A Machine Learning-Based System for the Prediction of the Lead Times of Sequential Processes

In this paper, a system for the prediction and analysis of the lead times of sequential manufacturing processes is proposed. The system includes two machine learning regression models, which estimate the process lead times based on the values of different independent variables: the lead times of up-stream processes, organizational variables, product specifications, and quality inspection reports. Additionally, the proposed system feeds the downstream processes lead time prediction modules with information regarding the lead time estimation of previous operations. The approach has been tested with real-world data from the case study of a wind turbine tower manufacturer. In particular, the applied system includes two modules for the prediction of the lead times of the bending and longitudinal welding processes. The system produces somewhat inaccurate predictions for the former, but significantly increases its predictive power for the longitudinal welding process, partly with the help of the bending lead time estimations. The output of the proposed system, that is, the lead time estimations, has direct applications in two key aspects of production planning and control: job scheduling and anomaly control.