In-Process Joint Strength Estimation in Pulsed Laser Spot Welding Using Artifical Neural Networks

Abstract

Pulsed laser spot welding is used in the manufacture of

many goods. Because weak joints can lead to product

defects, it is important to monitor and control the joint strength

precisely. This paper introduces a method to estimate the joint

strength of spot welds during the welding process. A point

infrared sensor is used to measure temporal radiation on the

top face of the spot weld. Because variable measuring conditions

affect the radiation power, a scale-free radiation feature

is extracted from the measured radiation and used as a monitoring

criterion. An artificial neural network (ANN) uses this

feature to estimate joint strength. In experiments, significant

welding parameters are varied within a controllable range,

and 640 weld parts are used for ANN learning. The correlation

coefficient between the estimated and measured strength

is more than 0.98 for learned parts. Another 180 weld parts

are used to appraise the efficiency of the learned ANN, and

the mean square error of estimation is 0.78 kgf.