Complexity-entropy feature plane for gear fault detection

The Complexity-Entropy Causality Plane (CECP) is a representation space with two dimensions: normalized permutation entropy (H
_s
) and Jensen-Shannon complexity (Cjs). CECP has wide found applications in non-linear dynamic analysis to classify a given signal according to its randomness and complexity which is a motivation to investigate its application for machine fault diagnostics. In this work we extract features from vibration signals (from gear box) using CECP approach to classify normal operational condition and faulty condition. We observe that the method using CECP is able to identify the changes in underlying dynamics of the input signal, which enables high accuracy classification. The method using CECP generates two-feature vectors with minimal preprocessing of the raw signals. In addition CECP is insensitive to external noise, non-stationarity, and trends; this makes CECP a good candidate for machine fault classification.