Anomaly detection for orbiting satellites has become a paramount research focus in the aerospace domain. Data-driven methodologies, employing high-dimensional telemetry data, have exhibited significant potential for t...
详细信息
Anomaly detection for orbiting satellites has become a paramount research focus in the aerospace domain. Data-driven methodologies, employing high-dimensional telemetry data, have exhibited significant potential for timely anomaly detection without the need for pre-existing models or rules. In this paper, we introduce a novel satellite anomaly detection method, the Deconvolution-reconstructed temporal convolutional autoencoder (DRTCAE), which utilizes preprocessed telemetry data. The DRTCAE consists of a convolutional encoder and a deconvolutional decoder, each leveraging the parallelism of stacked Advanced Dilated Causal convolutional (ADCC) blocks. This innovative architecture facilitates efficient multi-scale nonlinear transformations, which are crucial for extracting time-dependent features from various telemetry data sources and enabling accurate anomaly detection. Through extensive experiments with telemetry data from the Zhuhai-1 OVS3A satellite and an unspecified GEO satellite, our findings demonstrate that the DRTCAE-based approach outperforms traditional multivariate regression techniques. In conclusion, our results underscore the DRTCAE as an exceptionally proficient method for detecting anomalies in orbiting satellites, thereby contributing to the improvement of overall reliability and safety in space missions.
temporal convolutional autoencoder has an important value of application in time-series analysis. In the paper we aim to use temporal convolutional autoencoder to help find out abnormal stocks quickly in the scenario ...
详细信息
temporal convolutional autoencoder has an important value of application in time-series analysis. In the paper we aim to use temporal convolutional autoencoder to help find out abnormal stocks quickly in the scenario of financial trading market. However, there are still two critical problems to solve during the application of temporal convolutional autoencoder. First, trading data of each stock are multidimensional time-series data, while classical temporal convolutional autoencoder only applies to one-dimension data. Second, stock trading data in a market are huge and their analysis consumes a long time, which contradicts the demand of quick decision in stock trading. To solve the above problems, we improve temporal convolutional autoencoder based on multidimensional sampling, convolution kernel generated by prior knowledge, temporal feature reuse, parallel training on clouds. All the techniques help temporal convolutional autoencoder find abnormal stocks quickly and well. Extended experiments demonstrate that our proposed temporal convolutional autoencoder could raise F1 score to more than seventy percent. The largest time efficiency of finding abnormal stocks can be increased by ninety percent as well.
Student engagement is an important factor in meeting the goals of virtual learning programs. Automatic measurement of student engagement provides helpful information for instructors to meet learning program objectives...
详细信息
Student engagement is an important factor in meeting the goals of virtual learning programs. Automatic measurement of student engagement provides helpful information for instructors to meet learning program objectives and individualize program delivery. Many existing approaches solve video-based engagement measurement using the traditional frameworks of binary classification (classifying video snippets into engaged or disengaged classes), multi-class classification (classifying video snippets into multiple classes corresponding to different levels of engagement), or regression (estimating a continuous value corresponding to the level of engagement). However, we observe that while the engagement behavior is mostly well defined (e.g., focused, not distracted), disengagement can be expressed in various ways. In addition, in some cases, the data for disengaged classes may not be sufficient to train generalizable binary or multi-class classifiers. To handle this situation, in this paper, for the first time, we formulate detecting disengagement in virtual learning as an anomaly detection problem. We design various autoencoders, including temporalconvolutional network autoencoder, long short-term memory autoencoder, and feedforward autoencoder using different behavioral and affect features for video-based student disengagement detection. The result of our experiments on two publicly available student engagement datasets, DAiSEE and EmotiW, shows the superiority of the proposed approach for disengagement detection as an anomaly compared to binary classifiers for classifying videos into engaged versus disengaged classes (with an average improvement of 9% on the area under the curve of the receiver operating characteristic curve and 22% on the area under the curve of the precision-recall curve).
To prevent potential abnormalities from escalating into critical faults, a rapid and precise algorithm should be employed for detecting power battery anomalies. An unsupervised model based on a temporalconvolutional ...
详细信息
To prevent potential abnormalities from escalating into critical faults, a rapid and precise algorithm should be employed for detecting power battery anomalies. An unsupervised model based on a temporal convolutional autoencoder was proposed. It can quickly and accurately identify abnormal power battery data. Its encoder utilized a temporalconvolutional network (TCN) structure with residuals to parallelly process data while capturing time dependencies. A novel TCN structure with an effect-cause relationship was developed for the decoder. The same-timescale connection was established between the encoder and decoder to improve the model performance. The validity of the proposed model was confirmed using a real-world car dataset. Compared with the GRU-AE model, the proposed approach reduced the parameter count and mean square error by 19.5% and 71.9%, respectively. This study provides insights into the intelligent battery pack abnormality detection technology.
Gearboxes are a crucial component of the transmission systems in many devices. Due to prolonged operation and high loads, it is inevitable that their condition will degrade over time. Therefore, intelligently dividing...
详细信息
Gearboxes are a crucial component of the transmission systems in many devices. Due to prolonged operation and high loads, it is inevitable that their condition will degrade over time. Therefore, intelligently dividing health state stages and conducting timely and effective health state assessments are essential for ensuring the safe and reliable operation of gearboxes. In response to the acoustic signals generated during the operation of gearboxes, a health state division and assessment method based on acoustic signals is proposed. Initially, fast Fourier transform (FFT) is utilized to convert the measured sound signal into a spectral signal that is relatively less disturbed by noise. Subsequently, the temporal convolutional autoencoder (TCAE) is proposed and constructed to encode and decode the spectrum signals at different moments, so that the trained encoder can be used to extract the deep features of the signals adaptively. After that, K-Means clustering method was used to automatically divide the health state of the gearbox combined with the extracted deep features. Finally, the one-dimensional convolutional neural networks (1DCNN) model is constructed and trained, and the deep features extracted by TCAE are input to identify the health state stage of the test sample, so as to realize the health state assessment of the gearbox. The experimental results show that, in the gearbox data set of three working conditions, the proposed method is closer to the health stage of manual calibration, which proves the rationality of the proposed intelligent method. The accuracy of the proposed health assessment method can reach 95%, 90%, and 90%, respectively, and the effect is obviously better than that of the more commonly used models at present, achieving effective health state assessment of the gearbox under non-destructive testing conditions.
暂无评论