A method for constructing rolling bearing lifetime health indicator based on multi-scale convolutional neural networks
- 10 Downloads
The degradation of rolling bearings is complex, and traditional methods of degenerating feature degradation are highly dependent on previous research and expertise. However, the traditional methods’ ability for learning the complex relationship between degraded features and large amounts of measured data is limited. So, it is difficult to construct a single health indicator (HI) to predict the degradation state of the bearing. In order to solve this problem, a multi-scale convolutional neural network (MSCNN) is proposed, which uses the frequency signal as input to construct HI. In combination with the “bathtub curve,” the HI we proposed is constructed by the inverse hyperbolic tangent function (atanh) method which is different from the proportional HI construction method. Five kinds of processing methods, such as original vibration signal, frequency signal and three kinds of modal decomposition preprocessed signals, are tested with PHM 2012 bearing lifetime data. And the proposed method is compared with other three HI construction methods. The experimental results show that under the HI construction method of atanh, the MSCNN model with the input of frequency signal can better represent the degradation state of the bearing compared with other methods. Meanwhile, the remaining life prediction is more accurate compared.
KeywordsRolling bearing Degradation prediction Multi-scale convolutional neural network Bathtub curve
- 16.Nectoux P, Gouriveau R, Medjaher K, Ramasso E, Chebel-Morello B, Zerhouni N, Varnier C (2012) PRONOSTIA: an experimental platform for bearings accelerated degradation tests. In: Proceedings of the IEEE international conference on prognostics and health management, Denver, Colorado, USA, June 2012, pp 1–8Google Scholar
- 17.Jing L, Zhao M, Li P et al (2017) A convolutional neural network based feature learning and fault diagnosis method for the condition monitoring of gearbox. Meas J Int Meas Conf 111:1–10Google Scholar
- 18.Krizhevsky A, Sutskever I, Hinton G (2012) ImageNet classification with deep convolutional neural networks. In: International conference on neural information processing, vol 60, no 2, pp 1097–1105Google Scholar
- 19.Ioffe S, Szegedy C (2015) Batch normalization: accelerating deep network training by reducing internal covariate shift. In: International conference on international conference on machine learning. JMLR.orgGoogle Scholar