A Comparative and Hybrid Machine Learning Framework for IoT-Based Predictive Maintenance of Rotating Machinery

Abstract

Predictive Maintenance (PdM) has emerged as a critical application of Industrial Internet of Things (IIoT) and artificial intelligence for improving reliability and reducing unplanned downtime in industrial rotating machinery. While existing studies demonstrate high predictive accuracy using either classical machine learning (ML) or deep learning (DL) techniques, most approaches are evaluated in isolation and fail to address deployment feasibility, interpretability, and computational constraints inherent in industrial IoT systems. This paper proposes a comparative and hybrid predictive maintenance framework that integrates feature-based machine learning models and convolutional neural network (CNN)–based deep learning models within a unified IIoT architecture. Building upon prior work on ML-based classification and vibration-based CNN time-series learning, the proposed framework systematically evaluates both paradigms across predictive performance, computational complexity, and deployment suitability. Extensive experiments using IoT-derived sensor datasets demonstrate that ensemble ML models provide efficient and interpretable solutions for edge-level deployment, whereas CNN-based models achieve superior fault sensitivity for high-frequency vibration signals. Based on quantitative analysis, a hybrid decision algorithm is introduced to guide model selection under practical industrial constraints. The results confirm that decision-oriented hybrid PdM architectures offer superior scalability and industrial applicability compared to standalone modeling approaches.