IoT-based multi-sensor fusion predictive maintenance of 3D printers using an ensemble model framework

Abstract

The rapid adoption of additive manufacturing in smart industrial settings necessitates dependable and scalable maintenance solutions. This study proposes a smart IoT-based predictive maintenance system for 3D printers, integrating edge computing, multi-sensor data collection, and machine learning. The system exploits embedded sensors to track important operational parameters such as temperature, vibration, humidity, and smoke levels. An ESP32-based edge device for data acquisition and preprocessing enables real-time data collection and low-latency response to a monitoring dashboard. An optimised hybrid ensemble learning strategy integrating threshold-based detection, Isolation Forest, Random Forest, Support Vector Machine (SVM), and Logistic Regression (LR) classifiers is proposed to improve fault detection accuracy in IoT-enabled 3D printers. Additionally, Explainable Artificial Intelligence (XAI) methods were used to enhance the interpretability of the models. Experimental evaluation of the proposed optimised ensemble model achieved an accuracy of 99.53%, precision of 96.91%, recall of 100.00%, F1-score of 98.43%, and ROC-AUC of 0.9985 on the held-out test set. Repeated print-job-level validation further demonstrated stable performance, with a mean accuracy of 99.43% across ten GroupShuffleSplit evaluations. The findings indicate the importance of multi-sensor fusion and preprocessing in improving the reliability of anomaly detection. The suggested framework outlines a cost-efficient, scalable, and smart predictive maintenance solution that can contribute to the progress of smart manufacturing systems related to Industry 4.0. Not applicable.