TY - GEN
T1 - Research on the Reconstruction Method of Missing Data of Mechanical Failure Based on Bayesian Meta-Learning
AU - Teng, Zhenpeng
AU - Hou, Yongai
AU - Wang, Biao
AU - Yi, Xiaojian
N1 - Publisher Copyright:
© Beijing HIWING Scientific and Technological Information Institute 2025.
PY - 2025
Y1 - 2025
N2 - In unmanned systems, motor failures are frequent, and multi-sensor data fusion technology is an important strategy to improve the diagnostic performance. Aiming at the common missing problem in data fusion, this paper proposes a data reconstruction method based on Bayesian meta-learning (RIBM). First, the parallel self-learning network is used to extract fault features, and a priori weighting mechanism is constructed to maintain the time-frequency characteristics and mean values of the data, and to predict the variance; secondly, based on the a priori weighting mechanism, the data reconstruction network generates the complete reconstructed data and sets constraints to ensure that the samples are as close as possible to the real data; finally, for the network bias and feature degradation caused by the high missing rate, the feature regularization based on the Bayesian neural network is employed regularization method, which uses data uncertainty to reduce the bias. The experimental results verify that the method can effectively reconstruct the data, improve the diagnostic accuracy, and outperform the existing techniques.
AB - In unmanned systems, motor failures are frequent, and multi-sensor data fusion technology is an important strategy to improve the diagnostic performance. Aiming at the common missing problem in data fusion, this paper proposes a data reconstruction method based on Bayesian meta-learning (RIBM). First, the parallel self-learning network is used to extract fault features, and a priori weighting mechanism is constructed to maintain the time-frequency characteristics and mean values of the data, and to predict the variance; secondly, based on the a priori weighting mechanism, the data reconstruction network generates the complete reconstructed data and sets constraints to ensure that the samples are as close as possible to the real data; finally, for the network bias and feature degradation caused by the high missing rate, the feature regularization based on the Bayesian neural network is employed regularization method, which uses data uncertainty to reduce the bias. The experimental results verify that the method can effectively reconstruct the data, improve the diagnostic accuracy, and outperform the existing techniques.
KW - Bayesian meta-learning
KW - Data reconstruction
KW - Missing data
UR - http://www.scopus.com/inward/record.url?scp=105002483944&partnerID=8YFLogxK
U2 - 10.1007/978-981-96-3592-4_36
DO - 10.1007/978-981-96-3592-4_36
M3 - Conference contribution
AN - SCOPUS:105002483944
SN - 9789819635917
T3 - Lecture Notes in Electrical Engineering
SP - 347
EP - 356
BT - Proceedings of 4th 2024 International Conference on Autonomous Unmanned Systems, 4th ICAUS 2024 - Volume VII
A2 - Liu, Lianqing
A2 - Niu, Yifeng
A2 - Fu, Wenxing
A2 - Qu, Yi
PB - Springer Science and Business Media Deutschland GmbH
T2 - 4th International Conference on Autonomous Unmanned Systems, ICAUS 2024
Y2 - 19 September 2024 through 21 September 2024
ER -