Condition Monitoring of Internal Combustion Engine Using EMD and HMM

Abstract

The acoustic signature of an internal combustion (IC) engine contains valuable information regarding the functioning of its components. It could be used to detect the incipient faults in the engine. Acoustics-based condition monitoring of systems precisely tries to handle the questions and in the process extracts the relevant information from the acoustic signal to identify the health of the system. In automobile industry, fault diagnosis of engines is generally done by a set of skilled workers who by merely listening to the sound produced by the engine, certify whether the engine is good or bad, primary owing to their excellent sensory skills and cognitive capabilities. It would indeed be a challenging task to mimic the capabilities of those individuals in a machine. In the fault diagnosis setup developed hereby, the acoustic signal emanated from the engine is first captured and recorded; subsequently the acoustic signal is transformed on to a domain where distinct patterns corresponding to the faults being investigated are visible. Traditionally, acoustic signals are mainly analyzed with spectral analysis, i.e., the Fourier transform, which is not a proper tool for the analysis of IC engine acoustic signals, as they are non-stationary and consist of many transient components. In the present work, Empirical Mode Decomposition (EMD) and Hidden Markov Model (HMM)- based approach for IC engine is proposed. EMD is a new time-frequency analyzing method for nonlinear and non-stationary signals. By using the EMD, a complicated signal can be decomposed into a number of intrinsic mode functions (IMFs) based on the local characteristics time scale of the signal. Treating these IMFs as feature vectors HMM is applied to classify the IC engine acoustic signal. Experimental results show that the proposed method can be used as a tool in intelligent autonomous system for condition monitoring and fault diagnosis of IC engine.