Abstract
<title>Abstract</title> <p>Heat accumulation during the Laser Directed Energy Deposition (LDED) process is a key factor affecting the microstructure of the deposited layer. In addition to process parameters, the scanning strategy is an important factor influencing the thermal field. To improve the effects of heat accumulation in the Continuous Laser Deposition (CLD) process, this study proposes a scanning pattern optimization method that integrates the Quasi-Continuous Wave (QCW) process with a deep regression algorithm, referred to as the Interval Pulsed Laser Deposition (IPLD) process. First, a thermal field dataset is constructed based on the finite element simulation results of random scanning patterns. Second, a deep regression model based on a Skip-Connected 3D Convolutional Autoencoder (SC-3DCAE) is developed, and the model is trained using this dataset to achieve thermal field prediction. Finally, the optimal scanning pattern is fast screened out according to the requirement of the high cooling rate. Finally, the recommended optimized scanning pattern is used for finite element simulation and experiment verification of 20-layer thin-walled deposition, where DD5 material is deposited on the Inconel 718 substrate. The results show that the cooling rate of the IPLD process in all regions is significantly higher than that of the CLD process, resulting in fine columnar grains with a stronger < 001 > orientation. The IPLD process exhibits smaller dendrite spacing and less developed secondary dendrite arms. The proportion of large-sized carbides in the IPLD sample is significantly lower than that in the CLD sample. The elongation of the IPLD sample reaches 34.7%, which is better than the 29.9% of the CLD process, and the fracture dimples are smaller.</p>