Prediction of Remaining Useful Life for Aircraft Engines Using Convolutional and Recurrent Neural Networks

Prognostics and health management plays a key role in aerospace systems due to their complexity and the consequences it has on safety and reliability of the system. The project extends on the work conducted using convolutional layers to extract higher level features from raw sensor data. A hybrid model using convolutional layers and recurrent layers is tested on the CMAPPS aircraft engine dataset simulated by NASA. In order to automate the hyperparameter selection process, a genetic algorithm is implemented. The hybrid model was able to make RUL predictions with RMSE values that were competitive, and often times better, when compared to other models in literature. A time window approach was adopted, and various values were tested to show the effectiveness of different hybrid models with convolutional layers. The work conducted in this project with hybrid models showed great promise in RUL prediction using raw sensor aircraft engine data.