Title :
Helicopter engine performance prediction based on cascade-forward process neural network
Author :
Yao-ming, Zhou ; Zhi-jun, Meng ; Xu-zhi, Chen ; Zhe, Wu
Author_Institution :
Sch. of Aeronaut. Sci. & Eng., Beihang Univ., Beijing, China
Abstract :
In view of the difficulty of predicting engine performance effectively in traditional methods, a prediction method based on CFPNN (Cascade-Forward Process Neural Network) is proposed. By introducing a set of appropriate orthogonal basis functions into the input space, the input functions and weight functions are expanded. The time aggregation operation of the process neurons is simplified by this way. The RBP (Resilient Back-Propagation) learning algorithm based on orthogonal basis function expansion is proposed. The CFPNN based on RBP learning algorithm is compared with FFPNN (Feed-Forward Process Neural Network) based on RBP learning algorithm and CFPNN based on ABP (Adaptive Back-Propagation) learning algorithm respectively. The results show that the CFPNN based on RBP learning algorithm possesses good convergence and high accuracy. It provides an effective way for helicopter engine performance prediction.
Keywords :
aerospace engines; backpropagation; cascade systems; convergence; feedforward neural nets; helicopters; mechanical engineering computing; ABP; CFPNN; FFPNN; RBP learning algorithm; adaptive back-propagation learning algorithm; cascade-forward process neural network; convergence; feed-forward process neural network; helicopter engine performance prediction; orthogonal basis function expansion; process neuron; resilient back-propagation learning algorithm; time aggregation operation; weight function; Biological neural networks; Engine cylinders; Engines; Neurons; Prediction algorithms; Training; Cascade-Forward Process Neural Network; Resilient Back-Propagation learning algorithm; orthogonal basis function; prediction;
Conference_Titel :
Prognostics and Health Management (PHM), 2012 IEEE Conference on
Conference_Location :
Denver, CO
Print_ISBN :
978-1-4673-0356-9
DOI :
10.1109/ICPHM.2012.6299515