Research On Data-driven Methods For Identifying And Predicting Bottlenecks In The Apparel Production Process

The textile and garment industry is an important pillar of China’s economic development.However,China’s textile manufacturing industry still adopts the traditional "batch" production mode,with low automation level,serious problems of information islands,production capacity bottlenecks,and rigid production,which leads to process bottlenecks in garment production and affects the efficiency and quality of garment production.Therefore,to meet the automation needs of textile and garment enterprises and achieve rapid and efficient garment production,enterprises need to accurately identify and predict process bottlenecks in garment production.However,in the actual garment production process,the process flow is complex,and the workstation fluctuation is large,leading to low bottleneck identification accuracy and frequent bottleneck location shifts,which brings great difficulties for enterprises to identify and predict bottlenecks.To address the above challenges,this paper conducts research on garment production workflow modeling,production process and resource knowledge modeling,and dynamic bottleneck detection and spatiotemporal prediction methods.The research work in this paper includes:(1)Establishing a process model oriented to garment production event logs and identifying static bottlenecks in the production process.Exporting standard event log data from enterprise information systems,generating real garment production workflow models based on inductive process discovery mining algorithms,and based on this,obtaining histo rical production process performance statistics by replaying event logs on the process model,effectively identifying historical static process bottlenecks.(2)Designing an ontological knowledge model for garment production processes and resources,and cl assifying and associating production process data.According to the ontological model,importing data from enterprise production information systems,constructing a production process knowledge graph,achieving unified management and query of production pr ocess knowledge,and providing data sources for subsequent bottleneck detection and prediction.(3)Researching dynamic bottleneck detection and prediction methods for garment production processes,analyzing the layout structure of garment production lines,and proposing a garment production dynamic detection method based on the turning point method;performing spatiotemporal analysis on garment production bottleneck features,and analyzing the causes of production bottlenecks.A deep learning prediction mo del based on the spatiotemporal graph attention network is proposed,which can adaptively learn the spatial and temporal dimension features of the bottleneck state.Finally,a prototype system is designed and developed,and the proposed method is applied to the production process of a suit,verifying the effectiveness of the spatiotemporal graph attention network bottleneck prediction model.Compared with existing algorithm models,the accuracy is significantly improved.It can provide timely reminders of future bottlenecks and reduce the frequency of bottleneck occurrence.