Never Miss a Due Date: Continuous Order Completion Time Prediction in Manufacturing Using Deep Learning

Accurate order completion time (OCT) prediction is an important aspect in logistics and supply chain management. However, its inherent uncertainty due to factors such as random disruptions in manufacturing and logistic complexity makes it a difficult task. In this paper, we address the task by applying a novel transformer-based deep learning architecture that is designed to handle the temporal and sequential aspects of manufacturing data. Our transformer-based model predicts a log-normal probability distribution of OCTs, providing a time range with minimum and maximum OCTs, along with a confidence interval. The predictions are based on the properties of the order and all concurrent orders to reflect the workload of the manufacturing facility. The evaluation demonstrates that the predicted OCT distributions aligns with the observed distribution from simulated and real-world data. Our research enables more accurate and robust OCT predictions and improves the adherence to due dates.