Feature Extraction Method Reasearch For Weak Fault Signal Of Multiple Rotor Bearing Under Complex Path

Aero-engine is the power plant of aircraft, such as airplane and airship. Main shaft bearing acts as an important role in aero-engine, it works under the complex conditions of high temperature, high pressure and high rotation speed, and it mainly responsible for the safety of aviation device. Therefore, it is of great importance for the fault diagnosis of aero-engine bearing.The feature extraction methods in this paper generally include two aspects:feature extraction based on vibration signal in one-dimension and feature extraction based on gray-scale image in two-dimension. Through these two aspects, we aim to reach the common goal of bearing fault diagnosis.In one-dimensional vibration signal fault diagnosis methods, Multivariate Empirical Mode Decomposition (MEMD) algorithm is firstly introduced in this paper. It makes use of multi-channel vibration signals to do decomposition at the same time, and which makes up for the traditional Empirical Mode Decomposition (EMD) algorithm in mode aliasing. As a result, it can help to extract the fault characteristic frequency effectively. Besides, from the perspective of entropy and energy, this paper puts forward two new characteristic parameters, the sample entropy of wavelet packet and singular value decomposition of the short-time energy. These two characteristic parameter values can be calculated easily and own the quality of anti-noise, what’s more, they are able to deal with the vibration signals in early failure or in low SNR. Finally, this paper combines the two proposed characteristic parameters with extreme learning machine (ELM) to realize the pattern recognition of rolling bearing.For the fact that gray-scale images transformed from vibration signals often have special texture features, fault diagnosis methods based on gray-scale image are proposed in this paper. When vibration signals are transformed into images, the amplitude of vibration signal turns into the gray-scale value of the transformed image, and the periodical impact in the vibration signal is converted to the changes of gray-scale value in the image. In this paper, three different image processing algorithms are discussed, including Scale Invariant Feature Transform (SIFT), Speeded Up Robust Features (SURF) and Gray-level Co-occurrence Matrix (GLCM). Through these image processing methods, feature vectors can be extracted from the transformed images, and these feature vectors can be used to realize the pattern recognition of bearing. To verify the proposed methods, several experiments have done in this paper, and the experimental results shows that the newly proposed methods are effective in bearing fault diagnosis. In addition, bearing fault diagnosis based on gray-scale images seems to show a new way for the whole fault diagnosis field.Finally, based on the theoretical research above, a simple aero-engine bearing fault diagnosis system has been developed. The fault diagnosis system is developed in the platform of Visual Studio 2010, and it combines C# with the open source of image processing library named Emgu CV. Each module has completed the debugging work, and the system proved to meet the requirement of engineering application.