Predicting cycle times in textile manufacturing using artificial neural network

Abstract

In textile manufacturing systems, manual labor is considered as necessity due to difficulty of working with a non-rigid material and constantly changing product types. Using robots which have the ability to work with such materials are still quite expensive compared to manual-labor. Since textile processes depend on human capabilities, it is hard to predict processing times, which is essential for production planning. Many textile manufacturers use time study methods for planning, however it only considers the motion related with the sewing process, causing decreased accuracy for predicted cycle times. Yet, in reality, there are many factors affecting the cycle times, such as type of sewing machine, abilities of workers, material (e.g. fabric) type and product design. Including all these factors increase the complexity of the time model, but they can be necessary to increase prediction accuracy. In this study, multilayer perceptron, which is one of the most widely used approaches in machine learning, is used to predict cycle times of a common operation in textile manufacturing, as many studies have shown that machine learning methods are more effective while dealing with many variables and complex relationships compared to statistical methods.