Accuracy Comparison of Long Short-Term Memory and Linear Regression Methods for Welding Robots Predictive Maintenance

In the manufacturing industry, effective maintenance is required to ensure that the robots function as expected. Maintenance is generally carried out on a scheduled basis, but it is still considered ineffective due to the uneven workload of each piece of equipment. Based on this, a predictive maintenance method was developed based on equipment condition analysis. In this paper, the accuracy of predictive Maintenance in machine learning-based applications using the LR (linear regression) method and the LSTM (long short-term memory) method are compared to predict the remaining useful lifetime (RUL) of the hardware controller of one of the welding robots involved in manufacturing operations at PT. Toyota Motor Manufacturing Indonesia (TMMIN). The accuracy level is based on the mean absolute error (MAE) and mean squared error (MSE) values obtained from the results of running the application created for each method. By utilizing equipment operational data for 225 days, it was found that the LSTM method had better accuracy with MAE and MSE of 0.06399 and 0.00806, compared to the LR method, which had MAE and MSE of 0.14476 and 0.02673.