Research On Aero-engine Fault Diagnosis Based On Extreme Learning Machine

With rapid improvement of comprehensive ability, complexity and informatization, aero-engine is been difficult to accurately predict, locate and repair by traditional concepts, modes and methods, and also can’t guarantee the maintenance efficiency and benefits. Therefore, in order to detect the reliability of aero-engine, as is expected, when combining the advanced mechanical engineering systems with the tightly coupled electronic control systems, the critical system has the ability of widely detection and diagnosis.This paper presents an effective method to fault diagnosis of aero-engine by composing improved Extreme Learning Machine (ELM) in order to solve the prediction of aero-engine health problems. ELM is extended single hidden layer feedforward neural networks, which directly gives nodes instead of adjusting hidden layer of network, by reason of which can calculate and approximate arbitrary types of hidden nodes. So, it can greatly shorten the training time in the application process. ELM can not only obtain the minimum training error, but get minimum output standard, which means better generalization performance compared with traditional neural networks.The method inputs the feature extracted from the subspace into the ELM classifier to fault diagnosis of aero-engine by using the FDA and Cloud model. It further identified the corresponding fault types, and finally analyzes the failure model according to the actual data.The result is satisfactory that the experiment verified that the proposed method can classify aero-engine fault effectively. Numbers of test results proved that this improved ELM can go beyond the original algorithm. Therefore, the improved algorithm is more suitable for prediction of aviation engine health and has higher diagnosis accuracy and noise immunity.