Research On The Turbine Blade Failure Classification Based On Simulation Data

Gas turbine is an advanced power machinery equipment,which occupies an important strategic position in the shipbuilding,aerospace and other modern industrialized fields.As a high-tech technology-intensive equipment,gas turbine technology represents a country's technological strength.The equipment is so important,but for a long time our country is dependent on imports,but Western countries have restrictions on the export of key technologies to China,so that China's development in many areas controlled by others.Due to the complexity of more advanced gas turbine technology,the working environment and is very bad,it is particularly prone to failure.After failure of conventional repair process is time consuming more impact on the overall life of the machine,there is the best technique to monitor a gas turbine,and timely maintenance and repair of the machine,increasing the efficiency of the machine.Therefore combustion engine fault diagnosis technology has become a hot research.Gas turbine fault diagnosis technology is still in the stage of theoretical research and not into practical applications,but many countries have invested a lot in this area of human and material resources.The turbine blades as the main components of the gas turbine,is the main reason for the occurrence of the fault,and thus detection and fault diagnosis through the turbine blades is one of the main research directions.To troubleshoot the blade,the first step is blade monitoring,and then analysis of the extracted features,and fault information obtained through the machine algorithm.In this paper,fault diagnosis of turbine blades depend on the analysis of the temperature signal be monitored.Firstly,the use of radiation thermometry aspects of the work of the turbine blades to collect data,and then filtered to obtain,through accurate temperature data obtained by dividing a blade.When the blade fails,the temperature signal is generally non-stationary random signal,and short-time Fourier transform and wavelet transform has limitations in dealing with non-stationary random signals.On the basis,we use of empirical mode decomposition to extract the feature vector,and empirical mode decomposition is an advanced signal analysis method for non stationary signal processing.Through the empirical mode decompositionhe for the simulated fault siguals we get many intrinsic mode functions which can be used to extract feature vectors,and feature vectors extracted by this method can be used for troubleshooting stable.Finally,we do the fault diagnosis and simulation for the extracted feature vectors based on relevance vector machine algorithm.Compared to the support vector machine and BP neural network method we find that relevance is more accurate for fault diagnosis.And further the correctness of the extraction of the feature vector is validated.