Research On Technologies Of Aeroengine Blades Damage Image Fast Recognition

Aero-engine are working in high load, high speed and high temperature chronically,easy to cause blades damage which have enormous impact on engine safety work.In addition,the impact of wind and bird can cause blades varying degrees of damage. Through the patternrecognition of aero-engine blades damage image, the aero-engine blades damage conditioncan be fast and accurate detection and it is effective to predict malfunction problems. So, theresearch of aero-engine blades damage image fast recognition have important applicationvalue.Image segmentation and feature parameters are the first things for recognition.In thispaper, firstly, Probabilistic neural networks method for segmentation the blade damage imageof aero-engine is used. The results of experiment show that the probabilistic neural networkscan better achieve.Secondly, mathematical morphology method is used to further processsegmentation image for eliminating the noise and obtaining more accurate segmentationresults. Finally, The paper adopts the boundary tracking algorithm to extract the shapecharacteristic parameters and adopts gray level co-occurrence matrix to extract the texturecharacteristic parameters,then,the input of recognition model is obtained.As a result of the traditional blades damage detection methods can’t accurate judgespecific types of blades damage,and reliability and degree of automation is low. In this paper，the adaptive genetic algorithm is used to optimize the basis neural network parameters, anddynamic adaptive GA-RBF recognition model is established. The method is applied insimulation. The classification accuracy rate is93.33%. The method is compared with theradial basis neural network, the result show this method is more effective than the basis neuralnetwork.For the single optimized basis neural network don’t take into account the uncertainty ofaero-engine blades damage image in acquisition process. In order to further improve therecognition rate and the stability of model, the paper establish a combinatorial optimized model. That is a damage image recognition method based on decision-level informationfusion was proposed, which was based on D-S evidential theory and RBF network.Theinstance simulation shows that this combinatorial optimized model not only improve theaccurate pattern recognition ability of basis neural network,but also make full use of the D-Sevidential theory to express fuzzy information and the ability to deal with uncertain factors.Atthe same time,this method overcome the shortage of the single basis neural network whichdepend on single information and improve the robustness and stability of the identifier andaccuracy.