Autonomous quality monitoring for complex manufacturing process

Abstract

This research investigates the possibilities of using different types of incremental learning algorithms and deep neural networks to monitor the quality of products produced from the complex manufacturing process through binary and multi-class classification to identify defects quickly. To find out which algorithm or neural network has the best result, I compared the performance of 4 different algorithms and 3 different neural networks using a set of time-series data collected as a result of the production of a transparent mold from the Molding machine. The performance is done by comparing with binary classification results from 4 different test settings. Additionally, the neural networks will also be tested with multi-class classification. The result has shown that deep neural networks generally performed better than incremental

learning algorithms with a prediction accuracy of >93% in all test settings in both binary and multi- class classification. However, the prediction accuracy is slightly lower in the multi-class

classification. Therefore, future work such as combining Convolutional Neural Network (CNN) and the deep neural network can be explored to boost the accuracy of the multi-class classification further.