The purpose of this study is to investigate the specific method of laser ultrasonic nondestructive testing technology applied to the detection of arc surface defects commonly found in industrial production. To verify the influence of different dimensional parameters of defects on the reflected and transmitted surface waves, a finite element model for the detection of arc-shaped surface defects was established based on the ultrasound excitation mechanism by thermoelastic effect. Meanwhile, the empirical mode decomposition method was used to process the signal data to extract the reflected and transmitted waves of defects and superimpose the signal components of corresponding characteristic frequencies. Results reveal the changing laws of the relevant ultrasonic characteristic parameters. Based on this, two prediction models for the depth of arc surface defects are established. By comparison and analysis, the maximum error between the defect depth calculated by using the transmission wave parameter model, and the actual defect depth is only 0.6%. Therefore, the proposed defect depth prediction model provides a feasible solution for the on-site detection of arc surface defects with high accuracy.