# A Multivariate Linear Regression-Based Ultrasonic Non-Destructive Evaluating Method for Characterizing Weld Tensile Strength Properties

## TL;DR

This paper introduces a non-destructive method using ultrasonic signals and regression to estimate weld tensile strength, achieving 76.3% accuracy.

## Contribution

A novel ultrasonic-based non-destructive evaluation method using multivariate regression for weld tensile strength characterization.

## Key findings

- An ultrasonic-based model was developed to predict weld tensile strength with 76.3% accuracy.
- Ultrasonic signals from 240 weld points were analyzed alongside destructive testing data.
- A grading evaluation model was introduced to rapidly characterize welds.

## Abstract

Destructive testing is a common method for obtaining tensile strength properties of welds. However, it is inconvenient to characterize the overall weld, and it cannot be applied to in-service structures. Non-destructive testing and evaluation (NDT&E) methods have the potential ability of overcoming these limitations. In this paper, an ultrasonic-based non-destructive evaluating method for weld tensile strength was proposed. Multiple sets of fully automatic welded X80 steel pipes were prepared. Acoustic signals from a total of 240 measurement points of the welds were collected, and ultrasonic characteristic parameters were subtracted through signal processing. Subsequently, tensile strength values were obtained through destructive testing. Using the ultrasonic and tensile test databases, a multivariate regression-based (MLR) non-destructive evaluation model was established to predict the tensile strength value. Based on this, in order to rapidly characterize the welds, a grading evaluation model was introduced. The grading evaluation result of the 240 measurement points indicates that the accuracy of the proposed method is 76.3%. In order to improve accuracy, a deep learning-based method could be used in the future.