Reliability Optimization for Batik Printing Machine Maintenance Using Data Analytics by Considering a Hybrid of LSTM and GRU Methods

Abstract

For Batik production, specifically in the textile manufacturing industry, the reliability of machine maintenance is crucial for assuring production process continuity. Scheduled preventive maintenance and reactive corrective maintenance, two traditional maintenance strategies, are commonly used in manufacturing but often result in inefficiencies such as early, unneeded maintenance and late response after failure occurrence. Responding to said problems, this research aims to develop a predictive maintenance optimization model for Batik printing machine using deep learning techniques. Both Long Short-Term Memories (LSTM) and Gated Recurrent Unit (GRU) models were used and combined into a hybrid model to forecast the Remaining Useful Life (RUL) of the Batik printing machine. Of 100 maintenance check datasets collected, 80% were used for training data, while the remaining 20% were used for testing data to then produce the predicted RUL. The RUL predictions were then analyzed to determine the optimal safe buffer, which is the number of hours at which maintenance should be proactively performed before predicted failure. In account for predictive, corrective, and preventive maintenance scenarios, different values of safe buffer ranging from 16 to 30 hours were tested to identify the exact point that minimized the total maintenance costs. Based on the findings and applying the optimal safe buffer, it leads to an efficient maintenance strategy consisting of 136 maintenance actions with varying yet specific strategies for each maintenance action. This integrated approach demonstrates that predictive maintenance, supported by RUL predictions and optimal safe buffer, can prominently enhance both the operational reliability and cost efficiency of Batik printing machines, suggesting a practical alternative to traditional maintenance planning.