铝合金变极性等离子弧焊穿孔熔池多源信息智能表征及焊缝成形控制策略研究

In the process of aluminum alloy Variable Polarity Plasma Arc Welding (VPPAW), the keyhole behavior, the intelligent characterization of weld pool features and the controlling of stable weld appearance have long been challenges. This project aims to develop a new methodology of online monitoring and real-time controlling the perforation behavior in aluminum alloy VPPAW. The methodology will be based on the intelligent information processing of plasma arc sound signal and visual feature signal of the keyhole welding pool. The project consists of several parts: developing the voice recognition method, which is based on linear prediction model of variable polarity plasma arc channel, to tell the penetration status of perforated welding pool; designing the regression forecast model, which is based on the arc voltage,sound sensing and visual sensing of support vector machine and the keyhole area ; designing the generalized Choquest model, which is based on information fusion of sound ，visual image and arc voltage; giving the intelligent penetration status classification strategy of aluminum alloy VPPAW. The project will provide valuable scientific methodology and practical approaches to solve the problem of aluminum alloy VPPAW online monitoring and robust controlling of the weld appearance.

铝合金变极性等离子弧焊接小孔行为、熔池特征信息智能表征以及稳定成形控制一直是该领域极具挑战性的难题之一。本项目尝试发展一种基于等离子弧的电压信号、声音信号以及穿孔熔池的视觉特征信号等多源信息智能处理、铝合金变极性等离子弧焊穿孔行为动态在线监测及实时稳定控制的新途径。主要研究内容包括：基于变极性等离子弧声道线性预测模型的穿孔熔池熔透状态语音识别方法；基于部件模型（Part-based model）穿孔熔池图像特征实时计算方法；设计基于支持向量机的电弧电压、声音传感、视觉传感同穿孔熔池背面小孔面积的回归预测模型；结合等离子电弧弧声音、电弧电压及穿孔熔池视觉图像信息融合的广义Choquest模型，进一步给出铝合金变极性等离子弧焊熔透状态的智能分类策略。为解决铝合金变极性等离子弧焊接在线监测及实时稳定控制提供有价值的科学方法和技术实现途径。

铝合金变极性等离子弧焊接小孔行为、多源信息智能表征以及稳定成形控制一直是该领域极具挑战性的难题之一。本项目以航天2219铝合金变极性等离子弧焊在线监测及成形控制为主线，以变极性等离子弧焊过程“电弧-声音-穿孔视觉行为”为切入点，开展2219铝合金变极性等离子焊接动态过程在线监测及稳定成形控制的研究。主要研究内容包括：研制了集电弧声音、焊接电流、穿孔熔池视觉等一体的多源信息智能传感系统；开发了基于部件树的背面匙孔图像处理算法并提取了匙孔面积、倾角等特征信息；基于熔池“力-热”模型，构建了背面匙孔特征尺寸-焊接参数（电流）的熔透状态预测知识模型；提出了变极性等离子弧焊的“双声源特性”的隐马尔科夫（HMM）焊接熔透识别模型，并进一步融合穿孔熔池视觉特征信息，通过流线型学习算法进行降维，给出了2219铝合金变极性等离子弧焊不同熔透状态的深度学习网络分类策略。该项目的研究成果，为解决铝合金变极性等离子弧焊接在线监测及实时稳定控制问题提供了科学方法和技术途径。