Wire Drawing Die Recognition Based on Visual Inspection

The wire drawing die is a key consumable in metal-wire processing, and diamonds are generally implanted as die cores by metal metallurgy for its high hardness. But the uncertainty of diamonds position greatly obstucts the efficiency of the automatic control of early stage of this. What’s more, to achieve efficient and precise production, machine vision is widely used. Since diamonds are invisible, the primary difficulty lies in the detection of the diamond exposure in initial stages of processing. However, due to the unfixed position of diamonds, precise positioning is difficult to achieve during the automation control process. To address this issue, this paper leverages machine vision to enhance the flexibility of processing and proposes an improved U-Net network model specifically designed for detecting diamond exposure in wire drawing die. Among them, considering the model’s generality, the U-Net convolutional networks down-sampling part is replaced by the feature extraction part of visual geometry group network-16 for training. Appropriate online data enhancement methods were also used. The experimental results show that the method has a missed detection rate of 6.67% and a misjudgment rate of 3.67% on the wire drawing die dataset. Compared with the original model, the missed detection and misjudgment of diamond exposure are reduced, and the training time is shortened from the original 94792 to 48526 s, which greatly shortens the training time and reduces the time cost. The improved model reduces the misjudgment and missed detection to a certain extent, and greatly shortens the training time. Meanwhile, the paper achieves adaptive feedback of machine vision through artificial intelligence technology.