Prognostics and Health Management (PHM) of the aircraft gas turbine engine is essential in the safety of the aircraft. In this paper, engine remaining useful life (RUL) was predicted with a novel architecture based on a hybrid recurrent neural network. This hybrid model trains HMM firstly and then gives a small LSTM to get distributions of HMM states. These HMM states are further trained to fill in gaps in HMM. Subsequently, a jointly trained hybrid model is constructed, which can enhance stability and accuracy of prediction significantly.

As one of the most critical wind power generation components, wind turbine blades play a key role in generating wind power. Aiming at the problem that the wind turbine blades are subjected to multiple loads in combination, the crack problem is easy to occur. Through the analysis of the macroscopic expansion mechanism and microscopic damage mechanism of short cracks and main cracks, the hidden relationship between crack appearance and damage nature is deeply explored. A fault diagnosis algorithm for wind turbine blades established on the basis of the BP neural network is raised. On the multi-discriminator fusion network structure, BP neural network algorithm is used to train the multi-feature sample data including wind turbine blades, so that the network parameters tend to convergence and gradually approach the real tag. The experimental analysis shows that the algorithm effectively diagnoses and evaluates the damage degree of the blade structure, and has a high recall rate and accuracy, which proves the effectiveness and robustness of the algorithm.