Deep learning-based penetration depth prediction in Al/Cu laser welding using spectrometer signal and CCD image

Abstract

In the laser welding of thin Al/Cu sheets, proper penetration depth and wide interface bead width ensure stable joint strength and low electrical conductance. In this study, we proposed deep learning models to predict the penetration depth. The inputs for the prediction models were 500 Hz-sampled low-cost charge-coupled device (CCD) camera images and 100 Hz-sampled spectral signals. The output was the penetration depth estimated from the keyhole depth measured coaxially using optical coherence tomography. A unisensor model using a CCD image and a multisensor model using a CCD image and the spectrometer signal were proposed in this study. The input and output of the data points were resampled at 100 and 500 Hz, respectively. The 500 Hz models showed better performance than the 100 Hz models, and the multisensor models more accurately predicted the penetration depth than the unisensor models. The most accurate model had a coefficient of determination (R2) of 0.999985 and mean absolute error of 0.02035 mm in the model test. It was demonstrated that low-cost sensors can successfully predict the penetration depth during Al/Cu laser welding.