Research On The Engine Condition Monitoring Based On Deep Belief Net

Aero-Engine is the power source of the airpline,as an important part of it,it is an important guarantee of safe flight of an aircraft,and the performance will also affect the economy directly.With the advances in technology and engineering,the performance of the engine is developing fast,along with the developing of the performance rapidly,the work environment of all parts of the engine has become increasingly severe,and the reliability of the components directly determine the safe operation of the aircraft,so monitoring the engine condition and diagnosis the failure to timely is extremely important.Deep belief net which compositioned by Restricted Boltzmann have been achieved great progress in pattern recognition,natural language processing and information retrieval,providing a new solution.to solve engine performance high coupling and high-dimensional data,nonlinear and non-equivalent featureBased on the analysis of the deep belief network learning method,According to the engine oil system and the gas path fault diagnosis system which with the desired output,mainly to do the following work:Based on the deep belief network model for unsupervised learning,take the BP network as the output layer,pattern of actual consumption of lubricating oil and gas fault codes as desired output,so that achieving in the process of training the network error back-propagation;Combined oil system and gas system's QAR data,determine training sample and the parameter of network model,to further validate the feasibility and practical of the deep belief network in oil consumption forecast and gas system fault diagnosis,the oil consumption forecast system used DBN and BP while established network model,and gas fault diagnosis system used DBN and GA-BP established network model,according to system used different of results evaluation parameter,using evaluation parameter on network model for evaluation;As for oil consumption forecast,used normal QAR data of the oil system,according to parameters input parameters,the input sample space is normalized,Establishing the oilconsumption forecast system with the deep belief network based on normalized data,and build the same network layer BP network training as comparing;...