A Deep Learning Framework for Severity-Aware Fault Diagnosis in Industrial Suction Fans Using Multi-Modal Sensing
Year: 2025
Type: article
Abstract: This study presents a deep learning-based multi-label regression framework for severity-aware fault diagnosis of industrial suction fans, leveraging real-world multi-modal sensor data. The method predicts eight continuous fault similarity scores corresponding to distinct and most common mechanical faults. A comprehensive evaluation of five neural network architectures reveals that a Deep Neural Network (DNN) achieves the best overall performance. To bridge the gap between continuous predictions and practical maintenance decisions, a key and novel concept is introduced: a severity-aware fault labeling via an adaptive thresholding mechanism. This framework maps regression outputs to discrete severity labels, enabling reliable and interpretable fault detection that aligns with operational priorities. Experimental results on data from industrial radial fans demonstrate the framework’s robustness and high accuracy, particularly under dynamic operating conditions. This work provides a practical, data-driven diagnostic tool for predictive maintenance strategies in industrial settings.
Source: IEEE Access
Institution: Istanbul Arel University
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,075
OpenAlex ID: W4417284361