TY - GEN
T1 - A study on the influence of signal number on performance of aakr
AU - Li, Wei
AU - Qi, Zhenfeng
AU - Chen, Juan
AU - Yuan, Yidan
AU - Du, Shuhong
N1 - Publisher Copyright:
© ESREL2020-PSAM15 Organizers.Published by Research Publishing, Singapore.
PY - 2020
Y1 - 2020
N2 - Considering that the performance of a condition monitoring model directly determines the final execution effect of the CBM technology, this paper focused on the impact of the number of signals on the performance of a condition monitoring model. In the study, the influence factors except the number of signals used in the model are controlled by using the same training data, the same condition monitoring algorithm(AAKR) with optimal hyper-parameters determined according to the same standard, and the average performance of models formed by multiple random extractions of the same number of signals from the training data. The calculation results show that, overall, the performance of the model deteriorates as the number of selected signals increases. This may be because, as the number of signals increases, the dimensions of the space in which the training sample points are located in also increases. For a certain number of samples, the more dispersed in the higher dimensional space, and this is very disadvantageous for the AAKR method which essentially obtains the predicted value by interpolation. As a result, the performance of the model is degraded.
AB - Considering that the performance of a condition monitoring model directly determines the final execution effect of the CBM technology, this paper focused on the impact of the number of signals on the performance of a condition monitoring model. In the study, the influence factors except the number of signals used in the model are controlled by using the same training data, the same condition monitoring algorithm(AAKR) with optimal hyper-parameters determined according to the same standard, and the average performance of models formed by multiple random extractions of the same number of signals from the training data. The calculation results show that, overall, the performance of the model deteriorates as the number of selected signals increases. This may be because, as the number of signals increases, the dimensions of the space in which the training sample points are located in also increases. For a certain number of samples, the more dispersed in the higher dimensional space, and this is very disadvantageous for the AAKR method which essentially obtains the predicted value by interpolation. As a result, the performance of the model is degraded.
KW - Auto-associative kernel regression(AAKR)
KW - Auto-sensitivity
KW - Condition monitoring
KW - Cross-sensitivity
KW - Tennessee Eastman Process mode
KW - The condition-based maintenance
UR - https://www.scopus.com/pages/publications/85107314687
U2 - 10.3850/978-981-14-8593-0_4144-cd
DO - 10.3850/978-981-14-8593-0_4144-cd
M3 - 会议稿件
AN - SCOPUS:85107314687
SN - 9789811485930
T3 - Proceedings of the 30th European Safety and Reliability Conference and the 15th Probabilistic Safety Assessment and Management Conference
SP - 1683
EP - 1687
BT - Proceedings of the 30th European Safety and Reliability Conference and the 15th Probabilistic Safety Assessment and Management Conference
A2 - Baraldi, Piero
A2 - Di Maio, Francesco
A2 - Zio, Enrico
PB - Research Publishing, Singapore
T2 - 30th European Safety and Reliability Conference, ESREL 2020 and 15th Probabilistic Safety Assessment and Management Conference, PSAM15 2020
Y2 - 1 November 2020 through 5 November 2020
ER -