Event Title

Estimation of Bond Line Dimensions in Adhered Metal Joints Using Ultrasonic Lamb Waves: Developments Using Artificial Neural Networks

Location

Snowbird, UT, USA

Start Date

1-1-1999 12:00 AM

Description

The motivation for this work was to develop a scheme for the nondestructive quality assurance (QA) examination of adhered joints in experimental automotive body shell assemblies. The use of adhesives on automotive structures has potential for providing flexibility of design, including the use of mixed materials, reductions in vehicle weight, and improved impact and vibration performance; manufacturing cost may also be reduced. A requirement for adhered joints is that the adhesive layer dimensions are within specified ranges and this leads to a requirement to determine bond dimensions non destructively for the purposes of QA. However, in many automotive assemblies direct access to the joint region is not possible due to structure geometry and cladding. Ultrasonic Lamb waves can provide remote access to otherwise inaccessible joints, the idea being to excite such waves in the metal adherend on one side of the joint, and to receive them from a second adherend on the other side of the joint; comparison of Lamb wave signals that have traversed the joint with signals that have propagated along plane sheet could lead to a means to determine joint dimensions. The physics of Lamb wave interactions with typical automotive joints is complex and generally not amenable to inverse solutions that would yield joint dimensions directly. In this work we have applied artificial neural networks (ANNs) to relate patterns in Lamb wave signals to joint dimensions. After training, tests of the networks on adhered samples not used in training showed that the networks could recognise key bond dimensions in more than 90% of trials. Simplification of the networks (minimisation) gave improved interpolation performance in the recognition of bond dimensions not included in network training. The weights associated with trained networks could be used to identify salient features in received Lamb wave signals in a manner that gave an indication of the physics underlying Lamb wave propagation across bonded assemblies. This aspect will be discussed in a companion paper in this volume [1].

Book Title

Review of Progress in Quantitative Nondestructive Evaluation

Volume

18A

Chapter

Chapter 3: Simulations, Signal Processing, Tomography, and Holography

Section

Classification Techniques

Pages

829-834

DOI

10.1007/978-1-4615-4791-4_106

Language

en

File Format

application/pdf

This document is currently not available here.

Share

COinS
 
Jan 1st, 12:00 AM

Estimation of Bond Line Dimensions in Adhered Metal Joints Using Ultrasonic Lamb Waves: Developments Using Artificial Neural Networks

Snowbird, UT, USA

The motivation for this work was to develop a scheme for the nondestructive quality assurance (QA) examination of adhered joints in experimental automotive body shell assemblies. The use of adhesives on automotive structures has potential for providing flexibility of design, including the use of mixed materials, reductions in vehicle weight, and improved impact and vibration performance; manufacturing cost may also be reduced. A requirement for adhered joints is that the adhesive layer dimensions are within specified ranges and this leads to a requirement to determine bond dimensions non destructively for the purposes of QA. However, in many automotive assemblies direct access to the joint region is not possible due to structure geometry and cladding. Ultrasonic Lamb waves can provide remote access to otherwise inaccessible joints, the idea being to excite such waves in the metal adherend on one side of the joint, and to receive them from a second adherend on the other side of the joint; comparison of Lamb wave signals that have traversed the joint with signals that have propagated along plane sheet could lead to a means to determine joint dimensions. The physics of Lamb wave interactions with typical automotive joints is complex and generally not amenable to inverse solutions that would yield joint dimensions directly. In this work we have applied artificial neural networks (ANNs) to relate patterns in Lamb wave signals to joint dimensions. After training, tests of the networks on adhered samples not used in training showed that the networks could recognise key bond dimensions in more than 90% of trials. Simplification of the networks (minimisation) gave improved interpolation performance in the recognition of bond dimensions not included in network training. The weights associated with trained networks could be used to identify salient features in received Lamb wave signals in a manner that gave an indication of the physics underlying Lamb wave propagation across bonded assemblies. This aspect will be discussed in a companion paper in this volume [1].