These segments were resized to a 2-D matrix of 32 by 32. The segment-wise division is illustrated in Figure 1. 1000 segments are chosen at random as test data and rest were used for training.
Classification experiments were conducted on the raw time-series data and continuous wavelet transform (CWT) of the time-varying data. Continuous Wavelet Transform is a method to decompose a real signal into a set of elementary waveforms that provide a way to analyze the signal by examining different components related to its wavelets. The Akida model obtained an overall accuracy of 98% for the classification of raw vibration data without any pre-processing and with only 15 epochs for training. In order to investigate the effects of pre-processing instead of using raw data, CWT features were extracted. Without changing any additional parameters, the classification accuracy of the Akida model improved by 1% thus resulting in an overall accuracy of 99%. This is an excellent result that provides a point of reference for application of the Akida Neuromorphic processor in bearing fault detection.
As the world shifts away from human-driven error and AI implementations are continuously providing value in all applications, the Akida Neuromorphic IP bridges the gap between AI-enabled devices and implementational feasibility.
For organizations, this translates as a new standard of product development and maintenance practices as it allows for: