Reciprocating compressors intelligent fault diagnosis under multiple operating conditions based on adaptive variable scale morphological filter

Abstract

Faults of reciprocating compressors (RCs) are almost observed as impacts in the vibration signal. Thus, extracting impact features is considered an effective method to monitor RC health conditions and reduce sensor investment. However, the quantity and location of impacts are challenging to identify because of the severe background noise and the broad aliasing frequency. This study proposes a fault knowledge-independent adaptive variable scale morphological filter (AVSMF) and an automatic impact localization method to extract impact features from a vibration signal, making real-time intelligent fault diagnosis possible. The AVSMF method exploits the amplitude distribution of intrinsic mode functions and morphological filters to extract the impact components in a signal. Then, to identify impacts, the demodulation energy operator of symmetrical differencing is utilized to discover instantaneous energy changes. Furthermore, a new feature called the impact phase bin (IPB), which assembles the characteristics of impacts in the form of bins, is presented to improve the fault diagnosis performance. Simulations were carried out on a signal comprising multiple impulses with different frequencies and amplitudes. The results showed the effectiveness of the proposed methods on impact localization. Additionally, the potential application in liquid slugging fault diagnosis was demonstrated. Finally, the superiority of the proposed methods was experimentally substantiated by conducting fault diagnosis tests under single and multiple operating conditions on a two-stage double-acting RC. The fault classification accuracy under the single operating condition is 95.08% using the IPB feature only, and that is 99.06% under multiple operating conditions combining conventional statistical features with the IPB feature.