Collinear production cross-process variable quality tuning algorithm recommendation method and twin verification system

Abstract

The application discloses a kind of collinear production cross-process variable condition quality tuning algorithm recommendation method and twinborn verification system, belong to intelligent manufacturing and digital twinborn technical field.The method includes: data aggregation and condition perception step, real-time acquisition production whole-link data and build the condition feature vector of the current task;Algorithm intelligent recommendation step, similar historical condition is retrieved in case knowledge base based on the feature vector, the performance of each candidate algorithm is predicted using meta-learning model, and the recommended result is generated through multi-objective decision;Digital twinborn optimization verification step, the cross-process quality model chain is instantiated in digital twinborn environment, and the recommended algorithm is driven to carry out iterative optimization, to solve the optimal process parameter set;Scheme confirmation and entity execution step, the optimized scheme after verification is issued to physical workshop execution;Effect feedback and closed-loop learning step, actual production data are collected to form new case feedback to system, to realize continuous evolution.

Description

Technical Field

[0001] This invention relates to the fields of intelligent manufacturing, industrial big data and digital twin technology, and in particular to a method for recommending a quality optimization algorithm for cross-process variable operating conditions in co-line production and a twin verification system. Background Technology

[0002] With the deepening of intelligent manufacturing, digital twin technology has become an important means to build intelligent workshops that integrate virtual and physical environments and enable real-time interaction. Especially against the backdrop of increasingly prevalent multi-variety, small-batch, and customized production models, co-line production—that is, flexible organization and scheduling of production for different products or process paths on the same production line—has become a key mode for improving manufacturing resource utilization and response speed. However, while this production mode brings flexibility, it also makes quality optimization more complex, mainly in the following aspects.

[0003] In co-line production scenarios, products from the same batch or different batches may undergo different process routes. The final product quality is not determined by a single process step, but is influenced by the process parameters of multiple preceding and following processes, resulting in complex nonlinear coupling relationships between these processes. Traditional single-point quality optimization methods struggle to handle this cross-process global optimization problem.

[0004] In co-line production scenarios, frequent changes in production orders lead to dynamic changes in processing objects, equipment status, and environmental conditions. Optimization algorithms that perform well under specific conditions may drastically decrease in effectiveness under other conditions. Currently, shop floor engineers mainly rely on experience or trial-and-error methods to find suitable quality optimization algorithms (such as neural networks, support vector machines, and genetic algorithms), which is inefficient and makes it difficult to guarantee optimality.

[0005] Current applications of digital twin systems at the workshop level primarily focus on status monitoring, process visualization, and single-process simulation, lacking the ability to support overall quality optimization decisions for multiple processes and varying operating conditions. These systems often employ pre-set optimization algorithms or models, failing to dynamically select and recommend the most suitable optimization strategy based on real-time operating conditions. This results in insufficient adaptability and limited intelligence of digital twins in quality closed-loop control.

[0006] Therefore, in the complex manufacturing environment where co-line production, cross-process chains, and changing working conditions are intertwined, there is an urgent need for a system and method that can perceive the status of the entire production chain in real time, dynamically identify working condition characteristics, intelligently recommend and verify optimization algorithms, and ultimately achieve continuous self-learning, so as to break through the decision-making bottleneck of existing digital twin workshops in the quality optimization process and truly achieve accurate, efficient, and adaptive quality control. Summary of the Invention

[0007] This invention aims to address the technical problem of the lack of adaptive and intelligent quality optimization algorithm recommendation capabilities in existing co-line production environments, particularly under cross-process coupling and variable operating conditions in digital twin workshops. It provides a method for recommending quality optimization algorithms across processes and operating conditions in co-line production, along with a twin verification system. This system can automatically recommend the most suitable quality optimization algorithm based on real-time production data and dynamic operating conditions, thereby improving the accuracy, automation level, and universality of quality optimization.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] On the one hand, this invention provides a method for quality optimization algorithm recommendation in co-line production with varying process conditions, comprising the following steps:

[0010] Step 1: Data Aggregation and Operating Condition Awareness. Real-time access to order data from the Manufacturing Execution System (MES), cross-process parameters of work-in-process, equipment status, and environmental data monitors the start-up, shutdown, and status transitions of production tasks. Based on multi-dimensional features (including product features, process path features, dynamic process features, and contextual features), feature engineering and embedding coding techniques are used to construct a low-dimensional feature vector representing the current production task's operating condition. .

[0011] Step 2: Algorithm-based intelligent recommendation. Based on the aforementioned working condition feature vector... Using the query criteria, an approximate nearest neighbor search based on vector similarity is performed in a pre-built case knowledge base to obtain a list of similar historical cases. A meta-learning model is then utilized. Using the aforementioned feature vectors as input, the algorithm directly predicts the multi-dimensional performance indicators (such as optimization accuracy, convergence speed, and stability) of each candidate quality optimization algorithm under the current operating conditions. Combining the preset performance indicator weights, the algorithm calculates the comprehensive utility score using multi-attribute utility theory, or filters the non-dominated solution set using Pareto front analysis, ultimately generating a list of the main recommendation algorithm and candidate algorithms. Simultaneously, interpretability techniques are used to generate a recommendation reason report.

[0012] Step 3: Digital Twin Optimization and Verification. In the digital twin virtual environment, based on the current production task's process route, dynamically instantiate the corresponding cross-process quality transfer simulation model chain. Load and run the optimization algorithm recommended in Step 2, driving the algorithm and simulation model chain to iteratively interact and optimize within the containerized simulation engine, solving for the set of process parameters that optimizes the predicted quality indicators. And a preview is conducted through a 3D visualization interface.

[0013] Step 4: Solution Confirmation and Implementation. The optimal process parameter set obtained from virtual simulation verification will be applied. After engineers confirm, fine-tune, and electronically approve the data on the human-machine interface, it is sent to the manufacturing execution system or directly control the equipment controller to drive the physical workshop to execute the optimized production plan.

[0014] Step 5: Feedback and Closed-Loop Learning. Collect quality and process data after actual production execution, calculate the actual achievement of quality indicators and optimization effects. Form this process into a new "algorithm-operating condition-performance" triplet case, store it in the case knowledge base, and use it to update the meta-learning model, achieving continuous self-evolution and improved optimization capabilities of the system.

[0015] On the other hand, the present invention provides a collinear production cross-process variable operating condition quality optimization algorithm recommendation and twin verification system for implementing the above method, which adopts a layered architecture and includes the following modules:

[0016] The database module is used to store and manage multi-source heterogeneous data that supports the operation of the system. Specifically, it includes: a real-time database for high-throughput access to real-time data streams; a historical database for storing historical production end-to-end data; an algorithm model library for managing various quality prediction and optimization algorithm entities; and a case knowledge base for storing "algorithm-operating condition-performance" case knowledge.

[0017] The intelligent decision-making module, as the core recommendation engine of the system, includes: a working condition perception unit for real-time perception of production tasks; a feature extraction unit for constructing working condition feature vectors; a similarity matching and retrieval unit for retrieving similar cases in the knowledge base; an algorithm performance prediction unit for predicting algorithm performance; a multi-objective recommendation decision-making unit for generating final recommendation results; and a model interpretation unit for providing decision explanations.

[0018] The digital twin module provides a digital twin verification environment, including: a virtual workshop ontology layer that constructs a virtual mapping of the physical workshop; a virtual workshop simulation modeling layer that constructs a chain of cross-process quality simulation models using a mechanism-data hybrid modeling method; and an algorithm execution and simulation engine layer that provides a containerized environment to load and execute recommendation algorithms for simulation optimization.

[0019] The application service module provides users with interactive interfaces and functional services, including: an algorithm recommendation service page that displays recommendation results and virtual simulations; an optimization scheme management page for managing the approval and issuance of optimization schemes; a panoramic visualization monitoring page for global monitoring and traceability; and a system configuration and knowledge maintenance page for system backend maintenance.

[0020] The application services module provides a user interface and services, including:

[0021] Algorithm Recommendation Service Page: Displays task dashboards, recommendation algorithms, and virtual sandbox preview interface, integrating 3D simulation animations and indicator curve visualization.

[0022] The optimized solution management page supports solution confirmation, parameter fine-tuning, electronic approval, and instruction issuance to the manufacturing execution system or equipment controller, and archives execution records.

[0023] Panoramic visualization monitoring page: Provides workshop-level 3D digital twin walkthrough, cross-process quality traceability chain, and algorithm recommendation efficiency statistics dashboard.

[0024] The system configuration and knowledge maintenance page provides functions for algorithm library management, case knowledge base maintenance, model performance monitoring and retraining triggering, and user permission configuration.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] (1) This invention breaks through the limitations of traditional single-point optimization by constructing a cross-process quality transfer simulation model chain and performing overall optimization in a virtual environment. It can handle complex nonlinear coupling relationships between multiple processes and achieve global optimization of product quality.

[0027] (2) The system can dynamically perceive changes in production conditions through real-time working condition perception, feature vectorization representation and case similarity matching, and recommend algorithms that perform well under similar historical working conditions. This solves the problem of performance drop of fixed algorithms under changing working conditions and significantly improves the adaptability and robustness of the recommendations.

[0028] (3) This invention forms a complete intelligent closed loop from "perception-recommendation-verification-execution-feedback". The system can not only automatically recommend algorithms, but also pre-verify the optimization effect in a digital twin environment, and continuously learn based on actual production feedback, constantly accumulating and optimizing recommendation knowledge, so that the intelligence level of the system continuously improves over time.

[0029] (4) The system generates recommendation reasons by integrating interpretability technology and provides rich human-computer interaction interface and three-dimensional visualization pre-show function, which enables engineers to understand the recommendation logic and confirm and fine-tune the optimization scheme, thereby enhancing the credibility and practicality of the system and facilitating its application in actual workshops. Attached Figure Description

[0030] Figure 1 This is an overall framework diagram of the present invention.

[0031] Figure 2 This is a flowchart of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other. To achieve the above objectives, this invention adopts the following technical solution.

[0033] This invention proposes a method for quality optimization algorithm recommendation and a digital twin verification system for cross-process variable operating conditions in co-line production. The method comprises five core steps: data aggregation and operating condition perception, intelligent algorithm recommendation, digital twin optimization verification, scheme confirmation and entity execution, and effect feedback and closed-loop learning. The system adopts a layered architecture, including a database module, an intelligent decision-making module, a digital twin module, and an application service module.

[0034] The system architecture and workflow of this invention can be found in the appendix. Figure 1 and attached Figure 2 The specific implementation method is described in detail below:

[0035] Step 1: Data aggregation and operational condition awareness, constructing the current task operational condition feature vector.

[0036] This step aims to comprehensively perceive the status of the production site and construct a digital feature vector that can characterize the unique operating conditions of the current production task. Specifically, this is achieved through the collaborative operation of the condition perception unit and the feature extraction unit within the intelligent decision-making module.

[0037] Step 1.1: Real-time Access and Aggregation of Multi-Source Heterogeneous Data. The system collects dynamic data streams in real time through interfaces with sensors, PLCs, CNC controllers, and Manufacturing Execution System (MES) deployed at key nodes in the physical workshop. This data is accessed and processed with high throughput and low latency by a real-time database (such as a time-series database) in the database module. The collected data mainly includes:

[0038] Equipment status signals: such as spindle speed, feed rate, motor current, vibration, temperature, etc.

[0039] Process parameters: such as cutting depth, feed rate, heat treatment temperature, pressure, time and other process settings.

[0040] Environmental data: such as workshop temperature and humidity, cleanliness, etc.

[0041] The real-time database aligns these streaming data by production order number and timestamp, and supports aggregation calculations (such as mean and standard deviation) over sliding time windows, providing structured time-series data fragments for subsequent analysis.

[0042] Step 1.2: Historical and Contextual Data Association. Simultaneously, the system retrieves structured master data (such as product BOM and process route cards) related to the current work order from the historical database, as well as the entire production chain records of similar products from the past. These records, at the "work order-process" granularity, associate parameters of each process from raw materials to finished product with the final inspection quality data, providing a data foundation for understanding the process coupling relationship.

[0043] Step 1.3: Multi-dimensional Feature Extraction and Embedding Encoding. The feature extraction unit extracts four dimensions of raw features from the aggregated data:

[0044] Product features: product model, material grade, design complexity, etc.

[0045] Process path characteristics: process sequence, equipment type combination, process constraint relationship, etc.

[0046] Dynamic process characteristics: statistical features (such as mean, variance, extreme values), spectral characteristics, and stability indicators of key parameters extracted from real-time data.

[0047] Contextual features: order priority, production batch, shift information, etc.

[0048] The extracted raw features, after preprocessing such as standardization and normalization, are input into a pre-trained autoencoder. This encoder maps the high-dimensional, sparse raw features into a low-dimensional, dense, continuous vector, i.e., the working condition feature vector. . It retains the most critical variation information in the original data and is applicable to subsequent vector similarity calculations.

[0049] Step 2: Algorithm-based intelligent recommendation, generating recommendation results.

[0050] This step is the core of the system's intelligent decision-making process. Based on the characteristics of the current operating conditions, it dynamically recommends the most suitable optimization algorithm from the algorithm library. This is executed serially by multiple units of the intelligent decision-making module.

[0051] Step 2.1: Similar Historical Case Retrieval. The similarity matching and retrieval unit uses the working condition feature vector generated in Step 1. As a query vector, a hybrid retrieval is performed in the case knowledge base. The case knowledge base stores entity relationships (e.g., "Task A used algorithm B, achieving performance C") in a graph database, while also storing the working condition feature vectors of each historical case. Store in a vector database and create an index. During retrieval, calculate... With each in the library Return the highest similarity score. A list of similar cases is compiled from historical examples. These case studies document the actual performance of different algorithms (such as genetic algorithms, particle swarm optimization, Bayesian optimization, etc.) under similar working conditions.

[0052] Step 2.2: Candidate Algorithm Performance Prediction. The algorithm performance prediction unit will predict the current feature vector. The retrieved similar case information is input together into a pre-trained meta-learning model. This model (e.g., based on a model-agnostic meta-learning-MAML framework) can learn to quickly adapt to new tasks from a large number of historical "algorithm-condition-performance" triples. Its output is a summary of each candidate algorithm... Prediction vectors of multiple performance indicators under current operating conditions :

[0053] ;

[0054] in, These are the model parameters. Performance metrics for prediction can include prediction accuracy, the number of iterations (time) required for convergence, and the stability of the solution set.

[0055] Step 2.3: Multi-objective decision-making and recommendation generation. The multi-objective recommendation decision-making unit receives the performance prediction vectors of all candidate algorithms. System administrators or process experts can pre-set the weight vectors of each performance indicator according to production objectives (such as prioritizing quality or efficiency). The unit calculates the overall utility score for each algorithm. And sort them. Alternatively, Pareto front analysis can be used to find the "non-dominated solution set" that is not inferior to other algorithms on all performance metrics, providing engineers with multiple optimal balance solutions. The final output is the main recommendation algorithm. And an alternative list.

[0056] Step 2.4: Generate an interpretability report. The model interpretation unit uses techniques such as SHAP to analyze which operating condition characteristics (e.g., "abnormally increased spindle vibration" or "high-strength steel material") played a key role in recommending a particular algorithm. Simultaneously, it displays specific performance data from similar historical cases, generating a visually appealing report explaining the reasons for the recommendation, which is then presented to the user through the application service module interface.

[0057] Step 3: Digital twin optimization and verification, output optimization parameter set

[0058] This step, before deployment in physical production, verifies the effectiveness of the recommendation algorithm in a digital twin environment to ensure the safety and feasibility of the optimization solution. This is the sole responsibility of the digital twin module.

[0059] Step 3.1: Dynamic Instantiation of the Simulation Model Chain. The virtual workshop simulation modeling layer, based on the current production task's process route, calls the corresponding process-level mechanism-data hybrid models (e.g., turning force model, heat treatment phase transformation model, assembly deviation transfer model) from the model library. These models are then connected according to the actual process sequence, instantiating a complete cross-process quality transfer simulation model chain. The input to this model chain is a set of adjustable process parameters for each process step. The output is the predicted final product quality index.

[0060] Step 3.2: Containerized Simulation and Optimization Solution. The algorithm execution and simulation engine layer creates a lightweight containerized simulation environment. Within this environment, a recommendation algorithm is loaded from the algorithm model library. The packages and dependencies. Then, the engine-driven algorithm... With simulation model chain Iterative interaction: The algorithm continuously generates new combinations of process parameters. The algorithm takes a model chain as input and returns the quality of its predictions. Based on this, it evaluates the results and searches for better solutions. This process can be formalized as solving the following optimization problem:

[0061] ;

[0062] in, This is the feasible region of the process parameters (defined by constraints such as equipment capacity and process specifications). After solving, the theoretically optimal set of process parameters is obtained. .

[0063] Step 3.3: Virtual Sandbox Preview and Visualization. The optimization process and results are synchronized in real time to the algorithm recommendation service page of the application service module. Users can view the virtual device according to the optimized parameters in the "Virtual Sandbox" interface in the form of a 3D animation. During the operation, the predicted change curves of key quality indicators can be viewed simultaneously, thereby intuitively evaluating the optimization plan.

[0064] Step 4: Solution Confirmation and Implementation

[0065] This step completes the transition from the virtual world to the physical world, putting the validated optimization scheme into practice:

[0066] Engineers review recommendation algorithms and optimization parameters on the optimization scheme management page. The system provides simulation results and explanatory reports. Parameters can be fine-tuned. After confirmation, the system proceeds through an electronic approval process. The optimized process parameter set is then transmitted via standardized interfaces (such as OPC UA and REST API). Control commands are sent to the Manufacturing Execution System (MES) or directly to the equipment CNC system, PLC, and other controllers at the workshop level to drive physical equipment to perform production according to the new parameters.

[0067] Step 5: Feedback and Closed-Loop Learning

[0068] This step enables the system to evolve itself, allowing its recommendation capabilities to continuously improve over time:

[0069] After actual production is completed, the system automatically collects the actual quality inspection data (such as dimensional accuracy, surface roughness, and mechanical properties) and actual process parameter curves of the final product of the work order through the MES and quality inspection system. This data is then compared with the prediction results from step 3 to calculate the actual performance of the recommendation algorithm in actual production (such as the optimization target achievement rate and parameter robustness).

[0070] Subsequently, the system automatically constructs a new case knowledge unit: "Current operating condition characteristics". - Algorithm used - Actual performance This triple will be stored in the case knowledge base to enrich the system's knowledge reserves. Simultaneously, these new data pairs will be incorporated into the meta-learning model. The retraining dataset is used to periodically or trigger model updates, thereby making the next recommendation more accurate.

[0071] This invention also provides a system for recommending and verifying quality optimization algorithms across processes and operating conditions in co-line production, comprising:

[0072] The database module is used to store and manage multi-source heterogeneous data, including real-time databases, historical databases, algorithm model libraries, and case knowledge bases.

[0073] The intelligent decision-making module is used to perceive the current production conditions, extract condition feature vectors, retrieve similar cases, predict algorithm performance, and make multi-objective recommendation decisions.

[0074] The digital twin module is used to build a chain of virtual workshop ontology and simulation models, providing a containerized simulation environment to execute recommendation algorithms and perform virtual verification;

[0075] The application service module provides user interfaces and service interfaces for algorithm recommendation services, optimization scheme management, panoramic visualization monitoring, system configuration, and knowledge operation and maintenance.

[0076] The intelligent decision-making module includes:

[0077] The working condition sensing unit connects to the data stream of the manufacturing execution system in real time to monitor the task status;

[0078] The feature extraction unit extracts and encodes low-dimensional working condition embedding vectors from multi-dimensional original features;

[0079] The similarity matching and retrieval unit retrieves similar historical cases from the case knowledge base based on vector similarity.

[0080] The algorithm performance prediction unit uses a meta-learning model to predict the performance of each candidate algorithm under the current operating conditions.

[0081] A multi-objective recommendation decision unit generates a list of recommendation algorithms based on comprehensive utility or Pareto frontier;

[0082] The model interpretation unit performs interpretability analysis on the recommendation results and generates a reasoning report.

[0083] The digital twin module includes:

[0084] The virtual workshop entity layer constructs a three-dimensional geometric and lightweight physical model corresponding to the physical workshop.

[0085] The virtual workshop simulation modeling layer uses a mechanism-data hybrid modeling method to construct a cross-process quality transfer model chain and supports dynamic instantiation.

[0086] The algorithm execution and simulation engine layer provides a containerized environment to load and run recommendation algorithms, driving them and the simulation model chain to perform optimization iterations.

[0087] The application service module includes:

[0088] The algorithm recommendation service page integrates task dashboards, recommendation result displays, virtual sandbox simulations, and 3D visualizations.

[0089] The solution management page has been optimized to support solution confirmation, parameter fine-tuning, electronic approval, and instruction issuance.

[0090] The panoramic visualization monitoring page provides workshop-level 3D walkthrough, cross-process quality traceability chain, and recommended performance statistics dashboard;

[0091] The system configuration and knowledge maintenance page provides functions for maintaining the algorithm library and case knowledge base, monitoring model performance and triggering retraining, and configuring user permissions.

[0092] In summary, the system of this invention, through the collaborative work of the five steps and four modules, forms a complete intelligent closed loop of "perception-decision-verification-execution-learning." The database module provides the data foundation for the entire process; the intelligent decision-making module is the brain that generates intelligent recommendations; the digital twin module provides a safe and low-cost pre-simulation and verification optimization sandbox; and the application service module is the bridge connecting users and the system. Through this layered collaborative mechanism, this invention effectively solves the challenges of adaptive and intelligent selection of quality optimization algorithms in digital twin workshops under co-line production, cross-process, and variable operating condition scenarios, significantly improving the accuracy, automation level, and universality of quality optimization.

[0093] Contents not described in detail in this specification are prior art known to those skilled in the art. The above descriptions are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the 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 quality optimization and recommendation algorithm for cross-process variable operating conditions in co-line production, characterized in that, Includes the following steps: Step 1: Data Aggregation and Condition Awareness. Real-time access to order data, work-in-process parameters, equipment status, and environmental data from the Manufacturing Execution System (MES) is used to monitor task start / stop and status transitions. Based on product characteristics, process path characteristics, dynamic process characteristics, and contextual characteristics, a low-dimensional condition feature vector for the current production task is constructed through feature engineering and automatic encoder mapping. ; Step 2: Algorithm-based intelligent recommendation, using the aforementioned working condition feature vector. For querying, an approximate nearest neighbor retrieval based on cosine similarity is performed in a pre-built case knowledge base to obtain a list of similar historical cases; based on a meta-learning model... Predict the multidimensional performance indicators of each candidate algorithm under the current working conditions; calculate the comprehensive utility of each algorithm by combining the preset weights and use multi-objective Pareto front analysis to screen the non-dominated solution set, and generate the main recommendation algorithm and the candidate list; Step 3: Digital twin optimization and verification, dynamically instantiating the cross-process quality transfer simulation model chain corresponding to the current process route in the digital twin environment; The main recommendation algorithm is loaded and executed, driving iterative interaction between the algorithm and the simulation model chain in a containerized simulation environment to solve for the set of process parameters that optimizes the predicted quality indicators. ; Step 4: Solution Confirmation and Implementation, including the optimized process parameter set. After manual confirmation and adjustment, the data is sent to the Manufacturing Execution System or equipment controller to drive execution in the physical shop floor; Step 5: Effect feedback and closed-loop learning. Collect actual production data, calculate real quality indicators and optimization effects, form new "algorithm-operating condition-performance" cases, store them in the case knowledge base, and use them to update the meta-learning model and optimize subsequent recommendations.

2. The method according to claim 1, characterized in that, Step 1 includes: Step 1.1: Access equipment status signals, process parameters, and environmental data streams from the physical workshop through a high-throughput real-time database, and perform aggregation processing based on time windows; Step 1.2: Extract cross-process end-to-end parameters and quality records at the production work order granularity from the historical database, metadata associated with unstructured data, and product master data; Step 1.3: The feature extraction unit extracts raw features from multiple dimensions, including product features, process path features, dynamic process features, and contextual features. After feature selection, scaling, and encoding, the raw features are input into the autoencoder to obtain a low-dimensional operating condition embedding vector. .

3. The method according to claim 1 or 2, characterized in that, Step 2 includes: Step 2.1: In the case knowledge base, based on the working condition feature vector Feature vectors of historical cases in the database Find similar cases in historical cases ; Step 2.2: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require Input pre-trained meta-learning model Directly predict each candidate algorithm Output vector in multidimensional performance metrics : ； in, For the set of model parameters; Step 2.3: Based on the preset performance index weight vector Calculate the overall utility score of each algorithm. Alternatively, Pareto front analysis can be used to screen for non-dominated solutions and generate the final recommendation result. .

4. The method according to claim 3, characterized in that, Step 2 is followed by: Step 2.4: Model Explanation and Reason Generation. Use SHAP or LIME technology to perform feature attribution analysis on the recommendation decision and associate it with the performance of similar historical cases to generate an explainable recommendation reason report.

5. The method according to claim 1, characterized in that, Step 3 includes: Step 3.1: In the digital twin module, based on the process route of the current production task, dynamically instantiate the corresponding process simulation model from the mechanism-data hybrid model library, and connect them into a cross-process quality transfer model chain. ; Step 3.2: Load the recommendation algorithm into the digital twin module to Perform iterative interactions to solve the optimization problem: ； in, This represents the feasible region for process parameters; Step 3.3: Through 3D simulation animation and indicator curve visualization, the optimization process and results are previewed and displayed on the virtual sand table interface.

6. A twin verification system for recommending quality optimization algorithms across processes and operating conditions in co-line production, which implements the method described in any one of claims 1-5, characterized in that, include: The database module is used to store and manage multi-source heterogeneous data, including real-time databases, historical databases, algorithm model libraries, and case knowledge bases. The intelligent decision-making module is used to perceive the current production conditions, extract condition feature vectors, retrieve similar cases, predict algorithm performance, and make multi-objective recommendation decisions. The digital twin module is used to build a virtual workshop ontology and simulation model chain, providing a containerized simulation environment to execute recommendation algorithms and perform digital twin verification; The application service module provides algorithm recommendation services, optimization scheme management, panoramic visualization monitoring, and user interface and service interfaces for system configuration and knowledge operation and maintenance.

7. The system according to claim 6, characterized in that, The intelligent decision-making module includes: The working condition sensing unit connects to the data stream of the manufacturing execution system in real time to monitor the task status; The feature extraction unit extracts and encodes low-dimensional working condition embedding vectors from multi-dimensional original features; The similarity matching and retrieval unit retrieves similar historical cases from the case knowledge base based on vector similarity. The algorithm performance prediction unit uses a meta-learning model to predict the performance of each candidate algorithm under the current operating conditions. A multi-objective recommendation decision unit generates a list of recommendation algorithms based on comprehensive utility or Pareto frontier; The model interpretation unit performs interpretability analysis on the recommendation results and generates a reasoning report.

8. The system according to claim 6, characterized in that, The digital twin module includes: The virtual workshop entity layer constructs a three-dimensional geometric and lightweight physical model corresponding to the physical workshop. The virtual workshop simulation modeling layer uses a mechanism-data hybrid modeling method to construct a cross-process quality transfer model chain and supports dynamic instantiation. The algorithm execution and simulation engine layer provides a containerized environment to load and run recommendation algorithms, driving them and the simulation model chain to perform optimization iterations.

9. The system according to claim 6, characterized in that, The application service module includes: The algorithm recommendation service page integrates task dashboards, recommendation result displays, virtual sandbox simulations, and 3D visualizations. The solution management page has been optimized to support solution confirmation, parameter fine-tuning, electronic approval, and instruction issuance. The panoramic visualization monitoring page provides workshop-level 3D walkthrough, cross-process quality traceability chain, and recommended performance statistics dashboard; The system configuration and knowledge maintenance page provides functions for maintaining the algorithm library and case knowledge base, monitoring model performance and triggering retraining, and configuring user permissions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 5.

Images