PSI - Issue 14

Vamsi Inturi et al. / Procedia Structural Integrity 14 (2019) 937–944 Vamsi, Sabareesh, Vaibhav/ Structural Integrity Procedia 00 (2018) 000–000

943

7

5. Feature classification The identified significant features are supplied as input to the machine learning algorithms for classification. Machine learning algorithms are broadly categorized as supervised and unsupervised approaches. In supervised machine learning approach, the data consisting of input and output classes is applied to synthesize the network. Artificial Neural Network (ANN) and Support Vector Machine (SVM) are supervised approaches. Support vector machine (SVM) is a modern supervised machine learning algorithm which works on the principle of structural risk minimization [19]. Using SVM as feature classification tool provides many additional advantages such as high precision, least generalization error, ability to deal with small number of data sets, ability of solving non-linear problem etc. [20]. SVM algorithm is applied to classify the given data. Two different classes (healthy & faulty conditions) are separated by a boundary and it orients in such a way that the boundary gets maximized which is referred to as principle of structural minimization [18]. Because of its least generalization error property with small number of sample data sets, this algorithm is generally used for fault diagnosis problems. In this current investigation, SVM algorithm is applied to classify two types of gear faults (root crack & tooth chip). For each class (gear condition), four dominant features containing 49 feature value sets are collected. The data set consisting of the four dominant features is given as input to the SVM algorithm. The classification accuracies of various sensors while the gearbox is operating at 50% of the maximum speed of the motor are shown in Fig. 5. It can be observed that the microphones have higher classification efficiency than the accelerometers. Mic 2 gives highest classification accuracy of 92.8% while diagnosing the gear tooth chip fault. The accelerometers have shown moderate classification accuracies in the fault diagnosis of gear tooth chip. Similar types of observations are observed in the fault detection of gear tooth root crack also. Classification accuracies of microphones are found to be encouraging as compared to accelerometers. The maximum classification accuracy of 86.7% is recorded for mic 2. From these observations, it can be concluded that the fault detection capabilities of acoustic signal analysis are better than vibration analysis while diagnosing the local gear faults.

Fig. 5. Classification efficiencies of different sensors

6. Conclusions In this study, an attempt has been made to diagnose the local gear faults by Empirical Mode Decomposition (EMD) approach. Experiments are performed on a lab scaled model of wind turbine gearbox having the speed ratio