Incremental Weighted Support Vector Data Description Method for Incipient Fault Detection of Rolling Bearing

Incipient fault detection is a key technical link in the field of rolling bearing prognostic and health management. The traditional incipient fault detection models are generally built on offline data and unable to update timely for matching the online data of target bearing. Moreover, when using the anomaly detection algorithm represented by support vector data description (SVDD) for incipient fault detection, it is easy to cause high false alarm rate due to slight and anomalous fluctuation of online data. To solve the above problems, an incremental weighted support vector data description (IW-SVDD) is proposed for incipient fault detection of rolling bearing. First, we train an initial SVDD detection model based on a small amount of online data that exists at initial stage, and use this model to pre-detect the sequentially-arrived online data. Second, in order to adapt the detection model to the anomalous fluctuations of online data, we design a strategy to determine the sample state. This strategy divides the fluctuation of online data into four states: abnormal appears, abnormal appears in succession, abnormal disappears and abnormal re-appears. Then we assign proper weights on the corresponding samples in each state. Finally, we update training set repeatedly by replacing the earliest samples in the training set with the same amount of samples which violates KKT condition in pre-detection results. In this way, the detection model is re-trained in incremental mode. Experiment results on IEEE PHM Challenge 2012 show that the proposed IWSVDD model can effectively reduce false alarm rate with ensuring detection accuracy.

[1]  Ruqiang Yan,et al.  Machine health diagnosis based on approximate entropy , 2004, Proceedings of the 21st IEEE Instrumentation and Measurement Technology Conference (IEEE Cat. No.04CH37510).

[2]  Yaguo Lei,et al.  Machinery health prognostics: A systematic review from data acquisition to RUL prediction , 2018 .

[3]  Robert P. W. Duin,et al.  Support Vector Data Description , 2004, Machine Learning.

[4]  Gang Yin,et al.  Online fault diagnosis method based on Incremental Support Vector Data Description and Extreme Learning Machine with incremental output structure , 2014, Neurocomputing.

[5]  Fuchun Sun,et al.  Building feature space of extreme learning machine with sparse denoising stacked-autoencoder , 2016, Neurocomputing.

[6]  Jun-Geol Baek,et al.  Density weighted support vector data description , 2014, Expert Syst. Appl..

[7]  Lovekesh Vig,et al.  Long Short Term Memory Networks for Anomaly Detection in Time Series , 2015, ESANN.

[8]  Mahmood Fathy,et al.  Adversarially Learned One-Class Classifier for Novelty Detection , 2018, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[9]  Jürgen Schmidhuber,et al.  Long Short-Term Memory , 1997, Neural Computation.

[10]  Hai Qiu,et al.  Wavelet filter-based weak signature detection method and its application on rolling element bearing prognostics , 2006 .

[11]  Chee Kheong Siew,et al.  Extreme learning machine: Theory and applications , 2006, Neurocomputing.

[12]  Jieping Ye,et al.  Multi-Task Feature Learning Via Efficient l2, 1-Norm Minimization , 2009, UAI.

[13]  Yu Cheng,et al.  Early Fault Detection Approach With Deep Architectures , 2018, IEEE Transactions on Instrumentation and Measurement.