Perancangan Predictive Maintenance dengan Metode Long Short-Term Memory pada Robot Welding

Abstract

The role of robots in Industry 4.0 is becoming increasingly important due to their ability to replace human labor. Although robots can perform repetitive tasks quickly and accurately, effective maintenance is required to ensure they continue to function as expected. In the manufacturing industry, maintenance is generally scheduled for all operated equipment. However, this approach is considered less effective because the workload of each piece of equipment involved in the manufacturing process is uneven. This affects the normal operating time of each piece of equipment before entering different maintenance cycles. Some equipment requires faster maintenance cycles than others to avoid operational failure. Therefore, a predictive maintenance method based on equipment condition analysis has been developed. This research successfully designed and built a Machine Learning application to predict the remaining useful lifetime (RUL) of the hardware controller of a welding robot involved in the manufacturing operations at PT Toyota Motor Manufacturing Indonesia (TMMIN). The application was built using the LSTM (Long Short-Term Memory) method. This study also compared the prediction results with the Linear Regression method to determine the relative accuracy of the methods used. The application is also equipped with a dashboard to display the prediction results to machine maintenance operators. By utilizing operational equipment data for 225 days as input data for the application, it was found that the application built using the LSTM method had better accuracy in making predictions compared to the Linear Regression method. The LSTM method resulted in a Mean Absolute Error (MAE) of 0.06399 and a Mean Squared Error (MSE) of 0.00806, while the Linear Regression method resulted in an MAE of 0.14476 and an MSE of 0.02673 from the actual values provided by the machine.