GTAW时变过程多源异构输入与焊接质量预测模型

Based on application requirement of intelligent welding technology in advanced manufacturing, this project will research on GTAW time-varying process under effect of multi-source heterogeneous input coupling, and propose a GTAW input and welding quality output model. In order to form modeled output evaluation mechanism, the project explores synchronous extraction and identification method of full parameter dynamic information during time-varying process and builds welding quality comprehensive and multi-attribute accurate description. Analysis matching relation between ‘acoustic, electric, arc, temperature’ phenomenon and weld pool periodical variation finite state machine during welding time-varying process. Data mining by MMP algorithm, and observe time-varying process full parameter features which is marked by welding quality. Extract instant feature and timing feature of multi-source heterogeneous input based on LSTM deep learning method, and propose a hierarchical gray box depth modeling method by using the extracted feature to estimate time varying process. Execute tensor expression of time-varying full parameter feature and build welding quality output estimate model based on time-varying process by utilize DNN/SVM. Take time-varying process as medium, and build integrated responsive channel among multi-source heterogeneous input，time-varying process and welding quality output by M quaternion relation frame. Thus, this project develop the desired GTAW time-varying based multi-source heterogeneous input and welding quality out prediction model. The proposed model will reveal the affinity mechanism of quantitative relation between multi-source heterogeneous input and welding output quality. Besides, it will employed to guide the intelligent research of welding robot technology.

针对智能焊接技术在先进制造业的应用需求，研究多源异构输入耦合作用的GTAW时变过程，建立多源异构输入与焊接质量预测模型。探究时变过程全参量动态信息同步提取辨识方法，研究焊接质量全方位多属性精准描述，形成焊接质量模型化输出评价机制。分析时变过程“声电光温”现象与熔池金属周期性变化有限状态机匹配关系，利用MMP进行数据挖掘，得到指向焊接质量的时变过程全参量特征。采用LSTM深度学习算法提取多源异构输入瞬时特征和时序特征，提出分层灰箱深度建模方法，进行基于多源异构输入的时变过程预测；对时变过程全参量特征进行张量表达并通过DNN/SVM建立基于时变过程全参量特征的质量预测模型。以时变过程全参量特征为媒介，采用M四元组关系模型框架搭建多源异构输入、时变过程和质量输出的完整响应通道，提出GTAW时变过程多源异构输入与焊接质量预测模型，揭示两者间量化关系和调控机制，推进机器人焊接技术的智能化研究。

通过处理有效传感信息预测熔池熔透状态，对焊接过程中的单维信息和图像信息（主动视觉、被动视觉）进行传感和解析。采用表面堆焊建立时变过程多源信息融合输入与焊接质量的预测模型，采用对接焊接建立薄板铝合金对接焊缝质量成形预测模型，并分别搭载Python平台开发基于深度学习的熔池熔透动态评价算法。搭建完成表面堆焊多参量动态传感与预测一体式GTAW熔透研究实验平台和薄板铝合金对接焊接焊缝成形监测实验研究平台。表面堆焊研究方面，基于MISO神经网络建立表面堆焊多源信息融合输入与焊接质量预测模型，分析焊接传感所得单维信号和图像信号，并探索传感信息与熔池上表面之间的关联，探索熔池动态演变过程，揭示熔池生长行为并建立传感信息与焊缝熔透之间的映射关系。采用单维信号和图像信号同步传感并建立数据集，采用MLP、CNN和MISO分别建立单维信号、图像信号以及多信息融合模型并开展训练、验证和预测；分析了基于时间、电流、电压和热输入等单维信号、主动视觉和被动视觉的二维图像信号与基于多源信息融合的焊缝质量预测模型对焊接熔池熔透的预测效果；并进一步对比了基于主动视觉和基于被动视觉的融合信号焊接质量模型预测精度。熔池多源信息传感能够较为全面地揭示时变过程熔池的周期性生长和演变行为，所建基于多传感手段的多源信息融合输入与熔池熔透预测模型提升了焊缝成形预测精度。在薄板铝合金对接焊接研究方面，基于动态视觉传感建立薄板铝合金对接焊缝质量成形预测模型。采用U2-net网络分析焊接传感所得动态图像信号变化规律和演化过程，并模拟多种焊接扰动探索焊接熔池与熔透的映射关联。解析焊缝成形缺陷状态，采用AlexNet模型获得焊缝宏观成形预测和焊缝熔透预测，采用模型串联构架建立传感信息与焊缝熔透之间的映射关系，获得高精度焊缝成形预测。本项目为焊接制造过程中高效信息传感，熔透实时预测，智能化焊缝质量控制这一系列关键问题奠定了理论和应用基础。