A Research For Diagnosis Method Of Rotary Machinery Impact Fault Signals Based On Compressive Sensing

With the fast popularization of automobiles,further improving comfortability and safety of vehicle has become an urgent need.Bearings and gears are important components which bearing load and transferring energy in automotive systems.Once those components break down,it will affect the comfortability and the riding safety of the vehicle.Therefore,it is very significant to take effective measures in advance by real-time monitoring and fault diagnosis with signal processing method.Based on the sparsity of mechanical vibration signals,the sparse representation and compressive sensing(CS)methods of the rotating mechanical impact fault signal are analyzed.Further research about dictionary learning method and CS reconstruction algorithm in in terms of fault feature extraction is carried on.In the aspect of dictionary learning methods for impact fault characteristics,a new based Kurtosis Shift-invariant K-SVD dictionary constructed method is proposed.The method selects a small optimal sample signal for impact pattern learning by kurtosis value and sliding window operation,and reduces noise interference and the calculation burden accordingly.A ‘Zeropadding time shift' strategy based on the learned pattern is performed to build the sparsest dictionary of impact fault signal.At last,the rolling bearing fault signal is sparsely reconstructed by the proposed dictionary,and the rolling bearing fault features are extracted accordingly.At the same time,the influences of algorithm parameters are discussed,and their value ranges are optimized.In the aspect of CS reconstruction algorithms,based on greedy iterative class CS reconstruction algorithms,a new method called Multi-Compressive Matching Pursuit(MCMP)is developed.Combined with Shift-Invariant impact dictionary,the new method is used for impact fault diagnosis of rotating machinery.Impulse components are enhanced by the integrating multi-compressive samples using the ‘sum-proxy' strategy,and the fault features is directly reconstructed from compressed sampling signals by combining the idea of backtracking and combination algorithm.Because each atom of the invariant sparse dictionary corresponds to only one impact,the proposed algorithm has wide applicability,which is suitable for signals under steady speed conditions and slowly time-varying conditions.In addition,the influence of noise intensity,compression ratio and number of compression groups on the performance of the algorithm are explored,which provides a reference for practical engineering applications.