Fault Diagnosis Method Of Low-speed Helical Gear For The Whole Life Cycle

Low-speed and heavy-load helical gear transmission is widely applied in large scale machinery such as industrial equipment, helicopter, wind turbine, naval vessel and et al. Because of extreme operating condition and heavy load, helical gear is vulnerable to be out of operation, leading to failure in transmission system and affecting the safety and reliability of the mechanical equipment. Therefore, fault diagnosis algoruthms of low-speed helical gear for the whole life cycle, including the faults generation, development and failure, are studied in this dissertation, providing a basis for gearbox on-condition maintenance and life prediction.As low speed helical gear failure diagnostic signal characterized with nonstationarity, nonlinearity and multi component, a time frequency analysis method based local time scale decomposition is proposed. And measures against ending effect, smoothing processing of background spectrum and the solution method of instantaneous amplitude and instantaneous frequency are also introduced in this dissertation. The local time scale decomposition method is able to reduce a complicate and unstable signal to several intrinsic time scale components with instantaneous frequency by signal features, also the constituent parts of different frequency ranges can be reflected by those components.The donoising method based on the ensmble local time scale decomposition is porposed for the helical gear fault vibration signal containing strong noise. In this method, the basic condition of white noise needed to meet has been defined, and the signal could be decomposed into a sum of several noise components and a residual signal through continuous decomposition and filtration. So, the noise components can be separated from the original signal. As the early helical gear fault feature is very weak and difficult to extract, a new fault feature extraction method based on local energy amplification of vibration signal is proposed. By this method, vibration signal of helical gear with early fault is divided into several sections, and then each section signal is performed energy averaging and amplification to realize the purpose of highlighting the fault feature. The algorithm combining the donoising method based ensemble local time scale decomposition and the vibration signal local energy amplification method is applied to early helical gear fault diagnosis.In order to deal with the strong non-stationary of the time domain sampling signal caused by speed random fluctuation and load changing in actual work process, the optical encoder is used as an external trigger of data acquisition board to proceed uniform angle sampling of gear fault vibration signal, which changes non-stationary signal in time domain to stationary signal in angular domain. The time domain synchronous averaging method is introduced into angle domain, and a fault diagnosis method based on angle domain synchronous averaging and order tracking is proposed to realize early helical gear fault extraction when speed random fluctuation. In addition, a gear tooth location method is put forward based on vibration signal effective value distribution at angle domain. This method determines the gear fault location through calculating the uniform angle sampling signal effective value distribution of each gear tooth.It is hard to obtain the failure sample data of helical gear, and the existing failure pattern recognition methods based on small sample such as support vector machine apply insufficient fault feature information and their classification accuracy is very low. The process support vector machine (PSVM) classification model using time variant function as input is built. And fault feature signals of helical gear are fit to continuous time variant functions by orthogonal basis function expansion. This mode takes advantage of helical gear fault feature information under small sample condition, and improves the accuracy of fault recognition method. The multi-classifier helical gear lifetime process fault mode recognition has been converted to the combination of several two-classifiers. The different fault modes in the whole life cycle can be recognized accurately through the process support vector machine classification model.