Comparing deep learning and Fourier series models for equipment failure prediction in predictive industrial maintenance 4.0.

Current innovations in the field of smart manufacturing call for high machine availability, good product quality, and a high degree of flexibility in production processes. In addition to reducing downtime and increasing asset occupation, predictive maintenance also has the potential to improve safety and reduce maintenance costs by analyzing data to identify the root causes of equipment failures and prevent future occurrences. In this study, we develop a predictive maintenance model using both a Fourier series-based mathematical approach and a Long Short-Term Memory (LSTM) deep learning algorithm, applied to synthetic multivariate sensor dataset representing industrial equipment behavior. The goal is to identify the most accurate method for forecasting machine failures and minimizing production interruptions. Our results demonstrate that the LSTM model outperforms the Fourier series model, achieving a lower MAE (0.0385), MSE (0.1085), and RMSE (0.3294), highlighting the superior performance of data-driven sequential learning in capturing failure dynamics. Statistical analysis using a paired t-test confirmed that the performance differences were statistically significant (p < 0.001), and 95% confidence intervals were reported. Additional residual visualizations further supported the conclusions.