Abstract
Metallurgical defects in metal laser additive manufacturing (LAM) are inevitable due to complex non-equilibrium thermodynamics. A laser ultrasonic system was designed for detecting surface/near-surface defects in the layer-by-layer LAM process. An approach was proposed for ultrasonic imaging of defects based on variable time window intensity mapping with adaptive 2σ threshold denoising. The Gaussian mixture model hypothesis and expectation-maximization algorithm can automatically differentiate between components dominated by defects and background noises, thereby providing an adaptive threshold that accommodates detection environments and surface roughness levels. Results show that the ultrasonic wave reflection at defect boundaries diminishes far-field ultrasonic intensity upon pulsed laser irradiation on surface defects, enabling defect size and location characterization. This method is applicable to LAM samples with a significant surface roughness of up to 37.5 μm. It can detect superficial and near-surface defects down to 0.5 mm in diameter and depth, making it significant for online defect detection in additive manufacturing.