Towards non-destructive machine learning-based acoustic resonance testing of aluminium 2024 riveted joints - Abstract

Aug 3, 2026

A non-destructive inspection approach for aerospace riveted structures uses machine learning with acoustic resonance testing to estimate progressive damage. Conventional NDT options such as eddy current, ultrasonic and radiographic methods often demand costly equipment and skilled operators for scanning and interpretation. In fatigue tests, specimens were excited by hammer impulses and acoustic emissions were recorded over time. Feature extraction from frequency response functions and multiple ML models were evaluated, with a convolutional neural network achieving R²=0.98 for stiffness-loss prediction and 100% accuracy for defect detection.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Introduction

Non-destructive testing methods and limitations

Acoustic resonance testing as an alternative

Machine learning in acoustic resonance testing

Material & Methods

Specimen properties and rivet type

Test setup and data collection

Towards non-destructive machine learning-based acoustic resonance testing of aluminium 2024 riveted joints D. Loer, F. Holleitner, W. Flügge Fraunhofer Institute for Large Structures in Production Engineering IGP [david.loer@igp.fraunhofer.de](david.loer@igp.fraunhofer.de) Abstract In the aerospace industry, non-destructive testing is commonly used to ensure structural integrity. Methods such as eddy current, ultrasonic and radiographic testing are applied, but often require relatively expensive equipment and experienced operators for scanning and data interpretation. Acoustic resonance testing is an objective and cost-effective method that can test entire structures. This study investigated a machine learning-based acoustic resonance testing method for predicting relative stiffness loss as an indicator of progressive damage in riveted joints. Fatigue tests were performed in which the specimens were excited by a hammer impulse and acoustic emissions were measured at intervals. Two feature extraction approaches were applied to the frequency response function and different machine learning models for the prediction. The best results were obtained using a Convolutional Neural Network. It achieved an R² of 0.98 for predicting the loss of the joint stiffness and a classification accuracy of 100 % for defect detection based on these predictions. 1 Introduction Riveting is a common technology in the aerospace industry for joining materials such as aluminium and fibre-reinforced plastics [1]. Rivet pilot holes induce high local stress in the material [2], creating geometric notches that might become the initiation of structural damage such as fatigue cracking [3] . Wrought aluminium alloys, widely used in structural lightweight applications, are particularly sensitive to crack initiation at riveted joints due to their low fatigue crack growth threshold [2] . As a result, structural health monitoring is a standard practice in the aerospace industry to ensure structural integrity. Aircraft undergo several inspections during their lifetime, such as the D-check every 30 000 flight hours [4], where safety critical areas such as riveted joints are inspected in detail. Non-destructive testing (NDT) methods, including eddy current testing (ECT) [5], ultrasonic testing [6] and radiography [7], are commonly used to detect potential defects without causing damage to the structure itself. In ECT, pulsed eddy currents are induced in a conductive material by an alternating magnetic field [5] . Defects such as cracks, pores or corrosion disrupt the flow of these eddy currents, causing changes in the signal. This can be used to determine the size, depth and location of a flaw. Ultrasonic testing, on the other hand, uses high-frequency sound waves to detect internal flaws in materials [6] . By measuring the echo of these sound waves and analyzing variations in the reflection time, this method can identify defects based on differences in the acoustic properties of the material and its imperfections [6] . In radiography, X-rays penetrate the material and create an image based on the different densities of the material and its inhomogeneities. However, these methods also have disadvantages. They require expertise to accurately scan and interpret the test results, leading to potential human error [8] . Radiographic testing is costly and less effective for small, near-surface cracks [6], while ECT and ultrasonic methods can be challenging to apply in hard-to-reach areas, such asthe aircraft wing box [7] . In addition, rivets can restrict the application of transducers, potentially obscuring defects close to the rivet. Acoustic resonance testing (ART), on the other hand, is capable of detecting defects throughout the entire component as any structural change will alter the resonance response. This eliminates the need for precise scanning [8]. ART is a potential alternative for structural health monitoring in the aerospace industry as it is fas

FAQs

What problem does the study address in aerospace non-destructive testing?

Conventional NDT methods are accurate but often costly, require experienced operators, and can be difficult to apply in hard-to-reach regions. The study targets these limitations by using machine learning-based acoustic resonance testing for riveted joints.

How were the fatigue tests performed and what data were collected?

Specimens were excited using a hammer impulse, and acoustic emissions were measured at intervals during fatigue testing. These measurements were then used for feature extraction and model training.

Which machine learning model produced the best defect detection performance?

A convolutional neural network provided the best overall results, achieving R²=0.98 for predicting joint stiffness loss and 100% classification accuracy for defect detection based on the predictions.

Aug 3, 2026

Upload by Valentina