Fault diagnosis of centrifugal pumps based on the intrinsic time-scale decomposition of motor current signals

Centrifugal pumps are widely used in various manufacturing processes, such as power plants, and chemistry. However, pump problems are responsible for large amount of the maintenance budget. An early detection of such problems would provide timely information to take appropriate preventive actions. This paper investigates the application of Machine Learning Techniques (MLT) in monitoring and diagnosing fault in centrifugal pump. In particular, the focus is on utilising motor current signals since they can be measured remotely for easy and low-cost deployment. Moreover, because the signals are usually produced by a nonlinear process and contaminated by various noises, it is difficult to obtain accurate diagnostic features with conventional signal processing methods such as Fourier spectrum and wavelet transforms as they rely heavily on standard basis functions and often capture limited nonlinear weak fault signatures. Therefore, a data-driven method: Intrinsic Time-scale Decomposition (ITD) is adopted in this study to process motor current signals from different pump fault cases. The results indicate that the proposed ITD technique is an effective method for extracting useful diagnostic information, leading to accurate diagnosis by combining the RMS values of the first Proper Rotation Component (PRC) with the raw signal RMS values.