Digital-twin-based online optimization method and platform for complex equipment manufacturing process

Abstract

The present application relates to a kind of complex equipment manufacturing process online optimization method and platform based on digital twinning, belong to digital twinning, manufacturing execution and process optimization control technical field.The platform of the present application includes field data access module, twin model management module, state synchronization module, disturbance identification module, candidate strategy generation module, virtual commissioning module, multi-objective evaluation module, strategy backwrite module, execution monitoring module and model evolution module.The present application is unified modeling to equipment, process, logistics, quality and resource state, identifies beat shift, buffer congestion, insert single, repair backflow and quality fluctuation etc.perturbation, generates and commissioning multiple candidate process strategies in virtual space, and carries out comprehensive evaluation, filters target strategy and safely backwrite field execution system, then continuously corrects twin model according to execution result, realizes the online optimization, closed-loop control and rolling evolution of complex equipment manufacturing process.

Description

Technical Field

[0001] This invention belongs to the field of digital twin, manufacturing execution and process optimization control technology, specifically relating to an online optimization method and platform for complex equipment manufacturing processes based on digital twins. Background Technology

[0002] Complex equipment manufacturing processes are typically characterized by long process chains, strong resource coupling, strict cycle time constraints, complex quality transfer relationships, and frequent on-site disturbances. In actual production, offline-developed process formulas, work step sequences, buffer strategies, and resource allocation plans are affected by factors such as equipment wear and tear, logistics congestion, material fluctuations, order insertions, rework and return, and personnel differences, causing them to deviate rapidly from the original optimal state during the execution phase.

[0003] Existing manufacturing systems typically handle process planning in the design phase, production control in the execution phase, and quality evaluation in the results phase across different systems. This leads to process optimization largely relying on manual experience revisions, local parameter adjustments, or post-event reviews. Although some production lines have deployed digital twin systems, most remain at the stage of status visualization, operation monitoring, and offline simulation, and have not yet formed an online optimization closed loop based on digital twins.

[0004] Existing technologies have at least the following shortcomings: First, twin models tend to focus on equipment state mapping or 3D visualization, lacking unified modeling of process, logistics, quality, and resource constraints; second, many systems can only reflect what happened on-site, but cannot answer how to adjust the process next; third, existing optimization solutions often focus on single equipment or single indicators, making it difficult to take into account multiple objective constraints such as cycle time, quality, energy consumption, work-in-process, and equipment load; fourth, optimization results often remain at the suggestion level and cannot be safely written back to the manufacturing execution level to form a closed loop; fifth, twin models lack continuous back-indexing and rolling evolution mechanisms, and the model accuracy gradually decreases over time.

[0005] Therefore, a new technical solution is urgently needed that can synchronize the field status in real time during the manufacturing execution phase, identify disturbance events, generate multiple candidate process optimization strategies, complete trial operation and risk assessment in a virtual environment, write the verified target strategy back to the field execution layer, and continuously revise the twin model based on the execution results, thereby realizing online optimization and closed-loop control of complex equipment manufacturing processes. Summary of the Invention

[0006] (a) Technical problems to be solved The technical problem this invention aims to solve is how to provide an online optimization method and platform for complex equipment manufacturing processes based on digital twins, in order to address the issue that while some existing production lines have deployed digital twin systems, most are still at the stage of status visualization, operation monitoring, and offline simulation, and have not yet formed an online optimization closed loop based on digital twins.

[0007] (II) Technical Solution To address the aforementioned technical problems, this invention proposes an online optimization method for complex equipment manufacturing processes based on digital twins. The method is characterized by including: real-time data acquisition, digital twin model construction and synchronization, disturbance identification, candidate strategy generation, virtual trial operation, multi-objective evaluation, strategy write-back, and execution feedback-driven model evolution. Real-time data acquisition is used to establish a unified data view around equipment objects, process objects, logistics objects, quality objects, and resource objects; the data includes equipment cycle time, equipment load, operating status, process formula number, process completion time, work-in-process location, buffer occupancy, test results, energy consumption information, alarm codes, rework and return events, and order insertion events. The construction and synchronization of digital twin models involves first uniformly encoding equipment, processes, logistics, quality, and resources in the physical manufacturing site; secondly, extracting equipment capacity parameters, process execution parameters, logistics flow parameters, quality judgment parameters, and resource occupancy parameters according to the decomposition logic of "physical entity—functional semantics—state parameters—constraint rules—relationship edges"; then establishing the association relationships between equipment twins, process twins, logistics twins, quality twins, and resource twins based on the same object encoding; and finally, writing real-time collected data into the state layer of each twin through the state synchronization module, and maintaining process sequence, resource occupancy, logistics flow direction, and quality transfer path in the relationship layer. Disturbance identification includes explicit disturbance identification and implicit disturbance identification. Explicit disturbance identification is used to identify direct events such as equipment failure, cycle time drift, buffer congestion, order insertion, rework return, and quality fluctuations. Implicit disturbance identification is used to identify state combinations that have not yet triggered alarms but have shown a trend of cumulative deviation. Candidate strategy generation is used to generate multiple executable strategies within the constraint boundary. The strategies include one or more combinations of process parameter adjustment strategies, process step sequence adjustment strategies, cycle time coordination strategies, buffer allocation strategies, equipment diversion strategies, material flow path adjustment strategies, and detection resource switching strategies. Virtual trial operation is used to quickly simulate various candidate strategies in a twin environment to obtain indicators such as cycle time, work-in-process, quality fluctuation, equipment load, changeover loss and energy consumption. Multi-objective evaluation is used to rank candidate strategies based on a comprehensive objective function; Policy writeback is used to write the target policy, after passing security boundary verification, process discipline verification, and permission verification, back to the manufacturing execution layer. Model evolution is used to update the parameters, weights, and disturbance factors in the digital twin model based on changes in cycle time, quality results, energy consumption, resource utilization, and deviation accumulation after the execution of the target strategy, so that the twin model can maintain a high degree of consistency with the physical site in the long term.

[0008] This invention also provides an online optimization platform for complex equipment manufacturing processes based on digital twins. The platform includes: a field data access module, a twin model management module, a status synchronization module, a disturbance identification module, a candidate strategy generation module, a virtual trial operation module, a multi-objective evaluation module, a strategy write-back module, an execution monitoring module, and a model evolution module. The field data access module connects to the equipment controller, manufacturing execution system, scheduling system, quality inspection system, energy acquisition system, and logistics tracking system to collect manufacturing field data in real time. The on-site data access module is used to establish a unified master index according to products, processes, equipment, logistics units and resource units. For the same manufacturing event, the platform simultaneously records the event start time, event end time, object identifier, process context, quality context and resource context. The twin model management module is used to build equipment twins, process twins, logistics twins, quality twins, and resource twins. The state synchronization module is used to map the collected field data to the corresponding twin in real time, forming a state vector Xt={Xdev, Xproc, Xlog, Xqua, Xres}. When the state of the field object changes, the state synchronization module not only updates the current state, but also records the difference before and after the state change and the impact path. The disturbance identification module is used to perform rule judgment and deviation detection on the state vector; The candidate policy generation module is used to generate multiple policy objects Uk based on the disturbance type, disturbance intensity, and boundary constraints; The virtual trial operation module is used to perform rapid simulations of various candidate strategies in a twin environment; The multi-objective evaluation module is used to score candidate strategies based on a comprehensive objective function; The strategy write-back module is used to write the target optimization strategy back to the manufacturing execution layer; The execution monitoring module is used to monitor the optimization effect in real time after the target optimization strategy is implemented, and to compare the changes in cycle time, quality, energy consumption and work-in-process before and after optimization. The model evolution module is used to back-iterate the twin model based on the execution monitoring results. (III) Beneficial Effects This invention proposes an online optimization method and platform for complex equipment manufacturing processes based on digital twins. Compared with existing technologies, this invention has at least the following advantages: First, a unified digital twin model oriented towards process, equipment, logistics, quality, and resource constraints was constructed, elevating the digital twin from the "display layer" to the "decision layer." This invention establishes a twin through multiple dimensions, enabling a more comprehensive and accurate simulation of the state changes of real physical equipment. Second, it can identify disturbances and generate multiple candidate strategies during the manufacturing execution phase, and complete trial runs and screening in virtual space to reduce the risk of trial and error on site. Third, the optimization results are securely written back to the on-site execution layer, forming a true online closed loop; Fourth, continuously update the twin model based on execution feedback to enable the model to evolve in a rolling fashion; Fifth, it can simultaneously take into account multiple objectives such as cycle time, quality, energy consumption, work-in-process inventory, and equipment load, making it more suitable for complex equipment manufacturing scenarios. Attached Figure Description

[0009] Figure 1 This is a diagram of the overall platform architecture of the present invention; Figure 2 This is a flowchart of the method of the present invention; Figure 3 A mapping diagram of multi-layered digital twin models; Figure 4 A flowchart for evaluating and selecting candidate strategies; Figure 5 To optimize the relationship between closed loop and model evolution online. Detailed Implementation

[0010] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.

[0011] To address the aforementioned technical problems, this invention proposes an online optimization method and platform for complex equipment manufacturing processes based on digital twins. This method is driven by real-time data from the manufacturing site, uses a multi-layered digital twin model as its computational foundation, employs candidate strategy generation and virtual trial operation mechanisms as its optimization core, and concludes with strategy write-back and model evolution as its closed-loop endpoints.

[0012] To achieve the above objectives, the present invention adopts the following technical solution: an online optimization method for complex equipment manufacturing processes based on digital twins, comprising the following steps: real-time data acquisition, digital twin model construction and synchronization, disturbance identification, candidate strategy generation, virtual trial operation, multi-objective evaluation, strategy write-back, and execution feedback-driven model evolution.

[0013] Real-time data acquisition is not simply sampling the status of a single piece of equipment, but rather establishing a unified data view around equipment, process, logistics, quality, and resource objects. The data includes at least equipment cycle time, equipment load, operating status, process recipe number, process completion time, work-in-process location, buffer zone occupancy, test results, energy consumption information, alarm codes, rework / return events, and order insertion events.

[0014] The construction and synchronization of digital twin models involves first uniformly encoding equipment, processes, logistics, quality, and resources in the physical manufacturing site; secondly, extracting equipment capacity parameters, process execution parameters, logistics flow parameters, quality judgment parameters, and resource occupancy parameters according to the decomposition logic of "physical entity—functional semantics—state parameters—constraint rules—relationship edges"; then establishing the association relationships between equipment twins, process twins, logistics twins, quality twins, and resource twins based on the same object encoding; finally, writing real-time collected data into the state layer of each twin through the state synchronization module, and maintaining process sequence, resource occupancy, logistics flow direction, and quality transfer path in the relationship layer, thereby forming a multi-layered digital twin model that is computable, deductive, and rewritable. Preferably, the digital twin model consists of an object layer, a relationship layer, a constraint layer, and a state layer. The object layer describes the equipment twin, process twin, logistics twin, quality twin, and resource twin; the relationship layer describes the process sequence relationship, resource occupation relationship, logistics flow relationship, and quality transfer relationship; the constraint layer describes the safety boundary, process discipline boundary, authority boundary, and capacity boundary; and the state layer describes the real-time state and historical evolution state of each object.

[0015] Disturbance identification includes explicit disturbance identification and implicit disturbance identification. Explicit disturbance identification is used to identify direct events such as equipment failure, cycle time drift, buffer congestion, order insertion, rework return, and quality fluctuations. Implicit disturbance identification is used to identify state combinations that have not yet triggered alarms but have shown a trend of cumulative deviation, such as slow cycle time drift at critical workstations, edge quality indicators continuously approaching the warning boundary, and abnormal growth in logistics waiting time.

[0016] Candidate strategy generation does not simply output a single corrective action, but rather generates multiple executable strategies within the constraint boundaries. These strategies include one or more combinations of process parameter adjustment strategies, process step sequence adjustment strategies, cycle time coordination strategies, buffer allocation strategies, equipment diversion strategies, material flow path adjustment strategies, and detection resource switching strategies.

[0017] Virtual trial operation is used to quickly simulate various candidate strategies in a twin environment to obtain results indicators such as cycle time, work-in-process inventory, quality fluctuation, equipment load, changeover loss and energy consumption.

[0018] Multi-objective evaluation is used to rank candidate strategies based on a comprehensive objective function, wherein the comprehensive objective function can be expressed as: F(Uk)=a·Gcycle+b·Gquality-c·Lenergy-d·Lwip-e·Lload-f·Lchange Wherein, Uk represents the k-th candidate strategy, G represents the benefit term, L represents the loss term, and a, b, c, d, e, and f represent weighting coefficients. Specifically, Uk represents the k-th candidate optimization strategy; Gcycle represents the cycle time benefit term, used to characterize the degree of improvement in cycle time for the entire production line or key workstations after strategy implementation; Gquality represents the quality benefit term, used to characterize the degree of reduction in key quality risks or the degree of improvement in pass rate; Lenergy represents the energy consumption loss term, used to characterize the increase in energy consumption per unit product or per unit time; Lwip represents the work-in-process loss term, used to characterize the adverse effects of buffer backlog and increased work-in-process levels; Lload represents the equipment load loss term, used to characterize the degree of deterioration in the load of key equipment, key inspection positions, or key resources; Lchange represents the changeover loss term, used to characterize the additional changeover, switching, or recovery costs introduced by the strategy; a, b, c, d, e, and f represent the evaluation weighting coefficients corresponding to the above indicators, and can be dynamically adjusted according to manufacturing objectives.

[0019] Policy write-back is used to write the target policy, after passing security boundary verification, process discipline verification, and permission verification, back to the manufacturing execution layer. The manufacturing execution layer includes one or more of the following: manufacturing execution system, scheduling system, process recipe library, equipment recipe interface, PLC parameter area, and detection rule library.

[0020] Model evolution is used to update the parameters, weights, and disturbance factors in the digital twin model based on changes in cycle time, quality results, energy consumption, resource utilization, and deviation accumulation after the execution of the target strategy, so that the twin model can maintain a high degree of consistency with the physical site in the long term.

[0021] Compared with existing technologies, this invention has at least the following beneficial effects: First, it constructs a unified digital twin model oriented towards process, equipment, logistics, quality, and resource constraints, elevating the digital twin from the "display layer" to the "decision-making layer"; Second, it can identify disturbances and generate multiple candidate strategies during the manufacturing execution phase, completing trial runs and screening in virtual space, reducing the risk of trial and error on-site; Third, it securely writes the optimization results back to the on-site execution layer, forming a true online closed loop; Fourth, it continuously updates the twin model based on execution feedback, enabling the model to have rolling evolution capabilities; Fifth, it can simultaneously consider multiple objectives such as cycle time, quality, energy consumption, work-in-process, and equipment load, making it more suitable for complex equipment manufacturing scenarios.

[0022] Example 1: like Figure 1 As shown, the platform of this invention includes: a field data access module, a twin model management module, a state synchronization module, a disturbance identification module, a candidate strategy generation module, a virtual trial operation module, a multi-objective evaluation module, a strategy write-back module, an execution monitoring module, and a model evolution module. The field data access module connects to the equipment controller, manufacturing execution system, scheduling system, quality inspection system, energy acquisition system, and logistics tracking system, and is used to collect manufacturing field data in real time.

[0023] The on-site data access module is used to establish a unified master index based on products, processes, equipment, logistics units, and resource units. For the same manufacturing event, the platform simultaneously records the event start time, event end time, object identifier, process context, quality context, and resource context, avoiding inconsistencies in objects and timing errors between different platforms.

[0024] like Figure 3 As shown, the twin model management module is used to construct equipment twins, process twins, logistics twins, quality twins, and resource twins. The equipment twin represents equipment capacity parameters, real-time load, and health status; the process twin represents the sequence of work steps, process formulas, and parameter targets; the logistics twin represents the location of work-in-process, queue length, and path status; the quality twin represents key quality characteristics, inspection results, and quality risk levels; and the resource twin represents the occupancy status of resources such as fixtures, tooling, inspection stations, and personnel.

[0025] It should be noted that this does not involve mechanically creating five isolated twins for a single real physical device. Instead, it uses the same physical object or manufacturing event as the main index to generate corresponding twin representations from different analytical dimensions: an equipment twin is formed when representing the device's capabilities, health, and load; a process twin is formed when representing the sequence of steps, formulas, and parameter targets carried by the device; a logistics twin is formed when representing the relationship between the device and its buffer zones, logistics paths, and work-in-process locations; a quality twin is formed when representing the quality characteristics, risk levels, and judgment results corresponding to the processing or inspection results of the device; and a resource twin is formed when representing the occupancy relationships of fixtures, tooling, inspection stations, and personnel associated with the device. These multiple types of twins are not simply repetitive modeling, but rather semantic mapping and relational assembly oriented towards the same physical object across different constraint dimensions. This invention, by establishing twins through multiple dimensions, can more comprehensively and accurately simulate the state change process of real physical devices.

[0026] The state synchronization module maps the collected field data to the corresponding twin in real time, forming a state vector Xt={Xdev, Xproc, Xlog, Xqua, Xres}. When the state of a field object changes, the state synchronization module not only updates the current state but also records the difference before and after the state change and the impact path, providing a basis for subsequent disturbance identification and strategy evaluation.

[0027] Wherein, Xt represents the overall on-site status vector at time t; Xdev represents the equipment status sub-vector, which includes at least equipment cycle time, operating status, alarm code, load rate, and health status; Xproc represents the process status sub-vector, which includes at least the process step sequence, process recipe number, parameter target value, parameter measured value, and process completion time; Xlog represents the logistics status sub-vector, which includes at least the work-in-process location, buffer occupancy rate, queue length, path status, and estimated arrival time; Xqua represents the quality status sub-vector, which includes at least the key quality characteristic value, inspection conclusion, quality risk level, and warning boundary distance; and Xres represents the resource status sub-vector, which includes at least the fixture occupancy status, tooling occupancy status, inspection position occupancy status, personnel availability status, and resource switching time.

[0028] The difference before and after the state change can be expressed as ΔXt = Xt Xt 1, where ΔXt is used to characterize the overall state change at adjacent times, and the influence path is used to characterize the action link of a certain state change to other objects through process sequence relationship, material flow relationship or resource occupation relationship.

[0029] The disturbance identification module is used to perform rule judgment and deviation detection on the state vector. When the cycle time of critical workstation equipment deviates from the baseline cycle time multiple times in a row, the buffer occupancy rate exceeds the set threshold, rework parts are returned in a concentrated manner in a short period of time, a certain quality indicator continuously approaches the warning boundary, or the order insertion changes the established flow sequence, the disturbance identification module generates a disturbance event set Et and calculates the intensity, impact range and duration of each disturbance.

[0030] If the real-time cycle time Ct of a critical workstation satisfies |Ct-C0| / C0≥θ1 with the baseline cycle time C0, and this condition persists for more than a preset time window T1, then a cycle time drift disturbance is identified. If the buffer occupancy rate Bt exceeds the threshold θ2, and the waiting time exceeds the preset window T2, then a congestion disturbance is identified. If the critical quality indicator Qt falls into the warning zone for n consecutive samples, then a quality fluctuation disturbance is identified.

[0031] The candidate strategy generation module generates multiple strategy objects (Uk) based on disturbance type, disturbance intensity, and boundary constraints. Each strategy object includes at least a strategy identifier, strategy type, strategy parameters, applicable boundaries, and priority. Taking a cycle imbalance scenario as an example, it can generate strategies for fine-tuning local process parameters, coordinating cycle times between preceding and following workstations, reallocating buffer zones, diverting inspection tasks, and staggering logistics flow.

[0032] The candidate strategy generation module first calls the rule template library to output a set of basic strategies, and then combines the current twin state and resource constraints to expand the basic strategies to obtain multiple candidate strategy sets. Then, based on the process discipline boundary and safety boundary, inexecutable strategies are eliminated, and the set of strategies to be evaluated is retained.

[0033] The virtual trial operation module is used to perform rapid simulations of various candidate strategies in a twin environment. For parameter-level strategies, the virtual trial operation focuses on calculating cycle time changes, quality fluctuations, and equipment load changes; for sequence-level and scheduling-level strategies, the virtual trial operation focuses on calculating work-in-process changes, queuing changes, conflict risks, and resource utilization changes; for combined strategies, it calculates multiple of the above indicators simultaneously.

[0034] The multi-objective evaluation module is used to score candidate strategies based on a comprehensive objective function.

[0035] Preferably, the cycle time benefit term Gcycle is used to represent the degree to which the strategy improves the overall cycle time, the quality benefit term Gquality is used to represent the degree to which quality risk decreases, the energy loss term Lenergy is used to represent the increase in energy consumption, the work-in-process loss term Lwip is used to represent the increase in work-in-process, the load loss term Lload is used to represent the deterioration of the load on critical equipment, and the changeover loss term Lchange is used to represent the additional switching costs introduced by the strategy.

[0036] When a candidate strategy can significantly reduce buffer congestion but causes key quality characteristics to exceed the allowable boundary, the multi-objective evaluation module directly judges the strategy as unexecutable; when multiple strategies meet the boundary requirements, the system prioritizes the strategy with the highest comprehensive score and the lowest rollback cost.

[0037] The strategy write-back module is used to write the target optimization strategy back to the manufacturing execution layer. If the target optimization strategy is a parameter-level adjustment, it is written back to the process recipe library or equipment parameter interface; if the target optimization strategy is a sequence-level adjustment, it is written back to the process sequence control module in the manufacturing execution system; if it is a scheduling-level adjustment, it is written back to the scheduling system or logistics scheduling interface.

[0038] To avoid accidental writing and execution, the policy write-back module performs three types of checks before distribution: First, permission check, verifying whether the current user, system, and device have the permission to receive the policy; second, constraint check, verifying whether the policy exceeds security boundaries, process discipline boundaries, or resource capacity boundaries; and third, rollback plan check, verifying whether the policy can be restored to the baseline policy if the effect deteriorates after execution.

[0039] The execution monitoring module is used to monitor the optimization effect in real time after the target optimization strategy is implemented, and to compare the changes in cycle time, quality, energy consumption, and work-in-process before and after optimization. If a certain indicator is detected to be continuously deteriorating, or an out-of-bounds anomaly occurs, the system triggers a rollback mechanism to restore the execution layer to the baseline strategy or the previous stable strategy.

[0040] The model evolution module is used to recalibrate the twin model based on the execution monitoring results. If there is a stable deviation between the virtual trial run prediction and the field execution results, the twin model parameters are updated; if the impact of a certain type of disturbance in the field is stronger than the historical estimate, the corresponding disturbance impact factor is increased; if a certain type of objective is more important in actual production than expected, the weight coefficients in the comprehensive objective function are adjusted.

[0041] Example 2: Cycle time imbalance scenario. A complex equipment assembly line consists of pre-processing stations, key assembly stations, inspection stations, and transfer units. Due to wear and tear on equipment at the key assembly stations, the actual cycle time gradually exceeds the standard cycle time, leading to backlog in the pre-processing buffer and idle testing resources in the subsequent process. After identifying cycle time drift and congestion disturbances, the platform generates candidate strategies such as "redistribution of testing resources," "off-peak transfer," and "local cycle time coordination." After comparison through virtual trial runs, the system selects the combination strategy with the highest comprehensive score and executes it, reducing the peak occupancy of the buffer and restoring the overall line cycle time to stability.

[0042] Quality fluctuation scenario. A certain inspection process continuously produces edge-passing products. Although the final inspection does not reject them, the quality risk increases. The platform writes this event as a quality disturbance into the twin model, and generates multiple candidate strategies by combining upstream process formula changes, equipment status changes, and material batch information. Virtual trial operation revealed that simply raising the inspection threshold would significantly reduce production capacity, while the combined strategy of "fine-tuning the parameters of the preceding process + optimizing the inspection sampling strategy" balances quality and cycle time. Therefore, it was selected as the target optimization strategy and written back to the execution layer.

[0043] In scenarios involving order insertion and rework return flow, when high-priority orders enter the production line and rework parts are returned in a concentrated manner, the platform simultaneously injects order insertion and rework events into the twin environment, automatically generating strategies such as buffer reallocation, path switching, workstation priority adjustment, and inspection resource sharing. After multi-objective evaluation, the system outputs a combined strategy of "priority rearrangement + buffer allocation + path switching" to ensure that the order insertion meets delivery requirements while avoiding significant deterioration of the overall work-in-process level and equipment load.

[0044] Model rolling evolution scenario. If a certain strategy performs well in the twin environment, but the actual return in the physical field is lower than the predicted value, it indicates that the twin model has parameter drift or distortion of perturbation factors. At this time, the model evolution module updates the equipment capacity parameters, logistics delay parameters, quality risk weights, and perturbation factors based on the execution results, so that subsequent strategy evaluations are closer to the real field.

[0045] This invention is not limited to the embodiments described above. For machining units, flexible inspection units, automated transfer units, and other manufacturing scenarios with strong coupling of multiple processes, as long as real-time data acquisition, object mapping, candidate strategy calculation, and write-back execution interfaces can be established, the technical solution of this invention can be adopted.

[0046] Example 3: This invention discloses an online optimization method and platform for complex equipment manufacturing processes based on digital twins, relating to the fields of digital twins, manufacturing execution, and process optimization control. The platform includes a field data access module, a twin model management module, a status synchronization module, a disturbance identification module, a candidate strategy generation module, a virtual trial operation module, a multi-objective evaluation module, a strategy write-back module, an execution monitoring module, and a model evolution module. This invention achieves unified modeling of equipment, process, logistics, quality, and resource status, identifies disturbances such as cycle time drift, buffer congestion, order insertion, rework return, and quality fluctuations, generates and trials multiple candidate process strategies in a virtual space, comprehensively evaluates them according to indicators such as cycle time, quality, energy consumption, work-in-process, equipment load, and changeover losses, selects target strategies, and safely writes them back to the field execution system. Based on the execution results, the twin model is continuously corrected, realizing online optimization, closed-loop control, and rolling evolution of complex equipment manufacturing processes.

[0047] A method for online optimization of complex equipment manufacturing processes based on digital twins includes the following steps: S1. Collect real-time operational data from the manufacturing site and establish a site dataset for single products, single processes, and single resource objects; S2. Based on the aforementioned field dataset, construct equipment twins, process twins, logistics twins, quality twins, and resource twins, and form a digital twin model; S3. Based on the digital twin model, identify disturbances in the manufacturing site such as cycle time drift, equipment failure, buffer congestion, order insertion, rework return, material fluctuation and quality anomaly, and obtain a set of disturbance events; S4. Under the constraints of process boundary, resource boundary and safety boundary, generate multiple candidate optimization strategies based on the set of disturbance events; S5. Conduct virtual trial runs of each candidate optimization strategy in a digital twin environment to obtain the corresponding cycle time index, quality index, energy consumption index, work-in-process index, equipment load index and risk index. S6. Select target optimization strategies based on multi-objective evaluation results, and write the target optimization strategies back to the manufacturing execution layer; S7. Monitor the execution effect of the target optimization strategy, and perform rolling calibration and evolutionary updates of the digital twin model based on the execution feedback.

[0048] Furthermore, the real-time operational data includes at least equipment operation data, process execution data, logistics status data, quality inspection data, resource usage data, energy consumption data, and abnormal event data; among which, equipment operation data includes equipment cycle time, operating status, alarm codes, and health status; process execution data includes process formula number, process step sequence, target parameter value, and measured parameter value; and logistics status data includes in-process location, buffer occupancy rate, path status, and estimated arrival time.

[0049] Furthermore, the digital twin model includes an object layer, a relationship layer, a constraint layer, and a state layer; the object layer is used to describe equipment objects, process objects, logistics objects, quality objects, and resource objects; the relationship layer is used to describe process sequence relationships, resource occupation relationships, logistics flow relationships, and quality transfer relationships; the constraint layer is used to describe safety boundaries, process discipline boundaries, access boundaries, and resource capacity boundaries; and the state layer is used to describe the real-time operating status of the objects and their historical evolution status.

[0050] Furthermore, the disturbance identification includes explicit disturbance identification based on rule thresholds and implicit disturbance identification based on state deviations; when the deviation between the equipment cycle time and the reference cycle time exceeds a set threshold, the buffer occupancy rate exceeds a set threshold, the key quality indicators are close to the warning boundary, or the order insertion and rework return occur simultaneously, the generation of candidate optimization strategies is triggered.

[0051] Furthermore, the candidate optimization strategies include at least one or more of the following: process parameter adjustment strategy, process step sequence adjustment strategy, cycle time coordination strategy, buffer allocation strategy, equipment diversion strategy, material flow path adjustment strategy, and detection resource switching strategy; and each candidate optimization strategy is accompanied by applicable boundaries, priority, and rollback conditions.

[0052] Furthermore, the multi-objective evaluation uses a comprehensive objective function F(Uk) to score the k-th candidate optimization strategy. The comprehensive objective function includes at least a cycle time gain item, a quality gain item, an energy consumption loss item, a work-in-process loss item, an equipment load loss item, and a changeover loss item. When a candidate optimization strategy violates the safety boundary or process discipline boundary, it is directly determined to be an unexecutable strategy.

[0053] Furthermore, writing the target optimization strategy back to the manufacturing execution layer includes: writing the process parameter-level optimization results back to the process recipe library, writing the sequence-level optimization results back to the manufacturing execution system, writing the scheduling-level optimization results back to the scheduling system, and performing permission verification, constraint verification, and rollback contingency plan verification before writing back; when the effect deteriorates or amplifies abnormally after the target optimization strategy is executed, the rollback mechanism is activated to restore the baseline strategy.

[0054] Furthermore, the rolling calibration and evolution update includes updating the object parameters, relationship parameters, disturbance influence factors, and evaluation weights in the digital twin model based on the execution results of the target optimization strategy, subsequent quality results, and deviation accumulation, so that the digital twin model continuously approximates the physical manufacturing site.

[0055] This invention also provides an online optimization platform for complex equipment manufacturing processes based on digital twins, including a field data access module, a twin model management module, a status synchronization module, a disturbance identification module, a candidate strategy generation module, a virtual trial operation module, a multi-objective evaluation module, a strategy write-back module, an execution monitoring module, and a model evolution module. The field data access module is used to collect real-time operational data from the manufacturing site. The twin model management module is used to construct equipment twins, process twins, logistics twins, quality twins, and resource twins. The disturbance identification module is used to identify disturbance events that affect the stable operation of the manufacturing process. The candidate strategy generation module, multi-objective evaluation module, and strategy write-back module are used to generate, screen, and execute candidate optimization strategies in a closed loop. The model evolution module is used to continuously update the digital twin model based on the execution results.

[0056] Beneficial effects: Compared with the prior art, the present invention has at least the following beneficial effects: First, a unified digital twin model oriented towards process, equipment, logistics, quality, and resource constraints was constructed, elevating the digital twin from the "display layer" to the "decision layer." This invention establishes a twin through multiple dimensions, enabling a more comprehensive and accurate simulation of the state changes of real physical equipment. Second, it can identify disturbances and generate multiple candidate strategies during the manufacturing execution phase, and complete trial runs and screening in virtual space to reduce the risk of trial and error on site. Third, the optimization results are securely written back to the on-site execution layer, forming a true online closed loop; Fourth, continuously update the twin model based on execution feedback to enable the model to evolve in a rolling fashion; Fifth, it can simultaneously take into account multiple objectives such as cycle time, quality, energy consumption, work-in-process inventory, and equipment load, making it more suitable for complex equipment manufacturing scenarios.

[0057] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for online optimization of complex equipment manufacturing processes based on digital twins, characterized in that, The method includes: real-time data acquisition, digital twin model construction and synchronization, perturbation identification, candidate policy generation, virtual trial operation, multi-objective evaluation, policy write-back, and execution feedback-driven model evolution; Real-time data acquisition is used to establish a unified data view around equipment objects, process objects, logistics objects, quality objects, and resource objects; the data includes equipment cycle time, equipment load, operating status, process formula number, process completion time, work-in-process location, buffer occupancy, test results, energy consumption information, alarm codes, rework and return events, and order insertion events. The construction and synchronization of digital twin models involves first uniformly encoding equipment, processes, logistics, quality, and resources in the physical manufacturing site; secondly, extracting equipment capacity parameters, process execution parameters, logistics flow parameters, quality judgment parameters, and resource occupancy parameters according to the decomposition logic of "physical entity—functional semantics—state parameters—constraint rules—relationship edges"; then establishing the association relationships between equipment twins, process twins, logistics twins, quality twins, and resource twins based on the same object encoding; and finally, writing real-time collected data into the state layer of each twin through the state synchronization module, and maintaining process sequence, resource occupancy, logistics flow direction, and quality transfer path in the relationship layer. Disturbance identification includes explicit disturbance identification and implicit disturbance identification. Explicit disturbance identification is used to identify direct events such as equipment failure, cycle time drift, buffer congestion, order insertion, rework return, and quality fluctuations. Implicit disturbance identification is used to identify state combinations that have not yet triggered alarms but have shown a trend of cumulative deviation. Candidate strategy generation is used to generate multiple executable strategies within the constraint boundary. The strategies include one or more combinations of process parameter adjustment strategies, process step sequence adjustment strategies, cycle time coordination strategies, buffer allocation strategies, equipment diversion strategies, material flow path adjustment strategies, and detection resource switching strategies. Virtual trial operation is used to quickly simulate various candidate strategies in a twin environment to obtain indicators such as cycle time, work-in-process, quality fluctuation, equipment load, changeover loss and energy consumption. Multi-objective evaluation is used to rank candidate strategies based on a comprehensive objective function; Policy writeback is used to write the target policy, after passing security boundary verification, process discipline verification, and permission verification, back to the manufacturing execution layer. Model evolution is used to update the parameters, weights, and disturbance factors in the digital twin model based on changes in cycle time, quality results, energy consumption, resource utilization, and deviation accumulation after the execution of the target strategy, so that the twin model can maintain a high degree of consistency with the physical site in the long term.

2. The online optimization method for complex equipment manufacturing processes based on digital twins as described in claim 1, characterized in that, The digital twin model consists of an object layer, a relationship layer, a constraint layer, and a state layer. The object layer describes the equipment twin, process twin, logistics twin, quality twin, and resource twin. The relationship layer describes the process sequence relationship, resource occupation relationship, logistics flow relationship, and quality transfer relationship. The constraint layer describes the safety boundary, process discipline boundary, authority boundary, and capacity boundary. The state layer describes the real-time state and historical evolution state of each object.

3. The online optimization method for complex equipment manufacturing processes based on digital twins as described in claim 2, characterized in that, Equipment twins are used to represent equipment capacity parameters, real-time load, and health status; process twins are used to represent process steps, process formulas, and parameter targets; and logistics twins are used to represent work-in-process locations, queue lengths, and path status. Quality twins are used to express key quality characteristics, test results, and quality risk levels; resource twins are used to express the occupancy status of fixtures, tooling, testing stations, and personnel.

4. The online optimization method for complex equipment manufacturing processes based on digital twins as described in claim 1, characterized in that, The state synchronization module is used to map the collected field data to the corresponding twin in real time, forming a state vector Xt={Xdev, Xproc, Xlog, Xqua, Xres}. When the state of the field object changes, the state synchronization module not only updates the current state, but also records the difference before and after the state change and the impact path, providing a basis for subsequent disturbance identification and strategy evaluation. Wherein, Xt represents the overall on-site status vector at time t; Xdev represents the equipment status sub-vector, including equipment cycle time, operating status, alarm code, load rate, and health; Xproc represents the process status sub-vector, including process step sequence, process recipe number, parameter target value, parameter measured value, and process completion time; Xlog represents the logistics status sub-vector, including in-process location, buffer occupancy rate, queue length, path status, and estimated arrival time; Xqua represents the quality status sub-vector, including key quality characteristic values, inspection conclusions, quality risk level, and warning boundary distance; Xres represents the resource status sub-vector, including fixture occupancy status, tooling occupancy status, inspection position occupancy status, personnel availability status, and resource switching time; The difference before and after the state change is expressed as ΔXt = Xt Xt 1, where ΔXt is used to characterize the overall state change at adjacent times, and the influence path is used to characterize the action link of a certain state change to other objects through process sequence relationship, material flow relationship or resource occupation relationship.

5. The online optimization method for complex equipment manufacturing processes based on digital twins as described in claim 1, characterized in that, If the real-time cycle time Ct of a critical workstation satisfies |Ct-C0| / C0≥θ1 with the baseline cycle time C0, and this condition persists for more than a preset time window T1, then a cycle time drift disturbance is identified. If the buffer occupancy rate Bt exceeds the threshold θ2, and the waiting time exceeds the preset window T2, then a congestion disturbance is identified. If the critical quality indicator Qt falls into the warning zone for n consecutive samples, then a quality fluctuation disturbance is identified.

6. The online optimization method for complex equipment manufacturing processes based on digital twins as described in claim 1, characterized in that, The candidate strategy generation module first calls the rule template library to output a set of basic strategies, and then combines the current twin state and resource constraints to expand the basic strategies with variants to obtain multiple candidate strategy sets; then, based on the process discipline boundary and safety boundary, it eliminates unexecutable strategies and retains the set of strategies to be evaluated.

7. The online optimization method for complex equipment manufacturing processes based on digital twins as described in claim 1, characterized in that, During virtual trial operation, rapid simulations of each candidate strategy are performed in a twin environment. For parameter-level strategies, virtual trial operation focuses on calculating cycle time changes, quality fluctuations, and equipment load changes. For sequence-level and scheduling-level strategies, virtual trial operation focuses on calculating work-in-progress changes, queuing changes, conflict risks, and resource usage changes. For combined strategies, multiple indicators mentioned above are calculated simultaneously.

8. The online optimization method for complex equipment manufacturing processes based on digital twins as described in claim 1, characterized in that, The comprehensive objective function is expressed as follows: F(Uk)=a·Gcycle+b·Gquality-c·Lenergy-d·Lwip-e·Lload-f·Lchange Wherein, Uk represents the k-th candidate optimization strategy; Gcycle represents the cycle time benefit, used to characterize the degree of improvement in cycle time of the entire line or key workstations after the strategy is implemented; Gquality represents the quality benefit, used to characterize the degree of reduction in key quality risks or the degree of improvement in pass rate; Lenergy represents the energy consumption loss, used to characterize the increase in energy consumption per unit product or per unit time; Lwip represents the work-in-process loss, used to characterize the adverse effects of buffer backlog and the increase in work-in-process level; Lload represents the equipment load loss, used to characterize the degree of deterioration of the load of key equipment, key inspection positions or key resources; Lchange represents the changeover loss, used to characterize the additional changeover, switching or recovery costs introduced by the strategy; a, b, c, d, e, f represent the evaluation weight coefficients corresponding to the above indicators, and can be dynamically adjusted according to manufacturing goals.

9. The online optimization method for complex equipment manufacturing processes based on digital twins as described in claim 8, characterized in that, If the target optimization strategy is a parameter-level adjustment, it is written back to the process recipe library or equipment parameter interface; if the target optimization strategy is a sequence-level adjustment, it is written back to the process sequence control module in the manufacturing execution system. If it is a scheduling-level adjustment, then write it back to the scheduling system or logistics scheduling interface; The strategy performs three types of checks before being issued: Permission verification verifies whether the current user, system, and device have the permission to receive this policy; Constraint verification verifies whether the strategy exceeds safety boundaries, process discipline boundaries, or resource capacity boundaries; rollback contingency plan verification verifies whether the strategy can be restored to the baseline strategy if the effect deteriorates after execution.

10. A digital twin-based online optimization platform for complex equipment manufacturing processes, implementing the method of any one of claims 1-9, characterized in that, The platform includes: a field data access module, a twin model management module, a status synchronization module, a disturbance identification module, a candidate strategy generation module, a virtual trial operation module, a multi-objective evaluation module, a strategy write-back module, an execution monitoring module, and a model evolution module; the field data access module connects to the equipment controller, manufacturing execution system, scheduling system, quality inspection system, energy acquisition system, and logistics tracking system to collect manufacturing field data in real time; The on-site data access module is used to establish a unified master index according to products, processes, equipment, logistics units and resource units. For the same manufacturing event, the platform simultaneously records the event start time, event end time, object identifier, process context, quality context and resource context. The twin model management module is used to build equipment twins, process twins, logistics twins, quality twins, and resource twins. The state synchronization module is used to map the collected field data to the corresponding twin in real time, forming a state vector Xt={Xdev, Xproc, Xlog, Xqua, Xres}. When the state of the field object changes, the state synchronization module not only updates the current state, but also records the difference before and after the state change and the impact path. The disturbance identification module is used to perform rule judgment and deviation detection on the state vector; The candidate policy generation module is used to generate multiple policy objects Uk based on the disturbance type, disturbance intensity, and boundary constraints; The virtual trial operation module is used to perform rapid simulations of various candidate strategies in a twin environment; The multi-objective evaluation module is used to score candidate strategies based on a comprehensive objective function; The strategy write-back module is used to write the target optimization strategy back to the manufacturing execution layer; The execution monitoring module is used to monitor the optimization effect in real time after the target optimization strategy is implemented, and to compare the changes in cycle time, quality, energy consumption and work-in-process before and after optimization. The model evolution module is used to back-iterate the twin model based on the execution monitoring results.

Images