Search

Along with the strong development of the industrial revolution 4.0 in many fields, the artificial intelligence systems supporting learners have also received more attention. In this paper, we propose a welding practice training support system using artificial intelligence. The system is specially designed to be able to learn from the challenging dataset, that is small but contains many variations of weld, to classify student’s weld based on completeness of weld. To handle the multi-variant of weld defects, we propose cSTN to not only allow the network to adaptively capture the weld defect regions but also make the decision based on both local and global perspectives. Besides, to deal with the insufficient data problem, we propose to slightly modify the DenseNet architecture as our backbone network. Experimental results indicated that the proposed system still achieved good performance despite only training on a small dataset.

Applying the Laser Cladding method to create composites cladding Co+(10; 20; 30) % TiC on the surface of Hot Die, SKD61 steel. Using Micro-hardness tester and Microscope (SEM) to measure the hardness and analyze the distribution of TiC particles in the clad. The results indicated that the average hardness of the TiC/Co-based composites cladding cross-section Co-based+10% TiC, Co-based+20% TiC, Co-based+30% TiC in order is 589 HV0.2, 788 HV0.2, and 934 HV0.2. All of these claddings have the hardness higher than SKD61 steel in order is 2.6, 3.5, and 4.2 times. From that show that, the hardness of the claddings has a close relationship with percentage of TiC particles. Besides, the regression models exactly show that the hardness of the claddings is: y1 =23.8438-7.3494x, y2=9.7389x+44.6927, y3=13.6597x+35.511 by 10, 20, 30% wt TiC particles in the claddings, which create the scientific theory to apply from some other carbides to reinforced for the cladding.