Cracked rotor diagnostics using soft computing

A study is presented for detection and diagnostics of cracked rotors using soft computing techniques like adaptive neuro-fuzzy inference system (ANFIS), artificial neural networks (ANN) and genetic algorithms (GA). A simple model for a cracked rotor is used to simulate its transient response during startup for different levels of cracks. The transient response is processed through continuous wavelet transform (CWT) to extract time-frequency features for the normal and cracked conditions of the rotor. Several features including the wavelet energy distributions and the grey moment vectors (GMV) of the CWT scalograms are used as inputs for diagnosis of crack level. The parameters of the classifiers, ANFIS and ANN, along with the features from wavelet energy distribution and grey moment vectors are selected using GA maximizing the diagnostic success. The classifiers are trained with a subset of the data with known crack levels and tested using the other set of data (testing data), not used in training. The procedure is illustrated using the simulation data of a simple de Laval rotor with a 'breathing' crack for different crack levels during run-up through its critical speed. A comparison of diagnostic performance for the classifiers is presented. Results show the effectiveness of the proposed approach in detection and diagnosis of cracked rotors.