Appropriate Feature Selection for Creating Health Index in Rotary Machines Utilizing the Prognostic Health Management System

Prognostic Health Management (PHM) is a new philosophy in maintenance that deals with the diagnosis and prognosis of failures and defects in devices PHM in rotary machines is usually utilizing the analysis of vibration signals, acoustic emission, temperature or oil analysis. By having a proper health index obtained from signal analysis, it is possible to detect system defects and prepare the device for maintenance operation. In this paper, the acoustic emission signals of a milling machine are used to detect tool wear or breakage. First, with wavelet analysis, the signal noise was reduced in order to achieve a suitable analysis to select the health index. Here, three mother wavelet functions db4, sym5 and haar and three thresholding methods are used. Research has shown that the parent functions sym5 and haar with low penalize threshold method, with 3 levels of analysis, have the lowest MSE of 0.0018 and 0.0019, respectively. In the next step, fourteen signal feature functions were extracted and compared with each other. Among the functions studied for the health index, the result showed that from healthy to unhealthy instrument in addition to the root mean square (RMS) function with 10% change, signal root square with 10%, entropy 15%, energy 28%, impact factor 33%, the maximum signal index of 48% can also be suitable criteria for the health index.