The telemetry data obtained from an on-orbit spacecraft contain important information to indicate anomaly of the spacecraft. However, the large number of monitoring variables and the large amount of data points, as well as the lack of prior knowledge about anomaly due to complicated structure of spacecraft and its working conditions, pose great challenge to the anomaly detection. This article proposes an anomaly detection algorithm based on a spatial-temporal generative adversarial network (GAN) for the anomaly detection in telemetry data. This algorithm establishes a GAN-based model combining convolutional neural network (CNN) and long short-term memory (LSTM) to extract spatial and temporal features of the telemetry data, which facilitates the automatic and simultaneous representation of nonnegligible time-related characteristics of a monitoring variable and complex correlation between variables. Using these features, many kinds of anomalies including multivariate anomalies and contextual anomalies can be detected. Moreover, an anomaly score specifically designed to fit the GAN-based algorithm is also proposed to evaluate the possibility of anomaly by weighted fusion of the generator metric and the discriminator metric, which is proved to be significantly helpful to the accuracy of anomaly detection. Finally, experiments on one real telemetry data set and two public telemetry data sets are conducted, by which the proposed anomaly algorithm is demonstrated to be effective and accurate in detecting outliers in telemetry data. Comparison with three other state-of-the-art methods also reveals the advantages of our proposed algorithm.

Remaining useful life (RUL) prediction of rolling bearings brings benefits for maintenance of spacecrafts. Vibration signals are widely used for RUL prediction. However, under some situations such as high-speed rotation of bearings, vibration signals are quite easily disturbed by noise and might be tough to collect due to inappropriate installation of accelerometers. Therefore, in this paper, stator current signals are considered as health indicator for bearing RUL prediction. Based on stator current signals, feature extraction and trajectory tracking of signals suffer two challenges: 1) degradation tracking of downlinked stator current data is restricted due to low-frequency downlink from spacecrafts to stations; 2) the accuracy of health state estimation is constrained by artificial feature selection and prior knowledge about bearings. To overcome these issues, a novel RUL prediction approach is proposed in this work. An adaptive feature selection strategy based on autoregression model and backpropagation neural network is applied. Finally, an improved RUL prediction framework is introduced under down-sampled signals, which combines compressed sensing and deep learning model. A case from Paderborn university and a case from life test of bearing on control moment gyro were investigated. Experimental results support the validity of proposed approach in terms of lower prediction error than related works.

Detecting anomalies for multivariate time series is of great importance in modern industrial applications. However, due to the complex temporal dynamics in modern systems, finding a distinguishable judge criterion is hard, which makes accurate anomaly detection still a challenging task. In order to better capture the anomalous features and design a more informative judge criterion, this paper presents an unsupervised generative adversarial network (GAN) for multivariate time series anomaly detection, which highlights a novel Active Distortion Transformer (ADT) block. Different from the vanilla Transformer, the ADT block can make good use of the prior knowledge of time sequences’ overall associations by actively conducting distortion during the reconstruction of input sequences. Benefiting from the ADT block, the network simultaneously utilizes the sequence associations and reconstruction error to recognize anomalies. In the online detection phase, anomalous data points tend to be less correlated with the overall sequence and have greater reconstruction errors than normal ones, so that an irrelevance score and a reconstruction error score can be obtained. We combine the two scores to generate a more powerful anomaly score as the judge criterion. Extensive experiments are conducted on four publicly available sensor datasets, and we also make comparisons with the recent baselines. Results show that our model outperforms the recent state-of-the-art methods, demonstrating the effectiveness of our method.

Tool wear estimation and prediction are keys of maintenance decision-making for milling machine. Various discrete-state degradation models have been developed for tool wear estimation and prediction. However, previous research assume that the number of discrete wear states is fixed based on prior understanding of tool degradation process. To break this limitation, a data-driven approach based on Hierarchical Dirichlet process-Hidden Markov model (HDP-HMM) is proposed. The number of states, transition probability matrix and omission probability distribution of hidden Markov model (HMM) can be automatically updated using observation data through a hierarchical Dirichlet process (HDP). Compared with weighted HMM and Conventional HMM, experiments on real data from high-speed CNC milling machine cutters demonstrates that the proposed approach yielded greater accuracy on tool wear estimation and kept a high reliability in tool life prediction.

An effective health assessment guarantees the high accuracy of remaining useful life (RUL) prediction of machinery components. The key to health assessment is the health indicator (HI) of machinery components, which are generally constructed by feature fusion from the time and frequency domains in sensorial data. However, the existing HI construction methods are constrained due to the low-sampling rate and the susceptibility to environmental disturbances of sensorial data. To draw these issues, this article mainly proposes an HI construction method combining condensed image coding and a metrics-constrained deep learning model for improving the accuracy of RUL prediction. First, to extract more information in time series under a low-sampling rate, a recurrence plot-gray (RP-G) image coding is extended from RP images to extract and fuse both the global and local dynamics at multiple scales in time series. Then, a nested residual-convolution autoencoder (NR-CAE) model is proposed to extract degradation-related information from RP-G images containing information disturbed by the environment. Finally, to build a more solid connection between the image-based features and the RUL prediction, a metric-constrained gated recurrent unit (MC-GRU) is proposed by considering characteristics related to RUL prediction and temporal relationships in RP-G image-based features. A commercial case of wind turbine gearboxes in Liaoning, China, is investigated and demonstrates that the proposed method owns the ability to construct reliable HI of machinery components for accurate RUL prediction, using low-sampling sensorial data under environmental disturbances.

Predictive maintenance enhances the reliability and safety of industrial components and systems, which are pivotal to prognostic and health management (PHM) systems. Specifically, a remaining useful life (RUL) prediction can support the decision-making process for predictive maintenance. With the development of vision sensors and deep learning technologies, image-based RUL prediction, including acquired images from machine vision and synthetic images from image-encoding approaches, has attracted considerable attention. Against this background, this paper investigates the development trend, classification, characteristics, and typical applications of RUL prediction using image sensors and image encoding methods. Firstly, a global view is summarised from the perspective of development trends, the usage phase of RUL prediction in typical components and systems, and primary motivations of image-based methods. Subsequently, related work on acquired and synthetic images is reviewed from the perspectives of computational mechanisms, characteristics, and related research. Then, image-only processing and multimodal fusion in RUL prediction are discussed. Finally, this paper concludes with a summary of the key scientific issues and potential future research.