Spindle System Fault Diagnosis Method Based On Manifold Learning Research

With the development of space technology, atomic energy, microelectronics,information technology and biological engineering and other emerging science technology,mechanical working accuracy became more and more strict. High-speed and high precisionNC machine is the most important foundation in precision machining. Fault diagnosistechnology of spindle system can find the fault of CNC machine early, keep the runningaccuracy, save the maintenance cost, increasing utilization and safeguard the equipments.This paper introduces the fault signal feature of key components of spindle system.Main shaft, gears and bearings are research object. Starting with the failure mechanism andsignal characteristics, new signal processing methods for spindle system are studied, andapplying the non-linear and non-stationary signal processing techniques, a series of studieson the early fault diagnosis of spindle system are carried out.According to the feature extraction techniques of non-linear and non-stationary fault inearly weak signal, through the study of EEMD method and wavelet packet decompositionmethod which deal with nonlinear, non-stationary signal, an approach to sensitive faultfeature extraction based on EEMD and wavelet packet decomposition is put forward.Through the study of Manifold learning, an approach to sensitive fault feature extractionbased on Manifold learning and time-frequency domain statistical index is proposed. At thesame time, an approach to sensitive fault feature extraction based on Manifold learning andaxis trace is presented.This paper discusses several frequently-used methods to optimize the parameters ofmanifold learning and SVM, such as grid search algorithm, genetic algorithm and particleswarm optimization. And then the paper compares their advantages and disadvantages. Inthe last, grid search algorithm applies to parameter optimization of manifold learning andSVM. An approach to fault diagnosis based on Manifold learning and SVM is proposed.A fault diagnosis system of spindle system is developed by using MATLAB andLABVIEW. The system includes a signal analytic system of main shaft and a signalanalytic system of gears and bearings. And it can accomplish the functions of dataacquisition, showing axis trace, trend analysis and off-line diagnosis.