Gearbox is the key functional unit in a mechanical transmission system. As its operating condition being complex and the interference transmitting from diverse paths, the vibration signals collected from an individual sensor may not provide a fully accurate description on the health condition of a gearbox. For this reason, a new method for fault diagnosis of gearboxes based on multi-sensor data fusion is presented in this paper. There are three main steps in this method. First, prior to feature extraction, two signal processing methods, i.e. the energy operator and time synchronous averaging, are applied to multi-sensor vibration signals to remove interference and highlight fault characteristic information, then the statistical features are extracted from both the raw and preprocessed signals to form an original feature set. Second, a coupled feature selection scheme combining the distance evaluation technique and max-relevance and min-redundancy is carried out to obtain an optimal feature set. Finally, the deep belief network, a novel intelligent diagnosis method with a deep architecture, is applied to identify different gearbox health conditions. As the multi-sensor data fusion technique is utilized to provide sufficient and complementary information for fault diagnosis, this method holds the potential to overcome the shortcomings from an individual sensor that may not accurately describe the health conditions of gearboxes. Ten different gearbox health conditions are simulated to validate the performance of the proposed method. The results confirm the superiority of the proposed method in gearbox fault diagnosis.
An Adaptive Multi-Sensor Data Fusion Method Based on Deep Convolutional Neural Networks for Fault Diagnosis of Planetary Gearbox
