Application of Pearson Diversity Entropy as Prognostic Measure of Rotating Machinery

As a novel nonlinear measure, diversity entropy (DE) is a promising tool for prognosis of rotating machinery health. Being a core procedure in DE algorithm, cosine similarity (CS) focuses on the difference between the direction of two vectors instead of their distance or length. However, the direction difference can be easily affected by unwanted signal noise components, which will lead to low accuracy in DE-based signal complexity estimation. As a result, under heavy noise, DE not only fails to detect the incipient fault at the earliest stage of inception but also fails to trace the development of the fault. Aiming to solve the aforementioned problems, this paper incorporates the concept of Pearson similarity (PS) into DE calculation, instead of CS. PS measures the decentralized linear correlation between two vectors. Due to the involvement of PS, the newly proposed measure is termed as Pearson diversity entropy (PDE). Bearing run-to-failure data has been utilized to verify the performance of the proposed PDE. Result shows that the proposed PDE not only overcomes the limitations of original DE but also demonstrates better performance than conventional entropy algorithms such as permutation entropy (PE) and improved version of DE, namely multiscale diversity entropy (MDE).