Wire bonding is an important technique for forming semiconductor junctions, and statistical quality assurance is determined through sampling inspection. However, as the number of connections increases due to high semiconductor integration, wire bonding reliability becomes important, and connection evaluation is required for all products. In a previous study, we focused on the application of ultrasonic waves, which greatly influences the bonding strength of wire bonding, and proposed a quality estimation method using a thin film AE sensor and machine learning. In that study, samples made with manual wire bonders were destroyed by pull testing, and the quality was judged based on the fracture load. However, the loop shape of the manually prepared samples was not constant, and thus the results of the pull testing varied. In this study, we automated the fabrication of the bonding samples, stabilized the sample shape, and sought improvement in the quality evaluation performance. Since the number of defective samples was small, we developed a quality estimation method using a one-class SVM, an anomaly detection method involving machine learning. Experiments using actual samples confirmed that the accuracy rate in the proposed method was roughly 86%.