The Aero Engine Rotor System’s Vibration Fault Diagnosis Based On The Double Grid Corrected Wave Cluster Algorithm

The rotor system is the core part of the typical rotating machinery- aero-engine,whether it is normal or not has affection on the operating state of the whole machine.In order to improve the production efficiency and avoid the serious accidents, it is of great significance to carry on the fault diagnosis and monitoring of the rotor. With the development of information technology, intellectualization has become the developing direction of the fault diagnosis. And with the dramatic increase of fault information data and the existence of knowledge bottleneck, the traditional pattern recognition method is difficult to meet the requirements of modern fault diagnosis. Developing new method to realize the pattern recognition and clustering diagnosis of rotor faults quickly and intelligently is still a research hotspot.The essence of fault diagnosis is the automatic processing and judging of the data which contains the relevant fault information. In this paper, the application of Wave Cluster method in fault diagnosis of aero-engine rotor system was explored:In consideration of the high space complexity of the Wave Cluster algorithm in the High-dimension, the improved Wave Cluster algorithm stored by the cell array is proposed. The non empty element with valid data was stored in the cell array structure,which can avoid the invalid operation of the grid cells. Meanwhile, the data content and address were set up on the same lookup table to construct the mapping relationship with the characteristic that the array structure can be used to manipulate the data of any type and any size, which reduce the space complexity of the algorithm. The rotor fault diagnosis based on the Wave Cluster algorithm is carried out with the data got by the aero-engine rotor fault simulating experiment, the results show that the space complexity of the algorithm can be reduced and the efficiency of the algorithm can be improved with the cellular storage; It can effectively to distinguish the rotor fault state by use the Wave Cluster stored by the cell array.In consider of the low precision and efficiency of the clustering with the connection define using in the tradition Wave Cluster algorithm,an improved Wave Cluster algorithm based on Breadth-First Search algorithm was presented. The algorithm integrate with the advantages of the Wave Cluster algorithm and the Breadth-First-Search(BFS) neighbor clustering algorithm, and the definition of the mesh neighbor unit of the former is used to improve the latter sensitivity that requiremanual input parameters; the setting of the cluster threshold parameters is used to improve the former defect that the division of clusters is inaccurate. Then, the analysis of the Wave Cluster, Breadth-First-Search neighbor clustering and improved Wave Cluster were done respectively with the vibration signal data got by the aero-engine rotor fault simulating experiment, the results show that the Wave Cluster based on Breadth-First Search can effectively improve the clustering accuracy and speed of the traditional Wave Cluster, reducing the computational complexity significantly.In consider of the modifiable areal unit problem(MAUP) produced by grid clustering algorithm inevitably, an improved Wave Cluster algorithm based on the double grid correction algorithm was presented. In this algorithm, the original grid clustering was performed when the K values was obtained in the maximum number of dense grid; the corrected grid clustering was performed when the K values was obtained in uniform distribution, and the best clustering results were obtained by the correction algorithm. An example of the rotor fault diagnosis based on the double grid corrected Wave Cluster algorithm shows that the accuracy of Wave Cluster algorithm is improved,which reduces the effect of grid partition and grid density on the clustering quality; The parallel processing of double grid clustering improves the speed of the algorithm, which ease the conflict between the even grid division and an uneven distribution of the data object, and provides a solution for the MAUP.