Prediction model to design standard production period for steel plate mills

Steel plate products are manufactured through many refining processes. Among the refining processes, there are some processes where it is determined during the production whether the plates need to go through, and the uncertainty of the production period and the workload derived by these processes make the production control of the steel plate difficult. In this paper, we propose a method to predict the standard production period that is especially important for the production control. Since black-box models are avoided in the production field, we contrived a model which first predicts the process flow of the refining processes by decision trees and then predicts the probability density function for the production period by adding up the processing periods of the transit processes. These probability density functions for the processing periods are calculated by means of a maximum likelihood estimation under normal distribution assumption, but it was found that the values of the standard production period were not as much different as those under exponential distribution assumption. Moreover, although it does not satisfy the requirements of the production field, we found that the average of the standard production periods improved 0.7 days using quantile regression forest predicting the standard production period directly.