Detecting change point in Individual Control Chart using Neural Networks

Abstract

A control chart is a common tool that enterprises use to monitor their process stability Because a control chart can effectively decrease costs many different types of control charts have been developed for different purposes However not all industries can use common control charts Due to the process itself or sampling method these industries only sample one product at a time so an individual control chart is designed for these situations Because of a small sampling size individual control charts have two problems First an individual control chart will delay signaling of an alarm because its control limits are looser than other alternatives that decrease the probability of a false alarm rate Second it can not be used to calculate statisics such as mean and variance so it is difficult to learn more about the shifted data process To resolve these problems artificial neural networks (ANNs) are used in this study to build two models The first model is a detection model that uses a convolution neural network (CNN) to find the change point information including what time and how much of the margin the process changed This model developed can achieve a 90% accuracy rate in terms of changing time and a 95% accuracy rate for the margin if two observation errors are allowed The second model is a monitor model that uses a long short-term memory network (LSTM) that can be substituted for the role of the individual control chart The proposed method is more effective than traditional methods

Date of Award	2019
Original language	English
Supervisor	Tai-Yue Wang (Supervisor)