A novel completion status prediction for the aircraft mixed-model assembly lines: A study in dynamic Bayesian networks

In the context of Industry 4.0, amidst the normalization of multi-variety and low-volume custom flexible production models, aircraft mixed-model assembly lines (AMMALs) have become widely adopted in various aerospace companies. Within AMMALs, each aircraft has its own manufacturing cycles and delivery dates, making accurate prediction of the completion status of each model critical. However, owing to the complexity of production operation data, the dynamic time-varying nature of the process, its weak regularity, and other factors, current methods for predicting completion states struggle to ensure prediction accuracy. Moreover, owing to the product characteristics of aircraft and economic considerations, it is difficult to accumulate a substantial amount of data for different models, especially for newly improved models. To solve the problem of completion state prediction in the context of dynamic time-varying conditions and limited data accumulation, we propose a Dynamic Bayesian Networks (DBNs)-based method for predicting AMMAL completion states. First, we construct a hierarchical agent model for cross-products based on DBNs to effectively generalize the dynamic time-varying features across other multi-products. Then, we utilize expert knowledge and DBN parameter learning to solve small-sample problems in AMMALs. Simultaneously, we propose a model synchronization update mechanism based on improved particle filtering to enhance reasoning speed and prediction accuracy. Experiments conducted at AMMALs have demonstrated the approach's feasibility, yielding results showing improved predictive ability of the job model. This method enables aerospace companies to perceive development trends and make scientific decisions in advance.