Intelligent Identification Method For Characteristics Of Thermally Groen Oxide Layer

Aeroengine is known as "the Pearl on the crown of industrial manufacturing",and is the key component of the development of aerospace industry.However,its key technology is monopolized by the United States,Britain,Russia and other military powers,and China has not yet mastered it,which is a major national demand.Therefore,China set up two major special projects in 2016.The thermal barrier coating is the key protective material to ensure the reliable operation of turbine blade in high temperature environment.It is mainly used to reduce the surface temperature of substrate and resist various environmental degradation mechanisms.The service environment of TBC is very bad.Once its spalling failure,it will cause local overheating and burning damage of engine parts,which will lead to disastrous consequences.Thermal growth oxide(TGO)will be formed in TBC at high temperature,which leads to the rapid increase of growth stress at the interface of two bonding layers of TBC.If the stress of TGO interface reaches the critical value of damage,cracks and other defects will appear in the TBC.In this thesis,the working environment of TBC under close to real conditions is simulated by thermal vibration experiment.The TGO failure process is studied.Firstly,the samples after thermal vibration are cut and polished,and the TGO thickness image of TBC section is collected,and then the data set is established.Secondly,according to the complex texture of TGO image,the size of convolution kernel of vgg-16 model is adjusted.5 × 5 and 7 × 7 convolution kernels are used to increase the receptive field,and the moving stride of convolution kernel is adjusted.The improved algorithms are conv NET-1 and conv net-2,which increase the convolution kernel and the moving step size of convolution kernel to reduce the parameters.The test results show that the accuracy of vgg-16 network model recognition is about 93.40%;res net50 network model recognition accuracy rate is 79.40%;the improved conv NET-1 network convergence speed is the fastest,when the iteration is about 500 times,the model tends to converge,and the accuracy rate of model identification is about 95.50%.Compared with conv NET-1 model,the improved conv net-2 network model is unstable,and the accuracy rate of model identification is 90.20%.Finally,the experimental results are compared and analyzed.The results show that:the accuracy of vgg-16 network model recognition is about 92.01%;res net50 network model recognition accuracy rate is 85.30%;the improved conv NET-1 network model recognition accuracy rate is 93.68%,and the improved conv net-2 network model accuracy rate is 90.06%.The convergence speed of conv NET-1 network is faster,the stability of the model is better,and the accuracy rate is also improved.In this paper,a method of TBC feature recognition based on the combination of experimental test method and intelligent algorithm is proposed for the first time.It provides a theoretical basis for the detection of thermal barrier coatings.