A deep variational auto-encoder based dimensionality reduction for fault diagnosis in ball bearings

One of the main challenges faced by the industry in the context of failure diagnosis is the high quantity and high dimensionality of the available data. Due to the increasing capability and availability of sensing technology, nowadays it is possible to acquire a large amount. of (unlabeled) data on many operational and maintenance related variables from monitored machines. The problem lies on how to extract useful intbrmation from such data. A standard approach in fault diagnosis is to first apply a dimensionality reduction method. In this paper, we propose a method for dimensionality reduction based on Variational Auto-Encoders (VAEs). VAEs have shown good results in areas such as image processing, image generation and speech processing. In particular, in this paper, the-VAE based method works on spectrograms generated from vibration signals measured during system's operation. This approach is applied to the fault diagnosis of ball-bearings.