A novel non-destructive testing (NDT) system for detecting structural defects in Printed Circuit Boards (PCBs) using real-time B-scan imaging by air-coupled ultrasonic transducers and integrating deep learning is proposed in this article. Traditional inspection methods like visual inspection and X-ray inspection are likely to miss hidden defects, especially in multilayer PCBs. To bridge these gaps, we introduce a contactless inspection method with high-frequency air-coupled ultrasonic transducers for B-scan imaging and a convolutional neural network (CNN) for the classification of defects. Air-coupled transducers facilitate non-contact testing, minimizing handling risks and placing the system in a position for inline production testing. Noise reduction and contrast enhancement are preprocessing techniques used to enhance image quality. The CNN model is pretrained from B-scan images for detection of common PCB defects such as solder voids, delamination, trace breaks, and short circuits. The system has a classification accuracy of 96% in a specific dataset. The improved model is executed on an embedded edge AI device for the intent of real-time low-latency fault detection. Comparisons against benchmark performance indicate that this technique is faster, more accurate, and more suitable to use in automated manufacturing than current techniques of inspection. The applicability of the integration of air-coupled ultrasonic B-scan imaging and AI in providing sophisticated, noncontact PCB inspection is proven by this research.