©VIBROENGINEERING. VIBROENGINEERING PROCEDIA. NOVEMBER 2013. VOLUME 2. ISSN 2345-0533 103 Enhancement of Gear Fault Detection Using Narrowband Interference Cancellation J S Kang, X H Zhang, E Bechhoefer and H Z Teng ENHANCEMENT OF GEAR FAULT DETECTION USING NARROWBAND INTERFERENCE CANCELLATION. J S KANG, X H ZHANG, E BECHHOEFER AND H Z TENG J S Kang 1 , X H Zhang 2 , E Bechhoefer 3 and H Z Teng 4 1, 2 Mechanical Engineering College, Shijiazhuang, 050003, China 3 GPMS LLC, Cornwall, VT, USA 4 Maintenance Center, Lanzhou, 730060, China E-mail: dynamicbnt@gmail.com Abstract. The development of enhanced fault detection ability for gearbox systems has received considerable attention in recent years. Detecting the gear fault easier is very important for maintenance action. This has driven the need in research for enhanced gear fault detection method. The goal is to extract periodic impulse signal from the very noise signal which mainly contains the narrowband signals. This can be done by enhancing the impulsive signals while suppressing the narrowband signals. This paper used a new method, Narrowband Interference Cancellation, to detect the gear fault. This method reserves the impulsive signal produced by gear fault and removes the other signals out. The methodology is demonstrated on a gearbox run-to-failure test. The results show that Narrowband Interference Cancellation can enable the gear fault detection easier. 1. Introduction Gearbox is a key component in drive transmission systems, such as: mining machines, helicopters, and wind turbines. Any defects in gearbox will lead to catastrophic failure and results in a loss of production. So, detecting the gearbox defects as easy as possible to avoid fatal breakdowns of machines and reduce the cost are essential. Vibration analysis is the general technology applied in gearbox condition monitoring. Gear fault detection has been researched over many years. Many features based on the time synchronous averaging (TSA) have been proposed and the detail information can be found in [1]. Wang and Wong used autoregressive (AR) model to remove the regular toothmesh pattern for enhancing the impulse signal produced by gear fault [2]. Endo and Randall [3] proposed the use of minimum entropy deconvolution (MED) technique to enhance the ability of the existing AR model based filtering technique to detect localized faults in gears. Under this fundamental, some other revised AR models are also developed to detect the gear fault under constant load [4, 5]. Then, Yang and Makis further developed a ARX model based gearbox fault detection and localization under varying load conditions. This method considers load as additional information and tests validated it can be used in real situations [6]. The main objective of AR model based filtering is to separate the deterministic signal and random signal which produced by regular gear mesh and gear fault respectively. Li et al. [7] used the energy ratio between the random components and the deterministic components as the feature for bearing prognostics under varying load and speed condition. This property assumes when the bearing operating condition varies, both the energy of deterministic parts and random parts will change in the same direction. Therefore, the energy ratio will be more robust to the varying operating condition. In order to validate this method in gear fault feature extraction under non-stationary condition, Zhang et al. [8] conducted two gearbox test-to-failure experiments and extracted the energy ratio. The results showed that energy ratio can effectively reflect the degradation trend in both stationary and non-stationary condition. The comparison to some traditional features further explained the effectiveness of this method. However, there is big room for researchers to develop more methods which can be used to enhance the gear fault detection and reduce the missing detection ratio. Recently, 2 HePing West Road 97#, the sixth department of Mechanical Engineering College, Shijiazhuang, China.