Remaining Useful Life Estimation of Rotating Machines using Octave Spectral Features

Bearing failure is one of the most common causes of failure for electric machines. Acquiring the vibration data of a machine with suitably placed accelerometers is a noninvasive and widely adopted approach for obtaining information regarding the health condition of the mechanical bearings. This paper presents a robust condition monitoring method for wear state classification and remaining useful life estimation of the mechanical rolling element bearings using orthogonal vibration signals. This proposed method uses non-linear signal processing techniques in the frequency domain for feature subset selection of short-time Fourier Transform (STFT) spectra and non-linear temporal class boundaries for classification using Coarse and Weighted K-Nearest Neighbour. This method has been tested and validated using the IEEE PHM PRONOSTIA challenge dataset. The signal processing and ML based approach presented here has performed extremely well with correct classification results of up to 75.6% being achieved. This work is of significant merit and will be highly valuable for the electric machines community allowing for the implementation of a robust condition monitoring system for many industrial applications using vibration sensors.