GearDetectionNET: Detection of Gearbox Faults Under Different Load Conditions via 1D-CNN Architecture

Abstract

PurposeThis paper presents an improved method for early detection of gearbox failures using vibration signals and deep learning techniques. An innovative Lightweight 1-Dimensional Convolutional Neural Network (1D-CNN) model called GearDetectionNET has been developed.MethodGearDetectionNET introduces a Lightweight 1D-CNN architecture with optimized multi-scale feature extraction and adaptive regularization, specifically designed to enhance fault detection accuracy under varying load conditions. The model offers fast processing times, providing a significant productivity improvement over previous sophisticated techniques.Results and ConclusionThe results of the study show that the classification success of GearDetectionNET provides high accuracy and reliability. Broken and Healthy signals are classified with 100% accuracy, precision and recall and excellent Area Under Curve (AUC) ratio results are obtained. The results of the proposed architecture are compared with seven well-known machine learning (ML) and deep learning (DL) methods. In conclusion, this paper proves the effectiveness of the proposed 1D-CNN model for the detection of gearbox faults. GearDetectionNET achieved 100% classification accuracy, outperforming traditional machine learning methods (83.83%-87.56%) and outperforming many recent deep learning approaches on the same dataset with min. 1.27% higher prediction accuracy performance. The results obtained surpass the findings in the existing literature and offer a more efficient monitoring process in industrial applications.