incipient fault detection through vibration signature analysis

Abstract

Discretion is the better part of valour. Predictive maintenance of machines are the most efficient way of preventing agonizing accidents and losses. VHMNFS Vibration Signal Analysis based Hybrid multi Layer Neuro Fuzzy System has been developed to predict faults in rotating parts of machines in the nascent stage. For this purpose, real time vibration signals of a rotating machine (healthy or faulty) are acquired, analysed and matched with its vibration signature using a novel ensemble of latest, efficient and accurate signal processing and machine learning techniques. newlineModule 1 is the feature extraction module that uses IT-VMD based HHT for decomposing pre-processed vibration signals into mono-components and respective IFs and conversion to frequency domain; and DBN for extraction of required features Root Mean Square (RMS), Kurtosis (K) and Crest factor (C) for analysis from each Mode Function (MF). Currently popular FFT has been replaced with HHT for the correct treatment of non-linear and non-stationary signals; EMD in HHT has been replaced with IT-VMD for the improved and efficient decomposition of the signal into more high frequency mom-components. newlineModule 2 is the classification and fault identification module that uses Decision Tree Random Forest for identification of health status classes and randomization of the model; and lastly DENFIS for rule establishment, identification of fault and for the purpose of handling huge and continuous data. Random forest is one of the best CART and lends robustness to the output prediction system. DENFIS is a faint simulation of the human brain - which is the primal system used for fault identification. newlineOn comparison of the above mentioned system with the ones that are currently in use in the manufacturing industries, it was found that no other system can analyse vibration signals from machines that have developed faults due to reasons such as dope, cracks, load imbalance, electrical discrepancies, etc. newlineThe system can accurately predict incipient faults as compa