Predicting Production Cycle Time in a Real Disposable Medical Device Manufacturing System Using Semi-Supervised Deep Learning

Abstract

Manufacturing lot cycle time is the period required by a manufacturer for completion of a production process. It is an essential factor for determining the success of most manufacturing organizations, yet most research is based on studies made almost exclusively in the semiconductor industry and does not attempt to utilize the complete potential of recent breakthroughs in computational learning. Using real data collected from a medical device manufacturing company, this paper demonstrates the applicability of a semi-supervised deep learning framework for highly accurate cycle time prediction, using stacked Denoising Autoencoders to form fully connected deep neural networks and Convolutional Neural Network models. The proposed strategies for cycle time prediction can have a significant impact on product design decision optimization within the system which, in turn, facilitates reduction of costs, energy use, and the overall environmental impact.