Feasible Robustness in Adaptive Monitoring of Complex Harmonic Time Series Using In-Parameter-Linear Nonlinear Neural Architectures

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21220%2F25%3A00384196" target="_blank" >RIV/68407700:21220/25:00384196 - isvavai.cz</a>

Výsledek na webu

<a href="https://doi.org/10.1016/j.procs.2025.01.287" target="_blank" >https://doi.org/10.1016/j.procs.2025.01.287</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1016/j.procs.2025.01.287" target="_blank" >10.1016/j.procs.2025.01.287</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Feasible Robustness in Adaptive Monitoring of Complex Harmonic Time Series Using In-Parameter-Linear Nonlinear Neural Architectures

Popis výsledku v původním jazyce

This work studies the meta-parameter influence on adaptive prediction of complex harmonic signals such as ball bearing vibrations and its monitoring via neural weights using in-parameter-linear nonlinear neural architecture (IPLNA). We show that Higher-Order Neural Units (HONUs) also with higher nonlinearity orders have the potential to approximate and thus monitor complex behaviour of systems with inherent linearity (such as ball bearing vibrations) without overfitting during real-time learning. We show that the presence of multicollinearities in the input state does not necessarily reduce the generalizing properties of HONU models and that the increased nonlinearity of HONUs does not necessarily introduce an overfitting problem. Therefore, due to the analogy of the mathematical architecture of HONUs to time-variant linear dynamical systems and their underlying mathematical-physical understanding, we present a basis for novel research of explainable machine learning methods for predictive maintenance.

Název v anglickém jazyce

Feasible Robustness in Adaptive Monitoring of Complex Harmonic Time Series Using In-Parameter-Linear Nonlinear Neural Architectures

Popis výsledku anglicky

This work studies the meta-parameter influence on adaptive prediction of complex harmonic signals such as ball bearing vibrations and its monitoring via neural weights using in-parameter-linear nonlinear neural architecture (IPLNA). We show that Higher-Order Neural Units (HONUs) also with higher nonlinearity orders have the potential to approximate and thus monitor complex behaviour of systems with inherent linearity (such as ball bearing vibrations) without overfitting during real-time learning. We show that the presence of multicollinearities in the input state does not necessarily reduce the generalizing properties of HONU models and that the increased nonlinearity of HONUs does not necessarily introduce an overfitting problem. Therefore, due to the analogy of the mathematical architecture of HONUs to time-variant linear dynamical systems and their underlying mathematical-physical understanding, we present a basis for novel research of explainable machine learning methods for predictive maintenance.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

<a href="/cs/project/TN02000010" target="_blank" >TN02000010: Národní centrum kompetence Mechatroniky a chytrých technologií pro strojírenství</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2025

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Procedia Computer Science

ISBN

—

ISSN

1877-0509

e-ISSN

1877-0509

Počet stran výsledku

10

Strana od-do

2268-2277

Název nakladatele

Elsevier B.V.

Místo vydání

Amsterdam

Místo konání akce

Praha

Datum konání akce

20. 11. 2024

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—