A machine-learning architecture with two strategies for low-speed impact localization of composite laminates - scientific paper

Aug 3, 2026

Upload by Eliana

A machine-learning architecture is developed to integrate two strategies—data enhancement and an adaptive generation scheme—for impact localization on composite laminates. Low-speed impact response signals collected under varying working conditions are denoised using an Adaptive Sparse Noise Reduction Algorithm (ASNRA) to preserve pulse amplitude and prevent feature underestimation. A RIME-optimized Dual-layer Support Vector Regression (RDSVR) updates SVR hyperparameters in real time for stable, efficient localization, with numerical results confirming accurate and robust identification.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages30

Reading Time1 h 16 min

AccessFree to download

Introduction

Background and motivation

Challenges in signal acquisition and denoising

Regression-based impact localization

Hyperparameter optimization for SVR

A machine-learning architecture with two strategies for low-speed impact localization of composite laminates Junhe Shena , Junjie Yea,b* , Zhiqiang Qua , Lu Liua , Wenhu Yanga , Yong Zhanga,c , Yixin Chend Dianzi Liu e* a Research Center for Applied Mechanics, Xidian University, Xi’an 710071, China b Shaanxi Key Laboratory of Space Extreme Detection, Xidian University, Xi’an 710071, China c Guodian Nanjing Automation Co. , LTD. , Nanjing 211100, China d Key Laboratory of Expressway Construction Machinery of Shaanxi Province, Chang’an University e Engineering Division, Faculty of Science, University of East Anglia, Norwich , UK *Corresponding authors. Email address: [ronkey6000@sina.com](ronkey6000@sina.com) (Junjie Ye); [Dianzi.Liu@uea.ac.uk](Dianzi.Liu@uea.ac.uk) (Dianzi Liu) Abstract In this paper, a machine-learning architecture with the integration of two strategies including data enhancement and adaptive generation scheme for Impact Localization (IL) are developed to address the aforementioned issues for location identification of impacts on composite laminates. Two main contributions are included in this research: First, response signals collected from low-speed impact experiments under various working conditions are denoised using Adaptive Sparse Noise Reduction Algorithm (ASNRA), which aims at maximizing the preservation of the original signal amplitude, thereby avoiding the underestimation of pulse features during denoising. Then a RIME-optimized Dual-layer Support Vector Regression (RDSVR) method for the real-time update of hyperparameters is implemented in the machine-learning architecture to realize IL. The superior performances ofthe IL architecture over different IL models are validated throughout the numerical examples in terms of stability and efficiency. Results demonstrate that proposed architecture has the ability to realize the accurate and robust IL of composite laminates. Keywords: Composite materials; Impact localization; Machine learning; Sparse noise reduction; Optimization strategies 1. Introduction Composite materials are widely used in industrial fields, including automotive[1, 2], aerospace[3, 4], oilfields[5], military[6, 7], and others, owing to their exceptional properties of being lightweight, corrosion-resistant, and high stiffness-to-weight ratio[8]. However, during operational service, they can be vulnerable to low-energy impact. Despite of impact as low as a few joules, they can result in severe delamination of composite laminates, leading to a significant reduction in structural strength. Therefore, identifying low-speed impact location has become an essential aspect of practical engineering applications. With the rise of artificial intelligence, it has found widespread applications of Impact Localization (IL) in composite materials. Recently, many scholars have applied various machine-learning methods to IL in composite materials and obtained promising results[9, 10] . However, significant noise interference is often present during the signal acquisition process, resulting in poor-quality response signal datasets and making accurate localization challenging. Therefore, the removal of noise from response signals is a crucial strategy in the field of IL of composite material for fault diagnosis due to the presence of transient signals and noise components. In recent years, algorithms such as Empirical Mode Decomposition (EMD)[11], Ensemble Empirical Mode Decomposition(EEMD)[12], Spectral Kurtosis(SK)[13], and Wavelet Transform(WT)[14, 15] have been proven effective in enhancing the feature extraction capability of signals. Chen et al. [16] proposed a method for extracting weak fault features in rolling bearings using Improved Ensemble Noise-assisted Empirical Mode Decomposition (IENEMD) and Adaptive Threshold Denoising (ATD) . Based on Improved Adaptive Resonance Technology (IART) to remove noise components from vibration signals, Li et al. [17] designed an impro

FAQs

What are the two strategies integrated into the proposed impact localization architecture?

The architecture integrates data enhancement and an adaptive generation scheme to improve impact localization on composite laminates.

How does the method denoise low-speed impact response signals?

It uses an Adaptive Sparse Noise Reduction Algorithm (ASNRA) designed to maximize preservation of original signal amplitude and avoid underestimation of pulse features.

How are SVR hyperparameters updated in real time in the proposed approach?

A RIME-optimized Dual-layer Support Vector Regression (RDSVR) method performs real-time hyperparameter updates to realize impact localization.

Aug 3, 2026

Upload by Eliana