窄间隙P-GMAW多信息融合侧壁熔合状态识别研究

窄间隙焊缝坡口间距小且焊道较深，摆动中心与焊缝中心偏差较大时，坡口两侧侧壁受热不良，易发生未熔合缺陷。为了及时了解窄间隙侧壁熔合情况，掌握侧壁内部焊接质量，本文提出了一种基于BP神经网络和D-S证据理论的多信息融合方法，预测侧壁熔合状态。对窄间隙焊接未熔合缺陷产生机制进行了分析，研究发现焊接电弧信号和熔池变化与侧壁成形质量存在密切关系，为此进行了一系列偏差实验，建立了电弧电信号和电弧熔池图像信号的实时采集系统，采用批量特征提取算法，提取了与侧壁熔合状态密切关联的峰值电流、峰值电压、电弧弧长、熔池长宽比、熔池面积和熔池周长等特征参量。采用BP算法训练神经网络，在此基础上通过D-S证据理论进行决策级融合。实验结果表明，该模型识别率可达96.667%,避免了神经网络识别时的误诊，获得了比单一传感信息更好的预测结果，提高了熔合状态识别的准确度和可靠度。

Sidewall fusion state recognition of narrow gap P-GMAW based on multi-information fusion

In narrow gap welds, the groove spacing is small and the weld bead is deep. When the swing center deviates significantly from the weld center, the sidewalls on both sides of the groove are underheated, and non-fusion defects are prone to occur. In order to know the fusion status of the narrow gap sidewall in time and grasp the welding quality inside the sidewall, a multi-information fusion method based on BP neural network and D-S evidence theory was proposed to predict the sidewall fusion states. The mechanism of the non-fusion defect in narrow gap welding was analyzed. The welding arc signal and molten pool change were found to be closely related to the sidewall forming quality. Therefore, a series of deviation tests were carried out, and a real-time acquisition system of arc electrical signal and arc molten pool image signal was established. The peak current, peak voltage, arc length, ratio of length to width of the molten pool, and the area and circumference of the molten pool were extracted as characteristic parameters that were closely related to the sidewall fusion states. BP algorithm was used to train the neural network, and decision level fusion was carried out through the D-S evidence theory. Results show that the recognition rate of the model reached 96.667%, which can avoid the misdiagnosis in neural network recognition, obtain better prediction results than single sensor information, and improve the accuracy and reliability of fusion state recognition.