Intelligent Recognition of Gearbox Status by Wavelet Packet Decomposition

Energy level of signal is one of the characteristics which its ability to recognize condition of equipment, has been verified and proved. Fault in rotating devices affects the energy level of signal and its distribution type. Thus energy is an important feature for condition monitoring of rotating devices. Entropy is a concept to express uncertainly of random variables and Shannon entropy is one of most widely used manner to measure it. Data used in this research is from car gearbox in which many different pieces are involved. Data has also been collected in industrial environment. So the added noise is more than that would be the case if it was collected in laboratory condition with simple equipments containing few pieces. So the energy of these noises can significantly affect the energy level of the main signal and cause difficulty, in recognition. In contrast entropy only depends on the scatter of used data and a little change in level and distribution does not affect entropy. Data in this research has been collected from 20 car gearboxes which have been analyzed by 50 different wavelets to measure their energy and entropy as feature vector along with RBF neural network and support vector machine in recognition of the safe and damaged gearboxes. The mean and best recognition result was 57.8 and 65 percent for energy and RBF network, 59.9 and 70 percent for energy and support vector machine, 65.9 and 85 percent for entropy and RBF network and finally 71.4 and 90 percent for entropy and support vector machine. These results indicate is that entropy outperforms energy and support vector machine outperforms RBF network in condition monitoring of gearboxes.