of expected parameters index of finished castings, the range of normal order delivery cycle index, and the range of material flow status index.

[0025] Disturbance type determination: Equipment malfunction disturbances are determined based on the range of equipment parameter indices and normal equipment operation; casting quality abnormalities are determined based on the range of casting finished product parameter indices and expected casting finished product parameter indices; urgent orders are determined based on the range of order delivery cycle indices and normal order delivery cycle indices; and order cancellations are determined based on the range of material logistics information indices and material flow status indices.

[0026] As a preferred embodiment of the present invention, the reinforcement learning is used to dynamically adjust the production scheduling strategy in real time according to the type of disturbance, including the following steps:

[0027] The state space is defined according to the type of disturbance. The state space includes equipment status, quality anomalies, order queue, and resource constraints. Equipment status includes equipment operating / fault status. Quality anomalies include casting composition deviation and inclusion content. Order queue includes order priority, delivery date, and process path. Resource constraints include mold availability, raw material inventory, and energy supply.

[0028] Based on the state space, adjustment strategies are set; adjustment strategies include sequential adjustment, equipment switching, resource reallocation, and process parameter optimization; sequential adjustment includes inserting / removing orders and reordering priorities; equipment switching includes migrating tasks from faulty equipment to standby equipment; resource reallocation includes mold allocation, personnel shift adjustment, and dynamic energy scheduling; process parameter optimization includes adjusting melting temperature and cooling rate.

[0029] Define the action space: Output the adjustment strategy through the action space;

[0030] An incentive function is defined according to the state space and the action space, so as to guide the scheduling system to learn an optimal scheduling strategy in a disturbed environment; the incentive function is used to quantify a punishment and a reward mechanism; the incentive function comprises:

[0031] a resource utilization rate: a quality pass rate: a default cost: an equipment idle loss: a quality rework cost: ;

[0032] wherein, the actual completion time is represented by t, the planned delivery time is represented by T, the actual capacity is represented by C, the theoretical maximum capacity is represented by Cmax, the qualified product quantity is represented by Q, the total output is represented by Qtotal, the order default fine is represented by P, the delay compensation is represented by D, the equipment idle time is represented by Tidle, the energy consumption cost per unit time is represented by E, the defective product quantity is represented by D, the single-piece rework cost is represented by R.

[0033] As a preferred technical solution of the present application, the reinforcement learning further comprises:

[0034] when the equipment fails, the state space adjusts the equipment state to equipment failure according to the disturbance type, the adjustment strategy provides a task migration of the failed equipment to a backup equipment and a process parameter optimization strategy according to the equipment failure type, and the parameter information after the response of the action space is introduced into the incentive function calculation through the active response of the action space, the resource utilization rate encourages the recovery of the capacity after the migration, and the equipment idle loss punishes the original equipment idle loss;

[0035] when the quality is abnormal, the state space adjusts the quality abnormality to a casting component deviation or an inclusion content according to the disturbance type, the adjustment strategy provides a smelting temperature adjustment or a cooling rate adjustment strategy according to the quality abnormality type, and the parameter information after the response of the action space is introduced into the incentive function calculation through the active response of the action space, the quality pass rate ensures the quality pass rate, and the quality rework cost inhibits the rework caused by excessive adjustment;

[0036] When the emergency insertion order or order cancellation occurs, the state space adjusts the order queue or resource constraint according to the disturbance type, the adjustment strategy provides a rearrangement priority or removal order strategy according to the emergency insertion order or order cancellation type, and actively responds through the action space. The parameter information after the response of the action space is introduced into the incentive function calculation, the on-time delivery rate incentive completes the emergency order on time, and the default cost constraint order cancellation default cost.

[0037] As a preferred technical solution of the present application, the multi-objective optimization algorithm is used to accurately generate a global optimal scheduling method, comprising:

[0038] Initialize the population: adopt three-dimensional matrix encoding, randomly generate chromosomes of process, device parameters and time parameters after disturbance elimination or without disturbance intervention according to three-dimensional matrix encoding, and constitute an initial population of the algorithm, which represents an initial casting scheduling scheme;

[0039] Calculate the fitness: calculate the fitness function for each chromosome;

[0040] Selection: select chromosomes from the initial population according to the fitness function for generating the next generation of chromosomes;

[0041] Crossing: generate new chromosomes by crossing the selected device and time sequence of the chromosomes;

[0042] Mutation: randomly adjust the process, device or time control variation to generate new chromosomes;

[0043] Iteration: repeat the selection, crossing and mutation operations until the maximum iteration number is reached or the fitness is no longer significantly improved;

[0044] Local optimal judgment: if the fitness variance of 90% individuals in the population is less than 0.1, it is judged that the local optimal solution is reached;

[0045] After falling into the local optimal solution, the simulated annealing operation is entered to expand the search space and finely search for better solutions in the local field;

[0046] Output the global Pareto front solution set, according to the specific scene demand, including the shortest production cycle, the lowest device energy consumption or the least material loss, optimize the device load distribution, and obtain the global optimal solution;

[0047] The time parameter includes the time consumption and time constraint of each process on the device.

[0048] The fitness function is used to quantify the good or bad degree of the chromosome and measure the comprehensive performance of the scheduling scheme.

[0049] As a preferred technical solution of the present application, the simulated annealing operation comprises:

[0050] Select the top 20% of individuals with the highest fitness in the current population for local optimization;

[0051] fitness function Convert to energy function ,and , This represents the maximum fitness of the current population.

[0052] Set the initial temperature based on the current population energy fluctuations. Where N is the number of iterations to stop. For average energy, For the first The energy of an individual;

[0053] Neighborhood generation using Gaussian perturbation ,in , With a mean of 0 and a variance of The normal distribution and These are the neighborhood solutions before and after Gaussian perturbation processing, respectively. Let L be the cooling rate, and L be the number of annealing cycles at each temperature. This is the initial step size;

[0054] The Metropolis criterion accepts inferior solutions but not those generated in the neighborhood. Calculate energy changes ,when Accept directly Replace the current solution; when With probability accept ;

[0055] Set a threshold for the energy decrease; if the energy decrease is less than the threshold, accelerate the cooling process. ,otherwise ,in, and These are the temperature values ​​after the k-th and (k+1)-th iterations, respectively;

[0056] Until the temperature drops to Or, if the maximum number of iterations is reached, the simulated annealing will exit.

[0057] As a preferred embodiment of the present invention, the optimized equipment load distribution includes:

[0058] The equipment load adjustment formula is generated by weighted summation:

[0059] ;

[0060] wherein, , and are weight coefficients, the value range is 0~1, and , is the total production cycle, is the total energy consumption of the equipment, is the material cost loss.

[0061] In a second aspect, the present application also provides a casting scheduling system based on digital twinning, which executes the casting scheduling method as described above, including a physical entity layer, a digital twinning layer and a scheduling optimization layer,

[0062] The physical entity layer includes smelting equipment, pouring equipment, molding equipment and cooling equipment, which are used to execute metal casting production;

[0063] The digital twinning layer includes a three-dimensional visualization model, which is used to map the physical layer equipment state, process parameters and resource data in real time;

[0064] The scheduling optimization layer includes a dynamic disturbance judgment unit, a reinforcement learning adjustment unit, a multi-objective optimization unit and a man-machine interaction interface, which are used to accurately generate a globally optimal scheduling method and actively respond to dynamic disturbance events;

[0065] The dynamic disturbance judgment unit is used to determine the disturbance type according to the constraint condition;

[0066] The reinforcement learning adjustment unit dynamically adjusts the scheduling strategy in real time according to the determination result of the dynamic disturbance unit;

[0067] The multi-objective optimization unit takes the parameters after disturbance elimination or without disturbance intervention as input, and accurately generates a globally optimal scheduling method by using a multi-objective optimization algorithm, while optimizing the equipment load distribution.

[0068] As a preferred technical solution of the present application, the man-machine interaction interface includes a scheduling Gantt chart and a multi-dimensional index board;

[0069] The scheduling Gantt chart is used for process drag adjustment and version management, and records the reinforcement learning adjustment traces;

[0070] The multi-dimensional index board is used to display the equipment utilization rate, energy consumption curve and dynamic disturbance in real time.

[0071] The beneficial effects of the present application include:

[0072] The application can accurately obtain a globally optimal production scheduling method solution set through a multi-objective optimization algorithm, and meet the globally optimal production scheduling method under specific requirements according to an optimized equipment load distribution method; meanwhile, dynamic disturbance determination is performed in combination with constraint conditions and a reduced-dimension comprehensive quality feature vector, reinforcement learning adjustment is performed for dynamic disturbance events, production scheduling strategies are continuously adjusted through an incentive function, disturbance interference is eliminated, and the accuracy and robustness of the globally optimal casting production scheduling method are further improved, so that the casting production efficiency is effectively improved and the production cost is reduced.

[0073] For the prior art, the application can flexibly cope with four typical dynamic disturbances (equipment failure, quality abnormality, urgent order insertion and order cancellation) in the casting production scheduling, and eliminate the interference of the dynamic disturbances on the optimal production scheduling method; meanwhile, the application can avoid falling into a local optimum in the production scheduling process, and accurately obtain a globally optimal production scheduling method. BRIEF DESCRIPTION OF DRAWINGS

[0074] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0075] Figure 1 A schematic diagram of a casting production scheduling method based on digital twinning provided in the embodiments of the present application;

[0076] Figure 2 A flowchart of reinforcement learning and a multi-objective optimization algorithm in the present application;

[0077] Figure 3 A structural schematic diagram of a casting production scheduling system based on digital twinning provided in the embodiments of the present application. DETAILED DESCRIPTION

[0078] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor also belong to the protection scope of the present application.

[0079] The optimal embodiments of the present application will be further described in combination with the drawings;

[0080] Please refer to Figure 1 and 2 The present embodiment provides a casting production scheduling method based on digital twinning, including the following steps:

[0081] Parameter information of the foundry workshop is collected by workshop Internet of Things equipment, and the foundry workshop is mapped to a digital twin layer according to the parameter information, the digital twin layer constructs a three-dimensional visualization model to map device state, process parameters and resource data in real time, and the parameter information obtained by mapping is reduced, and the comprehensive quality feature vector after dimension reduction is imported into the production scheduling integrated database;

[0082] The three-dimensional visualization model includes:

[0083] Foundry equipment modeling: SolidWorks is used to construct three-dimensional geometric models of smelting equipment, pouring equipment, cooling equipment, cleaning equipment and testing equipment, and MATLAB / Simulink is used to construct physical behavior models of the heating curve of the smelting equipment and the vibration frequency of the molding equipment, and device operating parameters and state constraint parameters are embedded;

[0084] The device operating parameters include rated power, maximum capacity and maintenance period, and the state constraint parameters include mold replacement time and temperature threshold;

[0085] Process flow modeling: define the process logic chain of foundry pouring, and use a hybrid modeling method combining Petri nets and directed graphs to abstract each process node into a state machine containing event trigger rules and event dynamic parameters, which can be set as: the trigger condition for smelting completion is that the alloy liquid temperature is greater than or equal to and the composition spectrum analysis is qualified; the trigger condition for pouring start is that the pouring positioning accuracy is less than and the temperature is greater than or equal to ; the cooling switch is triggered when the surface temperature of the casting is less than or equal to (phase transition point);

[0086] Material resource modeling: record mold specifications, remaining life, occupancy status, and inventory and consumption rate of sand, alloy liquid and waste materials;

[0087] The parameter information reduction includes the following steps:

[0088] The parameter information includes device parameters, casting finished product parameters, order delivery cycle and material logistics information;

[0089] The device parameters include pouring ladle liquid level temperature, mold surface temperature, smelting furnace gas pressure and cooling water system pressure, etc., the casting finished product parameters include casting composition deviation, inclusion content, shrinkage volume ratio and tensile strength, the order delivery cycle includes order entry time, first piece inspection completion time, urgent order insertion buffer time and delivery period, and the material logistics information includes mold inventory parameters, smelting process unit energy consumption and energy supply consumption;

[0090] Remove data in the parameter information that exceeds the physical range or is invalid;

[0091] The parameter information is normalized. Several feature points are set in the random forest. Each feature point is assigned a feature score according to its importance. Parameters with an importance score greater than the importance score threshold are retained.

[0092] N1 sets of original spatiotemporal snapshots are extracted from the parameter information, the covariance matrix is ​​calculated, and the eigenvalue decomposition is performed based on the covariance matrix to obtain the eigenvector matrix. N2 sets of dominant modes with energy proportions exceeding the threshold are selected and retained (N2 < N1). The original spatiotemporal snapshots are projected onto the low-dimensional space spanned by the dominant modes to make the residual normal form accuracy error less than the accuracy error threshold, and the dimensionality-reduced mode coefficient matrix is ​​obtained.

[0093] Extract each modal coefficient from the column of the reduced modal coefficient matrix Y. The original coefficients of the k-th dominant mode are obtained, and the weighting coefficients are calculated. The weighted modal coefficient matrices are superimposed along the row direction to obtain the comprehensive quality feature vector;

[0094] The comprehensive quality feature vector includes equipment parameter index, finished casting parameter index, order delivery cycle index, and material logistics information index;

[0095] The projection formula is:

[0096]

[0097] The covariance matrix is ​​expressed as follows:

[0098] ;

[0099] The formula for calculating the weighting coefficients is as follows:

[0100] ;

[0101] Where X is the original spatiotemporal snapshot, Y represents the dominant mode, and Y is the modal coefficient matrix obtained by dimensionality reduction. For the first A snapshot, Let C be the mean field and C be the covariance matrix. These are modal weights.

[0102] The scheduling behavior is dynamically perturbed based on the integrated scheduling database. If there is a perturbation in the scheduling behavior, reinforcement learning is initiated to adjust the scheduling strategy in real time according to the type of perturbation. If there is no dynamic perturbation or the perturbation is eliminated, a multi-objective optimization algorithm is directly used to accurately obtain the globally optimal scheduling method.

[0103] The determination of dynamic disturbances in production scheduling includes:

[0104] Setting constraints: constraints include the interval range of equipment index when the equipment is running normally, the interval range of casting product expected parameter index, the interval range of order normal delivery cycle index, and the interval range of material flow state index;

[0105] Disturbance type determination: determine equipment failure disturbance according to equipment parameter index and interval range of equipment parameter index when the equipment is running normally; determine casting quality abnormality according to casting product parameter index and interval range of casting product expected parameter index; determine emergency insertion order according to order delivery cycle index and interval range of order normal delivery cycle index; determine order cancellation according to material logistics information index and interval range of material flow state index;

[0106] The reinforcement learning is used for dynamically adjusting production scheduling strategy in real time according to the disturbance type, and includes the following steps:

[0107] Setting state space according to the disturbance type; the state space includes equipment state, quality abnormality, order queue, and resource constraint; the equipment state includes equipment running / failure state; the quality abnormality includes casting composition deviation and inclusion content; the order queue includes order priority, delivery period, and process path; the resource constraint includes mold availability, raw material inventory, and energy supply;

[0108] Setting adjustment strategy according to the state space; the adjustment strategy includes sequential adjustment, equipment switching, resource reallocation, and process parameter optimization; the sequential adjustment includes inserting / removing order and rearranging priority; the equipment switching includes task migration of a failed equipment to a standby equipment; the resource reallocation includes mold allocation, personnel shift adjustment, and energy dynamic scheduling; the process parameter optimization includes smelting temperature adjustment and cooling rate adjustment;

[0109] Setting action space: outputting the adjustment strategy through the action space;

[0110] Defining incentive function according to the state space and the action space to guide the production scheduling system to learn the optimal scheduling strategy in the disturbance environment; the incentive function is used for quantifying the punishment and reward mechanism; the incentive function includes on-time delivery rate:

[0111] resource utilization rate: quality pass rate: default cost: equipment idle loss: quality rework cost: ;

[0112] wherein, represents actual completion time, represents planned delivery time, is actual capacity, for theoretical maximum capacity, for number of good products, for total production, for order penalty, for delay compensation, for equipment idle time, for energy consumption cost per unit time, for number of defective products, for single-piece rework cost.

[0113] The reinforcement learning further includes the steps of:

[0114] When the equipment fails, the state space adjusts the equipment state to equipment failure according to the disturbance type, the adjustment strategy provides a fault equipment task migration to a backup equipment and a process parameter optimization strategy according to the equipment failure type, and through an active response of the action space, the parameter information after the response of the action space is introduced into the incentive function calculation, the resource utilization rate encourages the production capacity recovery after migration, and the equipment idle loss punishes the original equipment idle loss;

[0115] When the quality is abnormal, the state space adjusts the quality abnormality to a casting composition deviation or an inclusion content according to the disturbance type, the adjustment strategy provides a smelting temperature adjustment or a cooling rate adjustment strategy according to the quality abnormality type, and through an active response of the action space, the parameter information after the response of the action space is introduced into the incentive function calculation, the quality qualified rate ensures the quality qualified rate, and the quality rework cost inhibits excessive adjustment leading to rework;

[0116] When the emergency order insertion / order cancellation occurs, the state space adjusts the order queue or the resource constraint according to the disturbance type, the adjustment strategy provides a priority rearrangement or an order removal strategy according to the emergency order insertion / order cancellation type, and through an active response of the action space, the parameter information after the response of the action space is introduced into the incentive function calculation, the on-time delivery rate encourages the timely completion of the emergency order, and the penalty cost constrains the penalty cost of the order cancellation.

[0117] The multi-objective optimization algorithm is used to accurately generate a global optimal scheduling method, and includes the following steps.

[0118] Initialize the population: adopt three-dimensional matrix encoding, randomly generate chromosomes of process, equipment parameters and time parameters after disturbance elimination or without disturbance intervention according to three-dimensional matrix encoding, and construct an initial population of the algorithm, which represents an initial casting scheduling scheme;

[0119] Calculate the fitness: calculate the fitness function of each chromosome;

[0120] Selection: select the chromosomes from the initial population according to the fitness function, which are used to generate the next generation of chromosomes;

[0121] Crossover: generate new chromosomes by crossing selected chromosomes of the population and time series;

[0122] Mutation: generate new chromosomes by randomly adjusting process, equipment or time control variation;

[0123] Iteration: repeat the selection, crossover and mutation operations until the maximum iteration number is reached or the fitness is no longer significantly improved;

[0124] Local optimum determination: if the fitness variance of 90% of the population is less than 0.1, it is determined to be trapped in a local optimum;

[0125] After falling into a local optimum, enter the simulated annealing operation to expand the search space and fine search for better solutions in the local field;

[0126] Output the global Pareto front solution set, including the shortest production cycle, the lowest equipment energy consumption or the least material loss, optimize the equipment load distribution, and obtain the global optimal solution according to the specific scene requirements;

[0127] The time parameters include the time consumption and time constraints of each process on the equipment;

[0128] The fitness function is used to quantify the pros and cons of the chromosome and measure the comprehensive performance of the production scheduling scheme;

[0129] The simulated annealing operation includes:

[0130] Select the top 20% of individuals with the highest fitness in the current population for local optimization;

[0131] Convert the fitness function to an energy function , and , is the maximum fitness of the current population;

[0132] Set the initial temperature according to the energy fluctuation of the current population , where N is the number of stop iterations, is the average energy, is the energy of the th individual;

[0133] Generate a neighborhood using Gaussian disturbance , where , is a normal distribution with mean 0 and variance , and and are the neighborhood solutions before and after Gaussian disturbance processing, is the cooling rate, L is the number of annealing at each temperature, is the initial step size;

[0134] Metropolis criterion accepts inferior solutions, for the generated neighborhood solution , calculate the energy change , when , directly accepted Replace the current solution; when , with a probability Accept ;

[0135] Set the energy drop amplitude threshold, if the energy drop amplitude is less than the energy drop amplitude threshold, then accelerate cooling , otherwise , wherein And The temperature value after the kth and (k+1)th iteration, respectively;

[0136] Until the temperature drops to Or reach the maximum iteration preset value, exit the simulated annealing.

[0137] The optimization equipment load distribution includes:

[0138] Generate the equipment load adjustment formula by weighted summation:

[0139] ;

[0140] Wherein, , And All are weight coefficients, the value range is 0~1, and , Total production cycle, Total energy consumption of equipment, Material cost loss.

[0141] Please refer to Figure 3 , the embodiment provides a casting scheduling system based on digital twinning, executes the casting scheduling method, including physical entity layer, digital twinning layer and scheduling optimization layer,

[0142] The physical entity layer includes smelting equipment, pouring equipment, molding equipment and cooling equipment, which is used for executing metal casting production;

[0143] The digital twinning layer includes a three-dimensional visualization model for real-time mapping of physical layer device state, process parameters and resource data;

[0144] The system includes a scheduling optimization layer for accurately generating a globally optimal scheduling method and actively responding to dynamic disturbance events, and the system executes to realize the steps of the above-mentioned method;

[0145] The production scheduling optimization layer comprises a dynamic disturbance determination unit, a reinforcement learning adjustment unit, a multi-objective optimization unit and a man-machine interaction interface, and is used for accurately generating a globally optimal production scheduling method and actively responding to dynamic disturbance events.

[0146] The dynamic disturbance determination unit is used for determining a disturbance type according to a constraint condition.

[0147] The reinforcement learning adjustment unit adjusts a production scheduling strategy in real time according to a determination result of the dynamic disturbance unit.

[0148] The multi-objective optimization unit takes parameters after disturbance elimination or without disturbance intervention as input, and accurately generates a globally optimal production scheduling method by using a multi-objective optimization algorithm, while optimizing equipment load distribution.

[0149] The man-machine interaction interface comprises a production scheduling Gantt chart and a multi-dimensional index board.

[0150] The production scheduling Gantt chart is used for process drag adjustment and version management, and records reinforcement learning adjustment traces.

[0151] The multi-dimensional index board is used for displaying equipment utilization, energy consumption curves and dynamic disturbances in real time.

[0152] The above is only a preferred embodiment of the present application, and is not used to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A casting scheduling method based on digital twins, characterized in that: Includes the following steps: The casting workshop collects parameter information through IoT devices in the workshop, maps the casting workshop to a digital twin layer based on the parameter information, and the digital twin layer builds a three-dimensional visualization model to map the equipment status, process parameters and resource data in real time. The parameter information obtained by mapping is then reduced in dimensionality, and the comprehensive quality feature vector after dimensionality reduction is imported into the production scheduling integration database. The scheduling behavior is dynamically perturbed based on the integrated scheduling database. If there is a perturbation in the scheduling behavior, reinforcement learning is initiated to adjust the scheduling strategy in real time according to the type of perturbation. If there is no dynamic disturbance or the disturbance is eliminated, the multi-objective optimization algorithm is directly used to accurately obtain the globally optimal production scheduling method; The dimensionality reduction of the parameter information includes: The parameter information includes equipment parameters, finished casting parameters, order delivery cycle, and material logistics information; Remove data from parameter information that is outside the physical range or invalid; The parameter information is normalized, and the random forest algorithm is used to retain parameters whose importance scores are greater than the importance score threshold. N1 sets of original spatiotemporal snapshots are extracted from the parameter information, and their covariance matrices are calculated. Eigenvalue decomposition is performed on the covariance matrix to obtain the eigenvector matrix, and N2 sets of dominant modes with energy proportions exceeding a threshold are selected and retained. The original spatiotemporal snapshot is projected onto the low-dimensional space spanned by the dominant modes, so that the accuracy error of the residual paradigm is less than the accuracy error threshold, and the dimensionality-reduced mode coefficient matrix is ​​obtained. Extract each modal coefficient from the column-wise modal coefficient matrix Y obtained by dimensionality reduction. The original coefficients of the k-th dominant mode are obtained, and the weighting coefficients are calculated. The weighted modal coefficient matrices are superimposed along the row direction to obtain the comprehensive quality feature vector; The comprehensive quality feature vector includes equipment parameter index, finished casting parameter index, order delivery cycle index, and material logistics information index; The projection formula is: ； The covariance matrix is ​​expressed as follows: ； The formula for calculating the weighting coefficients is as follows: ； Where X is the original spatiotemporal snapshot, Y represents the dominant mode, and Y is the modal coefficient matrix obtained by dimensionality reduction. For the first A snapshot, Let C be the mean field and C be the covariance matrix. Modal weights; The reinforcement learning method is used to dynamically adjust the production scheduling strategy in real time according to the type of disturbance, and includes the following steps: The state space is defined according to the type of disturbance. The state space includes equipment status, quality anomalies, order queue, and resource constraints. Equipment status includes equipment operating / fault status. Quality anomalies include casting composition deviation and inclusion content. Order queue includes order priority, delivery date, and process path. Resource constraints include mold availability, raw material inventory, and energy supply. Based on the state space, adjustment strategies are set; adjustment strategies include sequential adjustment, equipment switching, resource reallocation, and process parameter optimization; sequential adjustment includes inserting / removing orders and reordering priorities; equipment switching includes migrating tasks from faulty equipment to standby equipment; resource reallocation includes mold allocation, personnel shift adjustment, and dynamic energy scheduling; process parameter optimization includes adjusting melting temperature and cooling rate. Define the action space: Output the adjustment strategy through the action space; Based on the state space and action space, incentive functions are defined to guide the production scheduling system to learn the optimal scheduling strategy under disturbance conditions. These incentive functions quantify penalty and reward mechanisms and include on-time delivery rate. Resource utilization rate: Quality pass rate: Cost of breach of contract: Equipment idling losses: Quality rework cost: ; in, Indicates the actual completion time. Indicates the planned delivery time. Actual production capacity For the theoretical maximum production capacity, For the number of qualified products, For total output, For order cancellation fees, For delayed compensation, For equipment idle time, Energy cost per unit time For the number of defective products, This refers to the cost of reworking a single item.

2. The casting scheduling method based on digital twins according to claim 1, characterized in that: The dynamic perturbation determination of production scheduling behavior includes: Set constraints: Constraints include the range of equipment index when the equipment is operating normally, the range of expected parameters index of finished castings, the range of normal order delivery cycle index, and the range of material flow status index. Disturbance type determination: Equipment malfunction disturbances are determined based on the range of equipment parameter indices and equipment indices during normal operation; casting quality abnormalities are determined based on the range of casting finished product parameter indices and casting finished product expected parameter indices; urgent orders are determined based on the range of order delivery cycle indices and order normal delivery cycle indices; and order cancellations are determined based on the range of material logistics information indices and material flow status indices.

3. The casting scheduling method based on digital twins according to claim 1, characterized in that: The reinforcement learning also includes the following steps: When equipment fails, the state space adjusts the equipment state to equipment failure according to the disturbance type. The adjustment strategy provides the failure equipment task migration to standby equipment and process parameter optimization strategy according to the equipment failure type. It actively responds through the action space and imports the parameter information after the action space response into the excitation function calculation. Resource utilization encourages the recovery of production capacity after migration, and equipment idling loss penalizes the original equipment idling loss. When a quality anomaly occurs, the state space will adjust the quality anomaly to casting composition deviation or inclusion content according to the disturbance type. The adjustment strategy provides melting temperature adjustment or cooling rate adjustment strategy according to the quality anomaly type, and actively responds through the action space. The parameter information after the action space response is imported into the excitation function calculation. The quality pass rate is ensured, and the quality rework cost is suppressed to prevent rework caused by excessive adjustment. When an emergency order is inserted or cancelled, the state space adjusts the order queue or resource constraints according to the disturbance type. The adjustment strategy provides a reordering priority or order removal strategy according to the emergency order insertion / cancellation type. The action space actively responds and imports the parameter information after the action space response into the incentive function calculation. The on-time delivery rate incentivizes the timely completion of emergency orders, and the default cost constrains the penalty cost for order cancellation.

4. The casting scheduling method based on digital twins according to claim 1, characterized in that: The multi-objective optimization algorithm is used to accurately generate a globally optimal production scheduling method, including: Population initialization: Using three-dimensional matrix encoding, process, equipment, and time parameters after disturbance elimination or without disturbance intervention are randomly generated into chromosomes based on the three-dimensional matrix encoding to form the initial population of the algorithm. The initial population represents the initial casting production scheduling scheme; Fitness calculation: The fitness function is calculated for each chromosome; Selection: Chromosomes are selected from the initial population based on the fitness function to generate the next generation of chromosomes; Crossover: Generates new chromosomes by selecting chromosomes using devices and time series; Mutation: Randomly adjusting processes, equipment, or time controls to generate new chromosomes; Iteration: Repeat the selection, crossover, and mutation operations until the maximum number of iterations is reached or the fitness no longer increases significantly; Local optimum determination: If the fitness variance of 90% of the individuals in the population is less than 0.1, then the population is determined to be trapped in a local optimum. After getting stuck in a local optimum, simulated annealing is performed to expand the search space and conduct a fine search for a better solution in the local neighborhood. Output the global Pareto front solution set, optimize equipment load allocation based on specific scenario requirements, including shortest production cycle, lowest equipment energy consumption, or least material loss, and obtain the global optimal solution; The time parameters include the time consumption and time constraints of each process on the equipment; The fitness function is used to quantify the quality of chromosomes and measure the overall performance of the production scheduling scheme.

5. The casting scheduling method based on digital twins according to claim 4, characterized in that: The simulated annealing operation includes: Select the top 20% of individuals with the highest fitness in the current population for local optimization; fitness function Convert to energy function ,and , This represents the maximum fitness of the current population. Set the initial temperature based on the current population energy fluctuations. Where N is the number of iterations to stop. For average energy, For the first Individual energy; Neighborhood generation using Gaussian perturbation ,in , With a mean of 0 and a variance of The normal distribution and These are the neighborhood solutions before and after Gaussian perturbation processing, respectively. Let L be the cooling rate, and L be the number of annealing cycles at each temperature. This is the initial step size; The Metropolis criterion accepts inferior solutions, but not those generated in the neighborhood. Calculate energy changes ,when Accept directly Replace the current solution; when With probability accept ; Set a threshold for the energy decrease; if the energy decrease is less than the threshold, accelerate the cooling process. ,otherwise ,in, and These are the temperature values ​​after the k-th and (k+1)-th iterations, respectively; Until the temperature drops to Or, if the maximum number of iterations is reached, the simulated annealing will exit.

6. The casting scheduling method based on digital twins according to claim 4, characterized in that: The optimized device load allocation includes: The equipment load adjustment formula is generated by weighted summation: ； in, These are all weighting coefficients, with values ​​ranging from 0 to 1, and , For the total production cycle, This represents the total energy consumption of the equipment. This refers to material cost losses.

7. A casting scheduling system based on digital twins, executing the casting scheduling method as described in any one of claims 1-6, characterized in that, It includes a physical entity layer, a digital twin layer, and a production scheduling optimization layer; The physical entity layer includes melting equipment, casting equipment, molding equipment, and cooling equipment, used to perform metal casting production; The digital twin layer includes a three-dimensional visualization model for real-time mapping of physical layer equipment status, process parameters, and resource data; The production scheduling optimization layer includes a dynamic disturbance judgment unit, a reinforcement learning adjustment unit, a multi-objective optimization unit, and a human-computer interaction interface, which are used to accurately generate the globally optimal production scheduling method and actively respond to dynamic disturbance events. The dynamic disturbance determination unit is used to determine the disturbance type based on the constraint conditions. The reinforcement learning adjustment unit dynamically adjusts the production scheduling strategy in real time based on the judgment result of the dynamic perturbation unit. The multi-objective optimization unit takes the parameters after disturbance elimination or without disturbance intervention as input, and uses a multi-objective optimization algorithm to accurately generate the globally optimal production scheduling method, while optimizing equipment load allocation.

8. The casting scheduling system based on digital twins according to claim 7, characterized in that, The human-computer interaction interface includes a production scheduling Gantt chart and a multi-dimensional indicator dashboard; The production scheduling Gantt chart is used for process drag-and-drop adjustments and version management, and records reinforcement learning adjustment traces. The multi-dimensional indicator dashboard is used to display equipment utilization, energy consumption curves, and dynamic disturbances in real time.

Images