Machine Learning Applications for Condition-Based Maintenance of Critical Machinery Components

This study evaluates the performance of Machine Learning (ML) and Deep Learning (DL) models for Condition-Based Maintenance (CBM) of a raw mill, focusing on motor condition prediction. The work systematically analyzes and compares the predictive capabilities of Random Forest (RF) and Long Short-Term Memory (LSTM) models for three critical motors, Main Drive (MD), Separator (SR), and Fan (FN), using varying input timespans to identify the optimal approach. The RF model consistently outperformed the LSTM model across all motors, achieving superior accuracy and efficiency. Specifically, the RF model recorded an RMSE of 0.0235 and an R² of 97.9% for the MD motor, compared to the LSTM model's RMSE of 0.0858 and R² of 72.2%. Similarly, the RF model achieved an RMSE of 0.0466 and an R² of 91.1% for the SR motor, outperforming the LSTM model. While the LSTM model offers higher configurability and is well-suited for complex time-series tasks, its performance in this study was hindered by limited and noisy data, underscoring the robustness of shallow ML models like RF in such scenarios. A MATLAB-based dashboard was developed to visualize motor conditions, predict accuracy, and allow real-time updates. These results highlight RF as the preferred approach for CBM in industrial settings, offering greater consistency, computational efficiency, and ease of implementation compared to its deep learning counterpart.