【導(dǎo)讀】到舉足輕重的作用，然而滾動軸承又是其中極易損壞的元件之一。械的正常工作有重要意義。其中聚類算法在滾動軸承中得到了廣泛的應(yīng)用?？焖俚木垲愃惴?。的驗證實驗結(jié)果的正確性，做到對錯誤的檢測，從而可以減少經(jīng)濟損失。證實驗結(jié)果的正確率。減少摩擦損失的目的，是一種精密的機械元件。滾動軸承一般由內(nèi)環(huán)、外環(huán)、滾

　　

【正文】 d not get rid of the limitations of the Fourier analysis. EMD according to the local characteristics of the signal, adaptive to a plex signal is deposed into a series of intrinsic mode function IMF, each IMF ponent corresponds to a real physical significance. Window Fourier Transform, WignerVille, wavelet transform contrast, EMD is an adaptive multiresolution analysis method, it is very suitable for the analysis of nonlinear and nonstationary signals [8]. Many scholars on EMD and pattern recognition methods bine applied to fault diagnosis of mechanical systems, but most of the existing diagnostic methods have certain requirements for the quantity and quality of the training samples, and most of the diagnostic methods only on the type of fault classification For essential in the practical application of the degree of fault and not too involved. [9] EMD and neural work pattern recognition methods bined rolling element bearing fault diagnosis, this method requires a sufficient number of samples to train the neural work, and requires high quality training samples。 [10] proposed a kinds of diesel 16 engine vibration signal analysis method based on empirical mode deposition and support vector machine, the method also requires a certain amount of training samples with typical fault feature, in the case of a small amount of training samples, support vector machine overall classification performance will be reduced . Nonstable signal of rolling bearing vibration signal analysis for the traditional signal analysis methods do not have selfadaptability, as well as the actual samples of typical faults are not readily available, this paper presents a rolling bearing fault based on empirical mode deposition and optimization of the Kmeans clustering diagnostic methods, which can adaptively sample the fault type and degree of fault classification, this method has the advantages of algorithm is simple, and accurate results. The used data from the United States Case Western Reserve University Electrical Engineering Laboratory bench rolling bearing fault simulation [11], the bench, including a 2horsepower motor, a torque sensor and a power test meter. Bearings are located at both ends of the motor to be detected, the drive end bearing Model SKF6205 fan end bearing Model SKF6203 bearing fault point EDM from the damage point of a diameter of 0. 1778 mm, 0. 3556 mm, 0. 5334mm. Where in the outer bearing race damage points in the clock: 3 o39。clock, 6 o39。clock, 12 o39。clock in the three directions, the vibration data obtained by the collection of the acceleration sensors arranged on the motor housing, and the sampling frequency is 12 kHz, power, and speed by the torque sensor / decoder measured. The collected data are stored as *. mat format (MATLAB files). 6 Conclusion Recognize the new method based on EMD and optimization of the Kmeans clustering algorithm ball bearing fault diagnosis and fault extent. Rolling normal, the inner ring failure, failure of the outer ring and rolling the fault experimental signal analysis results show that: the empirical mode deposition signal noise can be removed to improve the signaltonoise ratio, thereby highlighting the fault characteristics。 extracted feature quantity can be accurately characterize the fault the type and degree of fault。 optimize the Kmeans clustering algorithm can correctly sample set with different fault types correctly classified。 Kmeans clustering algorithm with the same type of failure, different fault degree of sample set classification, optimization correctly pleted the classification of the different degree of fault. Bearing Fault Diagnosis Method Based on EMD and optimize the 17 Kmeans clustering on the basis of accurately extracting characteristic effectively identify the type of roller bearing fault, but also to correctly identify the same fault types, different degree of fault sample. The proposed method is based on the amplitude of the fault characteristic frequency at the feature quantity can be useful to explore further in terms of the selection of the feature quantity, in subsequent research work.