基於熔池影像的焊接外觀結果預測研究

Abstract

在製造業中，焊接品質的控制對於最終產品的穩定性和可靠性至關重要。尤其在薄片材料的焊接中，由於材料特性和厚度較薄，焊接過程極易出現變形及各種缺陷，對品質的即時監控有著更高的要求。然而，傳統檢測方法多為事後檢查，難以及時反映焊接過程中的異常與缺陷，如工件表面的裂紋或熔透不足等問題。尤其是在管道類環境中，接口焊接完成後甚至無法直接檢查內部是否焊接成功。此外，使用被動視覺工業相機記錄焊接過程，由於電弧光強度高和焊接區域溫度高，影像捕捉與分析面臨挑戰。因此，如何透過實時監視焊接過程並減少缺陷成為一項需解決的課題。 本研究提出一種基於被動視覺工業相機的實時監控系統，並利用機器學習對焊接過程中捕捉到的影像進行分析，從而實現對加工成品形貌的分類和預測。本研究將拍攝下來的加工影像進行分類，其後將影像做處理後抓取影像中的特徵信息，將熔池邊緣輪廓作為特徵進行計算。接著本研究第二部份利用隨機森林模型學習這些特徵信息，以達到對加工成品形貌的預測分類效果。此技術在提高工業焊接生產線自動化程度方面具有重要意義，能夠有效減少因材料特性帶來的焊接缺陷，提高產品的一致性和可靠性。

In the manufacturing industry, the control of welding quality is crucial to the stability and reliability of the final product. Particularly in the welding of thin sheet materials, the unique material properties and reduced thickness make the process highly prone to deformation and various defects, placing higher demands on real-time quality monitoring. However, traditional inspection methods are predominantly post-process and struggle to promptly detect anomalies and defects during welding, such as surface cracks or insufficient penetration. This issue is even more pronounced in pipeline environments, where it is often impossible to directly inspect the internal welding quality after joint completion. Additionally, using passive industrial cameras to capture welding processes is challenging due to the high intensity of arc light and the elevated temperatures in the welding area. Therefore, developing methods to monitor the welding process in real-time and reduce defects has become a critical issue to address. This study proposes a real-time monitoring system based on passive vision industrial cameras, and uses machine learning to analyze the images captured during the welding process, thereby achieving classification and prediction of the morphology of processed products. This study classifies the captured processed images and then processes them to capture feature information from the images. In the second part of this study, the random forest model is used to learn these feature information, in order to achieve the prediction and classification effect of the processed product morphology. The model is then put into the processing process to test the accuracy and speed of real-time judgment, thereby improving the real-time monitoring efficiency of welding quality and reducing the cost and time of manual inspection. This technology is of great significance in improving the automation level of industrial welding production lines, which can effectively reduce welding defects caused by material characteristics and improve product consistency and reliability.