CN-121981473-A - Coating workshop resource scheduling method based on multi-source heterogeneous data and federal learning

CN121981473ACN 121981473 ACN121981473 ACN 121981473ACN-121981473-A

Abstract

The invention discloses a painting workshop resource scheduling method based on multi-source heterogeneous data and federal learning, and belongs to the technical field of painting workshop resource scheduling. The method comprises the steps of collecting multisource heterogeneous data such as equipment running states, material circulation, environmental parameters and the like in real time, constructing a federal learning framework, training a global resource scheduling model by using locally collected data by each client, aggregating model parameters by a central server to generate an optimized global model, dynamically generating resource scheduling instructions according to real-time workshop state data by using the optimized model, automatically adjusting equipment tasks, logistics paths and production queues, finally, issuing the instructions to execution equipment, monitoring production progress and key performance indexes in real time through a visual interface, and early warning abnormality to form closed-loop control. The invention can obviously improve the utilization rate of resources and the production efficiency of a coating workshop and reduce the dependence on manual operation and the error rate while guaranteeing the data privacy of each production unit.

Inventors

ZHANG BO
KONG DEHUA
WANG CHENGHAN
XIAO XINLIN
Wang Gengkuan

Assignees

武汉东湖学院

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (15)

1. The coating workshop resource scheduling method based on multi-source heterogeneous data and federal learning is characterized by comprising the following steps of: The multi-source heterogeneous data acquisition and preprocessing stage comprises the steps of acquiring multi-source heterogeneous data in real time through a sensor network, a manufacturing execution MES system and an enterprise resource planning ERP system which are deployed in a coating workshop, and cleaning, formatting and fusing the acquired multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data; The distributed model training stage based on federation learning comprises the steps of constructing a federation learning frame comprising a central server and a plurality of local clients, wherein each local client corresponds to workshop equipment in a coating workshop, initializing a global resource scheduling model by the central server and issuing the global resource scheduling model to each local client, carrying out local training on the received global resource scheduling model by each local client by utilizing the corresponding preprocessed local multi-source heterogeneous data, and uploading model parameters obtained by training to the central server; Generating a dynamic resource scheduling strategy, namely generating a real-time resource scheduling instruction according to workshop state data acquired in real time by a global resource scheduling model optimized by federal learning, dynamically adjusting workshop equipment tasks, production logistics distribution and workshop personnel work, enabling a coating workshop to be always in a current optimal working state, improving the resource utilization rate and reducing the error rate; Scheduling execution and visual monitoring, namely, issuing the generated resource scheduling instruction to the execution equipment of the workshop layer to drive the coating production line to dynamically adjust, and simultaneously, displaying the production progress, the resource utilization rate and the equipment performance index in real time through a visual interface and early warning the scheduling abnormality.
2. The method for scheduling paint shop resources based on multi-source heterogeneous data and federal learning according to claim 1, wherein the multi-source heterogeneous data comprises equipment operation state data, material circulation data, environmental parameter data, production order data and personnel state data.
3. The coating shop resource scheduling method based on multi-source heterogeneous data and federal learning according to claim 1, wherein the multi-source heterogeneous data acquisition and preprocessing stage further comprises the following steps: And calculating the similarity of time sequences of different data sources by adopting a dynamic time warping algorithm, carrying out alignment treatment on unaligned time sequence data, carrying out feature extraction and dimension reduction on multi-source heterogeneous data acquired in real time based on an information entropy theory, and screening out feature vectors for model training.
4. The method for scheduling paint shop resources based on multi-source heterogeneous data and federal learning according to claim 1, wherein the distributed model training phase based on federal learning further comprises the following steps: Distributing aggregation weights to each local client according to the data size of each local client and the information entropy of the data and the current global resource scheduling model, and representing the local client with higher information entropy, wherein the higher the new information content of the data is, the higher the assigned aggregation weights are, so that the adaptability of the global model to new working conditions is accelerated; And selecting a cluster center node for each equipment cluster to be responsible for preliminary aggregation of model parameters in the cluster, and uploading an aggregation result to a central server for global aggregation so as to reduce communication expenditure and improve the convergence speed of a global resource scheduling model on similar workshop equipment groups.
5. The coating shop resource scheduling method based on multi-source heterogeneous data and federal learning according to claim 1, wherein the global resource scheduling model is a Dec-POMDP model based on multi-agent and reinforcement learning, each agent corresponds to one federal learning operation or one production resource unit, and a reward function of the agent is set according to a time sensitivity index and a work aggressiveness index, wherein the time sensitivity index is positively related to urgency of a production task and risk of equipment failure, and the work aggressiveness index is negatively related to scheduled frequency of equipment history and current health state of equipment.
6. The method for scheduling paint shop resources based on multi-source heterogeneous data and federal learning of claim 1, wherein the real-time resource scheduling instructions include equipment task allocation, adjusting motion trajectories of the painting robot to eliminate bottleneck stations, rescheduling AGV material delivery path personnel work arrangement, and production queue priority adjustment.
7. The paint shop resource scheduling method based on multi-source heterogeneous data and federal learning according to claim 1, wherein the dynamic resource scheduling policy generation stage further comprises the steps of: Acquiring real-time workshop state data, calculating real-time performance indexes of the painting workshop according to the real-time workshop state data, and simultaneously monitoring real-time dynamic events; A dynamic optimization target generation step, namely dynamically generating a current scheduling period optimization target according to real-time workshop state data and real-time performance indexes by a global resource scheduling model subjected to federal learning optimization, wherein the priority of the current scheduling period optimization target is adjusted according to the type of a real-time dynamic event; generating and executing a self-adaptive scheduling strategy, namely generating a real-time resource scheduling instruction based on a current scheduling period optimization target by a global resource scheduling model subjected to federal learning optimization, and transmitting the real-time resource scheduling instruction to workshop layer execution equipment; And the effect evaluation and continuous iteration step comprises the steps of returning to the real-time state sensing and performance evaluation step after the real-time resource scheduling instruction executes a scheduling period, sensing the real-time workshop state again and evaluating the change of the real-time performance index to judge the effectiveness of the current scheduling strategy, and starting a new round of optimization target generation and strategy adjustment by the global resource scheduling model based on the effectiveness evaluation result to form continuous circulation and self-optimization closed-loop control.
8. The method for scheduling coating shop resources based on multi-source heterogeneous data and federal learning according to claim 7, wherein the real-time performance indexes comprise conveyor chain effective utilization, equipment comprehensive efficiency OEE, order on-time completion rate and current work-in-process quantity, and the real-time dynamic events comprise machine faults, emergency order insertion, material supply delay and deviation of process parameters from standard ranges.
9. The method for scheduling paint shop resources based on multi-source heterogeneous data and federal learning according to claim 8, wherein the process parameters include spray room temperature humidity, drying room temperature and paint viscosity.
10. The method for coating shop resource scheduling based on multi-source heterogeneous data and federal learning according to claim 8, wherein the current scheduling period optimization objective includes one or more of minimizing maximum completion time, balancing machine load, minimizing total energy consumption, and maximizing order on-time completion rate, and increasing priority of maximizing order on-time completion rate when an emergency order insertion event occurs.
11. The paint shop resource scheduling method based on multi-source heterogeneous data and federal learning according to claim 6, wherein the bottleneck station is identified and eliminated by: and for the bottleneck generated by the cooperative operation waiting of the robot, predicting potential obstacle trajectories by adopting a Kalman filtering algorithm, and optimizing the robot motion trajectories in real time by utilizing a multi-objective optimization algorithm so as to reduce the waiting time.
12. A coating shop resource scheduling system based on multi-source heterogeneous data and federal learning, characterized in that the coating shop resource scheduling system based on multi-source heterogeneous data and federal learning according to any one of claims 1 to 11 is implemented, comprising: The system comprises a multi-source heterogeneous data acquisition and preprocessing module, a processing module and a processing module, wherein the multi-source heterogeneous data acquisition and preprocessing module is used for acquiring multi-source heterogeneous data in real time through a sensor network, a manufacturing execution MES system and an enterprise resource planning ERP system which are deployed in a coating workshop, and cleaning, formatting and fusing the acquired multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data; The distributed model training module based on federal learning is used for constructing a federal learning framework comprising a central server and a plurality of local clients, wherein each local client corresponds to workshop equipment in a coating workshop; the central server initializes the global resource scheduling model and transmits the global resource scheduling model to each local client, and each local client uses the corresponding preprocessed local multi-source heterogeneous data to locally train the received global resource scheduling model and uploads the model parameters obtained by training to the central server; The dynamic resource scheduling strategy generation module is used for generating real-time resource scheduling instructions according to workshop state data acquired in real time by using a global resource scheduling model subjected to federal learning optimization, dynamically adjusting workshop equipment tasks, production logistics distribution and workshop personnel work, enabling a painting workshop to be always in a current optimal working state, improving the resource utilization rate and reducing the error rate; The scheduling execution and visual monitoring module is used for sending the generated resource scheduling instruction to the execution equipment of the workshop layer to drive the coating production line to dynamically adjust, and simultaneously displaying the production progress, the resource utilization rate and the equipment performance index in real time through the visual interface and early warning the abnormal scheduling.
13. A coating workshop resource scheduling device based on multi-source heterogeneous data and federal learning comprises a memory and a processor, wherein the memory stores a computer program capable of running on the processor, and the coating workshop resource scheduling device is characterized in that the processor realizes the method of claim 1 when executing the computer program 11, A coating shop resource scheduling method based on multi-source heterogeneous data and federal learning.
14. A computer readable storage medium, wherein computer executable instructions are stored in the computer readable storage medium, and when the computer executable instructions are loaded and executed by a processor, the method for scheduling resources of a painting shop based on multi-source heterogeneous data and federal learning according to any one of claims 1-11 is implemented.
15. A computer program product comprising instructions which, when run on a terminal, cause the terminal to perform the method for coating shop resource scheduling based on multi-source heterogeneous data and federal learning according to any one of claims 1-11.

Description

Coating workshop resource scheduling method based on multi-source heterogeneous data and federal learning Technical Field The invention relates to the technical field of painting workshop resource scheduling, in particular to a painting workshop resource scheduling method based on multi-source heterogeneous data and federal learning. Background The painting workshop is used as a key link in the field of modern manufacturing industry, particularly high-end equipment manufacturing fields such as automobiles, aerospace and the like, and the production efficiency and the resource scheduling level of the painting workshop directly influence the apparent quality, the production cost and the delivery cycle of products. The typical painting workshop comprises a plurality of working procedures such as primer polishing, intermediate coating and spray painting, finish coating and spray painting, drying, forced cooling and the like, and is basically completed by an industrial robot and auxiliary equipment, and has the characteristics of high automation degree and complex process. As market demand for personalized customization increases, flexible spray systems capable of handling multiple vehicle models or products have become mainstream, which further complicates scheduling problems within workshops. Currently, research and practice for resource scheduling of a painting shop have advanced, but most methods still have limitations. First, the existing research is focused on scheduling optimization in a static environment, namely, assuming that production tasks and equipment states are fixed in a scheduling period. However, the actual production process is full of various dynamic events such as sudden machine failures, emergency order insertion, delays in material supply, or drift in critical process parameters (e.g., spray booth temperature and humidity). These dynamic events can significantly interfere with the original scheduling scheme, resulting in production disjoints, reduced resource utilization, even production line outages, etc. Although research has been attempted to establish a rescheduling mechanism to cope with such events, for example, a particle swarm algorithm (BAS-PSO) based on a longhorn beetle whisker search is adopted to perform optimization solution, the response mechanism is often lagged, and it is difficult to realize accurate sensing and quick response to the real-time state of a workshop while the global optimality of a scheduling scheme is ensured. Secondly, at the data utilization level, there is a large amount of heterogeneous multi-source data inside the paint shop, including equipment running state data from sensors, production order and material flow data from MES and ERP systems, environmental parameter data, and the like. These data differ in format, sampling frequency, and semantics, forming "data islands". The existing scheduling system often fails to fully integrate and mine the value of the multi-source heterogeneous data, and only depends on partial data to make decisions, so that the comprehensiveness and accuracy of a scheduling strategy are insufficient. More importantly, due to data privacy and security concerns, the data of different processes or devices often cannot be directly centrally shared, further limiting the possibility of building a high quality scheduling model based on global data. Furthermore, conventional approaches face challenges when seeking globally optimal scheduling schemes. The resource scheduling of the painting shop is a typical NP-hard combination optimization problem, and multiple objectives such as maximum finishing time, machine load balancing, order on-time completion rate and the like need to be optimized simultaneously. Although the intelligent scheduling algorithm (such as a genetic algorithm, an ant colony algorithm and the like) is applied, the intelligent scheduling algorithm is easy to sink into local optimum in a workshop environment with dynamic, multi-objective and complex constraint, has large calculation cost, and is difficult to meet the requirement of real-time scheduling. In summary, the existing coating shop resource scheduling method still has obvious defects in the aspects of real-time performance of dealing with dynamic events, fusion utilization and privacy protection of multi-source heterogeneous data, and efficiency and quality of multi-objective optimization solution. Therefore, a new method capable of sensing the state of the workshop in real time, cooperatively learning on the premise of protecting the privacy of data and dynamically generating an optimal scheduling strategy is urgently needed, so that the painting workshop is always kept in the current optimal cooperative working state, and the production efficiency and the resource utilization rate are remarkably improved. Disclosure of Invention The invention aims to overcome the defects of the prior art and provides a coating workshop resource scheduling method based on m