CN-121998792-A - Co-line production cross-process variable working condition quality tuning algorithm recommendation method and twin verification system

CN121998792ACN 121998792 ACN121998792 ACN 121998792ACN-121998792-A

Abstract

The invention discloses a collinear production cross-process variable working condition quality tuning algorithm recommendation method and a twin verification system, and belongs to the technical field of intelligent manufacturing and digital twin. The method comprises a data aggregation and working condition sensing step, an algorithm intelligent recommending step, a digital twin optimization verifying step, a digital twin environment instantiating cross-process quality model chain, a driving recommending algorithm to carry out iterative optimization, a scheme confirming and entity executing step, a physical workshop executing step and an effect feedback and closed loop learning step, wherein the working condition characteristic vector is used for representing a current task, the similar historical working condition is searched in a case knowledge base based on the characteristic vector, the performance of each candidate algorithm is predicted by utilizing a meta learning model, a recommending result is generated through multi-objective decision, the digital twin optimization verifying step is carried out in a digital twin environment, the recommending algorithm is driven to carry out iterative optimization, an optimal process parameter set is solved, the verified optimizing scheme is issued to a physical workshop to be executed, and an effect feedback and closed loop learning step is used for collecting actual production data to form a new case and feeding back to a system, and continuous evolution is realized.

Inventors

TAO FEI
QI QINGLIN
ZHAO SHILIN
CHENG YING

Assignees

北京航空航天大学

Dates

Publication Date: 20260508
Application Date: 20260123

Claims (10)

1. A collinear production cross-process variable working condition quality tuning algorithm recommendation method is characterized by comprising the following steps: Step 1, data aggregation and working condition sensing, real-time access to order form, product process parameters, equipment state and environment data of a manufacturing execution system, monitoring task start-stop and state switching, and constructing a low-dimensional working condition feature vector of a current production task through feature engineering and automatic encoder mapping based on product features, process path features, dynamic process features and context features ; Step 2, intelligent recommendation of algorithm, wherein the working condition characteristic vector is used For inquiring, performing approximate nearest neighbor search based on cosine similarity in a pre-constructed case knowledge base to obtain a similar historical case list, and based on a meta-learning model The multi-dimensional performance index of each candidate algorithm under the current working condition is predicted, and the comprehensive utility of each algorithm is calculated or a non-inferior solution set is screened by combining with the pre-set weight or multi-target pareto front analysis to generate a main recommendation algorithm and an alternative list; Step 3, digital twin optimization verification, namely dynamically instantiating a cross-process mass transfer simulation model chain corresponding to the current process route in a digital twin environment, loading and executing the recommendation algorithm, driving the algorithm to interact with the simulation model chain in a containerized simulation environment in an iterative manner, and solving a process parameter set which enables the prediction quality index to be optimal ; Step 4, scheme confirmation and entity execution, wherein the optimized technological parameter set After manual confirmation and adjustment, the data is issued to a manufacturing execution system or an equipment controller to drive a physical workshop to execute; and 5, effect feedback and closed-loop learning, collecting actual production data, calculating actual quality indexes and optimizing effects, forming new cases of algorithm, working condition and performance, storing the new cases into a case knowledge base, and updating a meta-learning model and optimizing subsequent recommendation.
2. The method according to claim 1, wherein the step 1 comprises: Step 1.1, high throughput access of equipment state signals, technological process parameters and environmental data streams from a physical workshop is realized through a real-time database, and aggregation treatment is carried out based on a time window; Step 1.2, extracting cross-process full-link parameters and quality records with production worksheets as granularity from a historical database, metadata associated with unstructured data and product main data; Step 1.3, extracting original features from multiple dimensions of product features, process path features, dynamic process features and context features by a feature extraction unit, and inputting the original features into an automatic encoder to obtain low-dimensional working condition embedded vectors after feature selection, scaling and encoding treatment 。
3. The method according to claim 1 or 2, wherein step 2 comprises: step 2.1, in the case knowledge base, based on the working condition feature vector Feature vectors with historical cases in a library Searching similar cases in the historical cases ; Step 2.2. Will be Inputting a pre-trained meta-learning model Direct prediction of candidate algorithms Output vector on multi-dimensional performance index : ; Wherein, the Is a model parameter set; step 2.3 weight vector based on preset Performance index Calculating the comprehensive utility score of each algorithm Or screening the non-inferior solution set through pareto front analysis to generate a final recommended result 。
4. A method according to claim 3, further comprising, after step 2: And 2.4, model interpretation and reason generation, namely performing feature attribution analysis on the recommendation decision by adopting SHAP or LIME technology, and correlating performance performances of similar historical cases to generate an interpretable recommendation reason report.
5. The method according to claim 1, wherein the step 3 comprises: Step 3.1, dynamically instantiating corresponding process simulation models from a mechanism-data mixed model library in a digital twin module according to the process route of the current production task, and connecting the process simulation models into a cross-process mass transfer model chain ; Step 3.2, loading recommendation algorithm in the digital twin module And combine it with Iterative interaction is carried out, and an optimization problem is solved: ; Wherein, the Is a feasible region of process parameters; And 3.3, performing previewing and displaying the optimization process and the result on the virtual sand table interface through three-dimensional simulation animation and index curve visualization.
6. The utility model provides a colinear production is to change operating mode quality accent algorithm recommendation twin verification system by technology, which characterized in that includes: The database module is used for storing and managing multi-source heterogeneous data, and comprises a real-time database, a historical database, an algorithm model library and a case knowledge base; The intelligent decision module is used for sensing the current production working condition, extracting the working condition feature vector, retrieving similar cases, predicting the algorithm performance and carrying out multi-objective recommendation decision; The digital twin module is used for constructing a virtual workshop body and a simulation model chain, providing a containerized simulation environment to execute a recommendation algorithm and carrying out digital twin verification; and the application service module is used for providing algorithm recommendation service, optimization scheme management, panoramic visual monitoring, system configuration and knowledge operation and maintenance user interfaces and service interfaces.
7. The system of claim 6, wherein the intelligent decision module comprises: the working condition sensing unit is connected with the manufacturing execution system data stream in real time and monitors the task state; the feature extraction unit extracts and codes the low-dimensional working condition embedded vector from the original multi-dimensional features; the similarity matching and searching unit is used for searching similar historical cases in the case knowledge base based on the vector similarity; the algorithm performance prediction unit predicts the performance of each candidate algorithm under the current working condition by using a meta learning model; the multi-target recommendation decision unit generates a recommendation algorithm list based on the comprehensive utility or the pareto front; and the model interpretation unit is used for carrying out interpretability analysis on the recommended result and generating a reason report.
8. The system of claim 6, wherein the digital twinning module comprises: the virtual workshop body layer is used for constructing a three-dimensional geometric and lightweight physical model corresponding to the physical workshop; The virtual workshop simulation modeling layer adopts a mechanism-data mixed modeling method to construct a cross-process mass transfer model chain and supports dynamic instantiation; And the algorithm execution and simulation engine layer provides a containerized environment to load and run a recommended algorithm and drive the recommended algorithm to perform optimization iteration with a simulation model chain.
9. The system of claim 6, wherein the application service module comprises: the algorithm recommends a service page, integrates a task billboard, recommended result display, virtual sand table previewing and three-dimensional visualization; Optimizing a scheme management page, supporting scheme confirmation, parameter fine adjustment, electronic approval and instruction issuing; Panoramic visual monitoring pages are provided for workshop-level three-dimensional roaming, a cross-procedure quality tracing chain and a recommended efficiency statistics instrument panel; the system configuration and knowledge operation and maintenance page provides the functions of maintaining an algorithm library and a case knowledge base, monitoring and retraining the model performance, triggering and configuring the user permission.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method according to any one of claims 1 to 5.

Description

Co-line production cross-process variable working condition quality tuning algorithm recommendation method and twin verification system Technical Field The invention relates to the technical field of intelligent manufacturing, industrial big data and digital twinning, in particular to a recommendation method and a twinning verification system for a cross-process variable working condition quality tuning algorithm for collinear production. Background With the deep advancement of intelligent manufacturing, digital twin technology has become an important means for constructing intelligent workshops with virtual-real fusion and real-time interaction. Particularly, under the background that the multi-variety, small-batch and customized production modes are increasingly popular, collinear production, namely flexible organization and scheduling production of different products or process paths on the same production line, has become a key mode for improving the utilization rate and response speed of manufacturing resources. However, this mode of production, while giving flexibility, also complicates the problem of quality optimization, mainly in the following respects. In the collinear production scene, the products in the same batch or different batches can undergo different process routes, and the quality of the final product is not determined by a single process, but is jointly influenced by the process parameters of a plurality of processes, and a complex nonlinear coupling relationship exists among the processes. The conventional single point quality optimization method has difficulty in handling such a global optimization problem across processes. In a collinear production scenario, frequent switching of production orders results in dynamic changes in process objects, equipment states, environmental conditions. An optimization algorithm that performs well under certain conditions may collapse in effect under another condition. Currently, shop engineers mainly rely on empirical selection or trial and error methods to find suitable quality optimization algorithms (such as neural networks, support vector machines, genetic algorithms, etc.), which are inefficient and difficult to guarantee optimality. The application of the current digital twin system on the workshop level is focused on state monitoring, flow visualization and single-procedure simulation, and the whole quality optimization decision support capability for multiple processes and variable working conditions is lacking. The system often adopts a preset optimization algorithm or model, and the most suitable optimization strategy cannot be dynamically selected and recommended according to the real-time working condition, so that the self-adaptive capacity of digital twin in quality closed-loop control is insufficient and the intelligent level is limited. Therefore, in the complex manufacturing environment of collinear production, cross-process chain and variable working condition interweaving, a system and a method capable of sensing the state of the full-link of production in real time, dynamically identifying the working condition characteristics, intelligently recommending and verifying the optimization algorithm and finally realizing continuous self-learning are needed to break through the decision bottleneck of the existing digital twin workshop in the quality optimization link, and accurately, efficiently and adaptively realize quality control. Disclosure of Invention The invention aims to solve the technical problem that a digital twin workshop lacks the recommendation capability of a self-adaptive and intelligent quality optimization algorithm under the conditions of cross-process coupling and variable working condition production in the existing collinear production environment, and provides a recommendation method and a twin verification system for the variable working condition quality optimization algorithm of the collinear production. The system can automatically recommend the optimal quality optimization algorithm according to the real-time production of the full-link data and the dynamic working condition, so that the accuracy, the automation level and the universality of quality optimization are improved. In order to achieve the above purpose, the invention adopts the following technical scheme: on the one hand, the invention provides a recommendation method for a cross-process variable-working-condition quality tuning algorithm for collinear production, which comprises the following steps: And 1, data aggregation and working condition sensing. Order data, cross-process technological parameters of products, equipment states and environment data from a manufacturing execution system are accessed in real time, and start-stop and state switching of production tasks are monitored. Based on multidimensional features (including product features, process path features, dynamic process features and context features), a low-dimensional feature