Accurate remaining useful life (RUL) estimation is crucial for the maintenance of complex systems, e.g. aircraft engines. Thanks to the popularity of sensors, data-driven methods are widely used to evaluate RULs of systems especially deep learning approaches. Though remarkably capable at non-linear modeling, deep learning-based prognostics techniques lack powerful spatio-temporal learning ability. For instance, convolutional neural networks are restricted to only process grid structures rather than general domains, recurrent neural networks neglect spatial relations between sensors and suffer from long-term dependency learning. To solve these problems, we construct a graph structure on sensor network with Pearson Correlation Coefficients among sensors and propose a method for combining the power of graph convolutional network on spatial learning and sequence learning success of temporal convolutional networks. We conduct the proposed method on aircraft engine dataset provided by NASA. The experimental results demonstrate that the established graph structure is appropriate and the proposed approach can model spatio-temporal dependency accurately as well as improve the performance of RUL estimation.

Accurate remaining useful life (RUL) prediction is of great significance for maintaining the safety and reliability of many industrial systems. In recent years, deep learning based-methods predict the RUL by automatically learning and fusing degradation features from signals and have shown great potential in improving prediction accuracy. However, these methods only focus on capturing degradation information from the sensor signals in the time domain while ignoring the characteristics of different sensors in the spatial domain. Our key motivation is that spatial characteristic is also critical in RUL prediction and cast the RUL prediction problem into a deep spatio-temporal graph fusion problem, which consists of two parts, i.e., spatial structures learning and spatio-temporal information fusion. We propose a framework, namely, adaptive spatio-temporal graph neural network (ASTGNN) to solve the above problem. In the spatial structures learning part, we propose two spatial graph convolution layers, i.e., ASTGNN-M and ASTGNN-A to learn the spatial structures adaptively from the time-varying signals. In the spatio-temporal information fusion part, we use dilation convolution to alleviate the over-smoothing problem when encountering long sensor signals. Moreover, based on ASTGNN, a model named adaptive spatio-temporal hypergraph neural network (ASTHGNN) is proposed for high-order spatio-temporal feature learning. The performances of ASTGNN-M, ASTGNN-A and ASTHGNN are investigated on the C-MAPSS turbofan engine dataset. Experimental results show that ASTHGNN achieves superior performance compared with the state-of-the-art methods and the proposed methods can effectively learn the graph and hypergraph structures of sensor signals.

This paper presents a parametric low-dimensional (LD) representation learning method that allows to reconstruct high-dimensional (HD) input vectors in an unsupervised manner. Under the assumption that the HD data and its LD representation share the same or similar local sparse structure, the proposed method achieves reconstructible dimensionality reduction via jointly learning dictionaries in both the original HD data space and its LD representation space. By regarding the sparse representation as a smooth function with respect to a specific dictionary, we construct an encoding-decoding block for learning LD representations from sparse coefficients of HD data. It is expected that this learning process preserves the desirable structure of HD data in the LD representation space, and simultaneously allows a reliable reconstruction from the LD space back to the original HD space. In addition, the proposed single layer encoding-decoding block can be easily extended to deep learning structures. Numerical experiments on both synthetic data sets and real images show that the proposed method achieves strongly competitive and robust performance in data DR, reconstruction, and synthesis, even on heavily corrupted data. The proposed method can be used as an alternative approach to compressive sensing (CS); however, it can outperform the traditional CS methods in: 1) task-driven learning problems, such as 2-D/3-D data visualization, and 2) data reconstruction at a lower dimensional space.

Synthetic aperture radar (SAR) image classification plays a key role in SAR interpretation. Due to the cost and difficulty of truth labeling for SAR images, the newly labeled samples available for image classification are very limited. This paper focuses on defining a new sample labeling method to solve the problem of truth acquisition for training data in SAR image classification. An efficient classification framework for high-resolution SAR images is presented in this paper, which is built on learning from uncertain labels. We use grid labeling for rapid training data acquisition by assigning a label to a group of neighboring pixels at a time. A novel SVM-based learning model is proposed to optimize the uncertain training data within the constraints of label proportions in each group and then predict the label of each sample for the test data based on the optimized training set. This work intends to explore a rapid labeling method called grid labeling for efficient training set definition and apply it to large-scale SAR image classification. The model demonstrates good performance in both accuracy and efficiency for scene interpretation of high-resolution SAR images.