Baseline Modeling For Gas Path Parameters And State Recognition Of Aeroengine Based On Intelligent Algorithms

As the core system of the aircraft,the performance of the engine system not only affects the overall performance of the aircraft,but also determines the flight safety of the aircraft.However engine always work in harsh environments,so it is very important to develop Engine Health Management(EHM)technology.With the development of the big data technology,EHM based on flight data is becoming mainstream.At present,the two main contents of EHM based on data model are engine parameter baseline modeling and engine state recognition.Therefore,the main research of the article depends upon the above contents.(1)The pre-processing of the real flight data mainly includes the missing value filling and the removal of the abnormal value.At the same time,the flight parameters are corrected to the standard atmospheric sea level state.In order to establish a baseline model,a stable point that represents the stable operating state of the engine is extracted from the status of cruise.(2)A baseline model of engine fuel flow,exhaust temperature,fan speed,and core speed are established using Stacked Denoising Autoencoders(SDAE).By comparing with the baseline model established by the traditional BP neural network,it is found that the baseline model established by the SDAE can be benefit for improving the precision and reducing the noise.Finally,the Particle Swarm Optimization algorithm(PSO)is used to optimize the number of nodes in each layer learning rate and iteration number of the SDAE.It is found that the hyperparameters which use PSO is obviously better than the manual search hyperparameters.(3)A method for converting data of flight into image data has been investigated.This conversion method uses data from various stages of flight and then establishes an engine state recognition model through a Convolutional Neural Network(CNN).Compared with the traditional method,it shows the superiority of the state classification modeling based on CNN.Secondly,by comparing with the use of only cruise phase data,it is proved that the classification model based on the data of all phases has better effect.Finally,the PSO is still used to optimize the hyperparameters of the convolutional network,and the superiority of the PSO in the model hyperparameter selection is verified.