Real-time Multi-resolution Decomposition of Degrading Fault Signals using Entropy measure

Trend identification in gradually degrading faults poses a challenge due to embedded local large data variations in signals, which sometimes can interfere with the gradual trend. This paper describes combining Multi-resolution Signal Decomposition (MSD) property and denoising effect of wavelet transforms with entropy measure to eliminate local high frequency variations and retain the gradual pattern present in the data. Information content in wavelet coefficients at every level given by entropy value serves as a measure for roughness present in the signal, and is explored as a means to automatically optimize number of MSD levels. Such automation is critical in real-time applications, since further decompositions than required can lead to signal pattern distortion. Trending algorithm using basic statistical measures is then applied on the wavelet-approximated signal. The proposed methodology was tested for degrading faults in gas turbines. Encouraging results were obtained, demonstrating the utility of proposed novel approach for monitoring applications. Keywords—entropy, multi-resolution decomposition, real-time applications, wavelets