Search

The stability and safety of the operation of an aero engine is very important for the flight of the aircraft, so its health status must be monitored at all times to ensure its safe operation. The traditional prediction algorithm for the remaining useful life (RUL) is often only shallow, and it is not possible to dig deep into the historical information of the sensor. Moreover, the traditional single sensor predicts the life, which often results in low or inefficient data utilization. Implementation issues. This paper presents a method for predicting remaining engine life based on multi-sense fusion. First, pre-process the data and extract, then, reconstruct the remaining sensor data through singular value decomposition (SVD); input the feature data into the deep GRU neural network for training to obtain the remaining engine life prediction Results; Finally, the commercial modular aero-propulsion system simulation (C-MAPSS) was selected for simulation experiments to verify the feasibility of the method.

For aero-engines, bearing fault monitoring and diagnosis can greatly improve its reliability. Based on the successful application of deep learning algorithms in the mapping of complex nonlinear systems, this paper proposes an aero-engine bearing fault diagnosis method that combines convolutional neural networks and support vector machine. Firstly, the Short-time Fourier transform is used to perform time-frequency analysis on time series signals (including nine faults and one normal) to obtain corresponding time-frequency images. Then, the convolutional neural network is used to extract the feature of the obtained time-frequency images. Furthermore, the support vector machine is used to classify and diagnose the faults from the extracted feature. Finally, based on the data set of Case Western Reserve University, the feasibility is validated. The results show that the proposed method has achieved a bearing fault recognition rate of more than 99%, which can effectively complete the fault diagnosis task.

Photometric ambient occlusion estimation is to recover the local visibility of a scene from multiple illuminations at a fixed viewpoint. Its effectiveness is highly dependent on the amount of illuminations. In this paper, we study how to reliably extract photometric ambient occlusion from sparsely sampled illuminations. We specifically propose an effective relative entropy minimization framework to encourage the consistency between the data observation and the theoretical distribution. Our extensive results on benchmark datasets show the effectiveness and accuracy of the proposed method. We also demonstrate the improvement of our method in producing high-quality reflectance.