• Chinese Journal of Lasers
  • Vol. 50, Issue 8, 0802102 (2023)
Tianyi Li1, Tuo Shi1、*, Kuan Li2, Rongwei Zhang2, Jianbin Li2, Yewang Sun1, and Guang Liu2
Author Affiliations
  • 1School of Optoelectronic Science and Engineering, Soochow University, Suzhou 215006, Jiangsu, China
  • 2School of Mechanical and Electrical Engineering, Soochow University, Suzhou 215021, Jiangsu, China
  • show less
    DOI: 10.3788/CJL220746 Cite this Article
    Tianyi Li, Tuo Shi, Kuan Li, Rongwei Zhang, Jianbin Li, Yewang Sun, Guang Liu. Height and Width Model of Cladding Layer Formed by Laser Cladding with Variable Angle[J]. Chinese Journal of Lasers, 2023, 50(8): 0802102 Copy Citation Text show less



    The fabrication of an overhang or large inclination structure with laser cladding must be completed on an inclined substrate. Regarding the parameters and morphology of laser cladding layer, previous studies have mainly been conducted on a horizontal base plane. Few studies have focused on the influence of different inclination angles of the base plane on forming morphology. The molten pool is often stretched or even displaced by gravity when conducting multilayer deposition with a large inclination, which affects the height and width of the single pass after solidification. A slight change in the height and width can affect the final forming accuracy; whereas, large changes in the height and width, particularly when the actual layer height is inconsistent with the preset layer height, will directly affect the forming quality and continuity. Therefore, this study explores the influence of different base plane inclinations on the height and width of a single track, and uses the base plane inclination as one of the inputs to establish a neural network prediction model.


    First, a single-factor experiment method was used to scan a single layer to determine the working range of each process parameter and the change step of the parameters. The laser power was varied from 800 to 1200 W in steps of 100 W. The scanning speed was varied from 4 to 8 mm/s in steps of 1 mm/s. The angle was varied from 0° to 135° at the step rate of 15°. Thin-wall deposition experiments were then carried out at 10 selected angles, and five groups of deposition with different power parameters at each angle were considered. Each group of thin wall was deposited with 30 layers, and the process included five groups of selected scanning speeds, which was changed every six layers. A CCD layer height measurement system was used to collect layer height data in real time during the deposition process of the thin wall.

    Results and Discussions

    The thin wall was cut from the middle, and the width of each layer of the cut section was measured. The mean values of the last three layers of every six layers in the measured layer height data are valid (Fig. 6). Finally, 250 sets of height and width data were obtained. Based on this data (Figs. 7 and 8), a BP neural network prediction model was established. The model considers the inclination of the cladding base plane, scanning speed, and power as the input, and the height and width of the cladding layer as the output. The data containing various angles, power, and speed information were regarded as the training set to enhance the comprehensiveness of the test set. The model was built using only the training set, and the remaining data were used as the test set. The test set was only used to test the predictive ability of the model and evaluate its generalization ability.


    The influence of variable angle cladding of 0°-135° on the single-pass morphology was studied. The experimental results show that the layer height first decreases and then increases with the change in the inclination angle, and that at 90° yields the lowest layer height, which can be attributed to the constant change in the angle between the gravity direction and the growth direction. The layer width first increases and then decreases with the angle change, reaching the highest value at 90°. The root mean square error of the two established neural network prediction models is controlled below 0.1, and the 90% confidence prediction accuracy A90% is 99% and 96%, respectively (Fig. 11), showing an excellent prediction effect of the established model.