Research on AR modeling method with SOFM-based classifier applied to gear multi-faults diagnosis

Gear mechanisms are an important element in a variety of industrial applications. An unexpected failure of the gear mechanism may cause significant economic losses. Efficient incipient faults detection and accurate faults diagnosis are therefore critical to machinery normal running. In this paper a novel method is presents to enhance the detection and diagnosis of gear multi-faults based on Autoregressive (AR) Model and Self-Organized Feature Map (SOFM) neural networks. The experimental vibration data acquired from the gear fault test-bed are processed for feature extraction. Firstly the vibratory signals in normal and fault states have been analyzed by AR modeling method respectively, so state features can be extracted by AR coefficients. The AR coefficients then make up the eigenvectors which are taken as inputs for SOFM training. Meanwhile, to avoid misdiagnosis, the architecture of Learning Vector Quantization (LVQ) is employed to further fault recognition. Finally the network is tested using the remaining set of data, the identification and diagnosis of gears in nine different working conditions, such as normal, single crack, single wear, compound fault of wear and spalling and so on, have been effectively accomplished, and the recognizable rate is 100%. The diagnosis results show that the proposed method is feasible for early and combined gear faults classification.