Ultrasonic welding is a well-known process for joining thermoplastics and has recently been introduced for joining carbon fiber reinforced polymer (CFRP) composite materials in the automotive industry. As a new joining method for CFRP materials, an understanding of the impact of the welding process on weld attributes and joint performance such as lap-shear strength is needed, as are methods to effectively classify weld quality. This paper investigates the relationship between joint performance and weld energy in ultrasonic welding of injection molded thin short-fiber CFRP sheets. Weld quality classes for training a generalized algorithm are determined from welded joint lap-shear strength and the microstructure of the weld zone. A simple and efficient method for feature selection is proposed to screen the most significant features for predicting from multiple weld quality classes. Several feature selection and weld quality classification methods were compared. A Bayesian Regularized Neural Network (BRNN) was found to be more accurate and robust when classifying weld quality in ultrasonic composite welding than the previously proposed methods of support vector machine (SVM), k-nearest neighbors (kNN), and linear discriminant analysis (LDA).

超声波焊接是用于接合热塑性塑料的公知方法，并且最近已被引入以用于接合汽车工业中的碳纤维增强聚合物（CFRP）复合材料。作为CFRP材料的一种新连接方法，需要了解焊接工艺对焊接属性和接头性能（例如搭接剪切强度）的影响，以及有效分类焊接质量的方法。本文研究了注射成型的短纤维CFRP薄板超声焊接中接头性能与焊接能量之间的关系。根据焊接接头的搭接剪切强度和焊接区的微观结构，确定用于训练通用算法的焊接质量等级。提出了一种简单有效的特征选择方法，以筛选最重要的特征，以便从多个焊接质量类别进行预测。比较了几种特征选择和焊接质量分类方法。发现贝叶斯正则神经网络（BRNN）在对超声复合焊接中的焊接质量进行分类时比以前提出的支持向量机（SVM），k近邻（kNN）和线性判别分析方法更准确，更可靠（ LDA）。