Identifying machinery anomalies using shape identification and classification algorithms

Abstract

The prime responsibility of anomaly detection and mitigation of rotating machinery distress lies with a certified vibration analyst. Vibration analysis is a multi-layered task and with huge number of assets in a large plant, the work can become tedious. With the introduction of a machine learning algorithm, the first layer of the task can be offloaded to an AI based module. The direct unfiltered numerical values of vibration with units such as mm/sec RMS, ips RMS, micron pk-pk, mils pk-pk which is normally used as marker for alarm and trip values. However, for anomaly detection, vibration signatures are analyzed in forms of various plots like Lissajous (Orbit) plots, band based FFT spectrums and shaft centerline plots. This paper presents a road map for detection of anomalies using a simplified multi-layered neural network technique based on geometrical functions of vibration signature plots. From a continuous streaming data, the algorithm can detect the plots of concern from a dataset of synthetic images designed to benchmark systems for understanding the spatial and logical relations among multiple shapes. This paper deals only on the cyclo-stationary-periodic signal for maintaining simplicity in understanding for readers.