TY - GEN
T1 - Overload prediction of hard-target penetration fuze based on the relevance vector machine and power exponential grey model
AU - Shao, Zhihao
AU - Xie, Yucen
AU - Wang, Yabin
AU - Fang, Anqi
AU - Gao, Wangxin
N1 - Publisher Copyright:
© 2026 IEEE.
PY - 2026
Y1 - 2026
N2 - The penetration overload signal is critical for the triggering reliability of high-value projectile systems. However, signal aliasing and scarce prior samples often lead to inaccuracies in estimating the overload features critical for detonation timing. This work is motivated by the idea that real-time prediction can generate dynamic prior information to bridge this data gap, providing reference information for timing judgment in fuze system. Accordingly, a novel prediction method combining a Relevance Vector Machine (RVM) and a Power Exponential Grey Model (PEGM) is proposed. In this RVM-PEGM framework, RVM's strong nonlinear fitting capability is leveraged for initial prediction, while PEGM is employed to correct the prediction errors and compensate for random disturbances, thereby enhancing overall modeling robustness. The proposed approach was validated using overload data from two independent penetration test cases, with the original RVM and the long short-term memory network (LSTM) methods used as the baselines. The results demonstrate its superior accuracy and generalization under the tested conditions. Furthermore, comparisons of single-step and multi-step predictions confirm the method's numerical stability, establishing it as a reliable signal processing tool for overload prediction and a critical step towards state estimation for adaptive fuze system.
AB - The penetration overload signal is critical for the triggering reliability of high-value projectile systems. However, signal aliasing and scarce prior samples often lead to inaccuracies in estimating the overload features critical for detonation timing. This work is motivated by the idea that real-time prediction can generate dynamic prior information to bridge this data gap, providing reference information for timing judgment in fuze system. Accordingly, a novel prediction method combining a Relevance Vector Machine (RVM) and a Power Exponential Grey Model (PEGM) is proposed. In this RVM-PEGM framework, RVM's strong nonlinear fitting capability is leveraged for initial prediction, while PEGM is employed to correct the prediction errors and compensate for random disturbances, thereby enhancing overall modeling robustness. The proposed approach was validated using overload data from two independent penetration test cases, with the original RVM and the long short-term memory network (LSTM) methods used as the baselines. The results demonstrate its superior accuracy and generalization under the tested conditions. Furthermore, comparisons of single-step and multi-step predictions confirm the method's numerical stability, establishing it as a reliable signal processing tool for overload prediction and a critical step towards state estimation for adaptive fuze system.
KW - Machine learning
KW - Overload prediction
KW - Power exponential grey model
KW - Relevance vector machine
UR - https://www.scopus.com/pages/publications/105054251936
U2 - 10.1109/AANN71145.2026.11709119
DO - 10.1109/AANN71145.2026.11709119
M3 - Conference contribution
AN - SCOPUS:105054251936
T3 - 2026 6th International Conference on Advanced Algorithms and Neural Networks, AANN 2026
SP - 743
EP - 749
BT - 2026 6th International Conference on Advanced Algorithms and Neural Networks, AANN 2026
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 6th International Conference on Advanced Algorithms and Neural Networks, AANN 2026
Y2 - 7 August 2026 through 9 August 2026
ER -