A comparative analysis of machine and deep learning models for predictive maintenance and fault estimation of tools in industrial machinery within small-scale settings

This study presents a comparative evaluation of machine learning (ML) and deep learning (DL) models for predictive maintenance (PdM) in small-scale industrial systems. A low-cost Arduino-based testbed equipped with vibration, temperature, and rotational speed sensors was developed to emulate real-world conditions. The primary focus of the study is the detailed implementation and analysis of a Recurrent Neural Network with Long Short-Term Memory (RNN-LSTM). For benchmarking, two baseline models—Linear Regression and K-Nearest Neighbors (KNN)—were also implemented. According to the evaluation results, RNN-LSTM achieved the highest performance, with 95.31% accuracy, 0.047 MSE, 0.217 RMSE, 0.047 MAE, and 23.4% SMAPE. In comparison, Linear Regression and KNN yielded lower accuracies (92.30% and 93.27%) and higher error values (e.g., SMAPE of 58.7% and 41.2%). These findings confirm the superiority of RNN-LSTM in modeling temporal dependencies, while baseline models demonstrated limited generalization. Overall, the study shows that advanced DL models can be deployed on resource-constrained embedded systems, supporting the wider adoption of Industry 4.0 practices in small and medium-sized enterprises.