Engine Mechanical Fault Diagnosis System Feature Extraction Algorithm

The system of automobile engine fault diagnosis is an intelligence instrument for an engine in debugging and operating online tests and fault diagnosis. The principle is use the vibration sensor gather the non-normal vibration signal from the automobile engine, and analysis the fault data through the computer. Finally, the result will display as the way of failure codes. This paper do research for this problem, the main work concentrates on such aspects as follows:(1) The sorts of typical vibration signal of engine fault including of the knot of valve, pole, cylinder, pole bearing, axle bearing, gear, piston pin, normal;(2) The time domain noise analysis for the engine fault signals;(3) The method of time-frequency arithmetic design for smooth and random signals based on Fast Fourier Transform;(4) The algorithm design based on periodogram power spectral analysis arithmetic;(5) The arithmetic design of feature evaluation criteria based on scatter matrix of Euclidean similarity;(6) The method research of engine fault feature extraction based on adaptive genetic strategy.This Paper proposes a time-domain analysis model based on RMS, Peak, Crest model, which would be describe abnormal vibration signals through the average energy, peak energy, range energy point of view. Further more the paper discusses the parameters of time domain inference rules amendments, as well as receive the alarm parameters and limits parameters through the training sample. It would be find a class of periodic isolate fault through frequency domain analysis method and solve the problem with fault location.Design a fast Fourier transform algorithm based on situ operation. It has high space efficiency, suitable for real-time online system, it would be gains the frequency-domain information online.Propose a power spectral analysis model and the relevant program design based on the periodogram. The algorithm based on the energy density function, through the energy point of view to make the frequency domain signal analysis, it would be describe the change trend of frequency information between the engine fault and energy.Establish a search method based on adaptive genetic strategy. The algorithm based on Euclidean similarity measure scatter matrix as the feature evaluation criteria, based on m-dimensional feature as the gene to construct chromosome string, and build the cross/ mutation operator model based on adaptive genetic strategy. As the feature selection problem with a non-continuous, multi-peak, noise domain space features, the algorithm overcomes the problem that SFFS, SFBS can not go back while select features and the B&B algorithm deal with non-continuous, multi-peak problem by the local optimum deception, it has a strong robustness, and be able to converge to the global optimal solution with a satisfied probability.Based on above, develop the relevant C-language application program. Unit and union test results shows that the algorithm and application program satisfied with the design requirements.