Model structure mining for electromechanical actuator systems : Signal characteristic analysis using high dimensional model representation and Kolmogorov- Arnold Modelling

Tools

Alimhillaj, Parid and Minisci, Edmondo and Dalla Vedova, Matteo Davide Lorenzo and Ferro, Carlo Giovanni and Maggiore, Paolo; Guxho, G. and Spahiu, T.K. and Xhafka, E. and Gjeta, A. and Sulejmani, A., eds. (2026) Model structure mining for electromechanical actuator systems : Signal characteristic analysis using high dimensional model representation and Kolmogorov- Arnold Modelling. In: Proceedings of the Joint International Conference: 5th Conference on Engineering and Entrepreneurship and 11th Textile Conference. Lecture Notes on Multidisciplinary Industrial Engineering . Springer, ALB, pp. 166-185. ISBN 978-3-032-11085-5 (https://doi.org/10.1007/978-3-032-11085-5_16)

[thumbnail of Alimhillaj-etal-2025-Model-structure-mining-for-electromechanical-actuator-systems]
Preview
 Text. Filename: Alimhillaj-etal-2025-Model-structure-mining-for-electromechanical-actuator-systems.pdf
Accepted Author Manuscript
License: Creative Commons Attribution 4.0 logo
Download (1MB)| Preview

Abstract

This paper presents a dual-surrogate modelling framework for analysing the internal structure of electromechanical actuator (EMA) simulations under fault conditions. Specifically, it investigates how fault parameters influence the actuator signal through interpretable surrogate models. Two complementary approaches are employed: High Dimensional Model Representation (HDMR), which decomposes the output variance into additive contributions from individual parameters and their interactions; and Kolmogorov-Arnold Modelling (KAM), which constructs the output via a superposition of locally dominant univariate components, enabling regime-specific analysis. The methodology is applied to a high-fidelity EMA simulation with five degradation parameters. The output signal is evaluated at two distinct time steps to capture both early and later behaviours. The HDMR results reveal dominant electrical faults and key second-and third-order interactions, while KAM uncovers localised structures, mode transitions, and evolving sensitivity patterns across input regimes. It can be noted that discrepancies between HDMR and KAM increase at later stages, highlighting the added value of regime-aware modelling for long-term degradation analysis. Moreover, the identified regimes in the KAM represent mathematical patterns in the signal response rather than physically validated operational states. Together, these results demonstrate that HDMR and KAM offer complementary insights-global and local, additive and compositional-that improve model transparency, support diagnostic interpretation, and provide a basis for future integration into fault-aware design and real-time monitoring strategies.

ORCID iDs

Alimhillaj, Parid, Minisci, Edmondo ORCID logoORCID: https://orcid.org/0000-0001-9951-8528, Dalla Vedova, Matteo Davide Lorenzo, Ferro, Carlo Giovanni and Maggiore, Paolo; Guxho, G., Spahiu, T.K., Xhafka, E., Gjeta, A. and Sulejmani, A.
CORE (COnnecting REpositories)