More Hierarchical Intelligent Diagnosis And Early Fault Indication Methods Of Multi Rotor-Bearing

With the development of mechanization, the ties between modern machinery and equipment become more closer.As the power unit of the modern equipment, rotating machinery has a complex system structure, its development trend towards to high speed.As an important part of rotating machinery parts such as bearings and gears are often damaged,the rotor-bearing system can produce all kinds of faults.It will affect the overall mechanical operation condition, research on the rotor-bearing system has important role for the actual production and life.The fault diagnosis technology is usually used in the actual production, fault diagnosis technology incorporates sensor application,signal processing method,artificial intelligence and computer technology. Based on the detection of rotor and bearing parts, we can monitor the working state of the rotor-bearing system. When different parts havefailure, which caused the frequency and amplitude of fault signal have limits with various structural parameters. When different structures have failure, the corresponding fault signals may present particular forms, through the analysis of fault signal we can get failure type and characteristics of the rotor-bearing system.First of all, in view of the different fault types of rotor-bearing system, this paper introduces the basic wavelet denoising method, because of the wavelet denoising shortage, the wavelet denoising method based on the adaptive threshold is adopted, the effect of the first method is more apparent than the traditional wavelet denoising method, which can extract weak signal from the noise background, which can effectively solve the noise modal aliasing, and the denoising effect is better.Secondly, the features of fault signals are extrated respectively, such as the parameters from the time domain, the frequency domain, the demodulation domain, the wavelet-packet-energy spectrum domain, which compose the parameter set that can reflect the rotor-bearing system feature. KPCA and the distance measuring method are used to compress dimension of the parameter set, for raising the efficiency and reducing the redundancy of data.KPCA can effectively solve the space dimension of nonlinear parameter set and identify the failure dynamically, and the distance measurement method is based on the size of the distance between the characteristic parameters. By improving the evaluation function of the traditional algorithm, we can effectively reduce the dimensions of the parameter set and recognize the patterns of the compressed data set of real-time.Finally we adopt the advanced neural network, the improved ant colony algorithm and the PNN model to identify and predict the fault types. Traditional BP neural network weights was optimized by using genetic algorithm, the probability of trapped in local optimal solution was reduced.Using ant colony algorithm can search the optimal solution of optimization. Using PNN method has higher training speed and accuracy relative to the above two methods, and the stability and relatively were increased, can effectively identify multiple rotor-bearing system fault type.Using the method of GRNN and WNN methods to forecast the failure mode.Eventually, the intelligent fault diagnosis precess was integrated to be a system, which can analyze hierarchically complex information of rotor bearing, achieving the accurate identification result and the purpose of the intelligent diagnosis.