Detecting dyeing machine entanglement anomalies by using time series image analysis and deep learning techniques for dyeing-finishing process

Abstract

Tangle anomalies represent a critical quality bottleneck in the dyeing and finishing process. The key problem is that most dyeing and finishing factories in Taiwan are small and medium-sized enterprises. The operating dyeing machines can only detect entanglement through the shutdown alarm or experienced craftsmen. In this study, machine data of operating dyeing machines are collected. The time-series data image technology of the Gramian Angular Summation Field, Gramian Angular Difference Field, and Markov Transition Field is combined with a convolutional neural network to develop an entanglement detection model. Industry manufacturers cooperating with universities collected 1,750,977 pieces of production data from 1,848 batches for verification and study. The results show that the optimal detection time for entanglement detection is 220 s, using the MTF conversion method with a moving window of 20, a parameter learning rate of 0.00005, and a batch size of 200. The results are systematically introduced into the field. It has been verified that when signs of entanglement occur, an alarm can be sent to the relevant personnel to solve the problem in the shortest time. As a result, manufacturers can greatly reduce the cost of quality loss and scheduling bottlenecks caused by downtime caused by entanglement.