Damage Detection and Localization by Learning Deep Features of Elastic Waves in Piezoelectric Ceramic Using Point Contact Method

Pragyan Banerjee, Pranjal Saxena, Nur M M Kalimullah, Amit Shelke, Anowarul Habib; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2024, pp. 63-70

The reliable detection and localization of damages in piezoelectric materials such as Lead Zirconate Titanate (PZT) pose significant challenges in various engineering applications. Conventional methods for damage detection often depend on manual inspection or basic signal processing techniques which are subjective labor-intensive and susceptible to human error. In this paper a novel approach for damage detection and localization in PZT materials using deep learning techniques is proposed. Leveraging a convolutional neural network (CNN) in tandem with methodologies such as class activation mapping (CAM) the objective is to enhance the accuracy and reliability of fault detection systems. In particular the VGG16 architecture is adopted as the foundation of the proposed framework due to its simplicity and effectiveness in large-scale image recognition tasks. By integrating CAM into the training process CNNs are equipped to precisely localize anomalies within PZT ceramic images facilitating improved fault detection performance. The study demonstrates the effectiveness of deep learning methods in addressing the challenges of fault detection and localization in PZT materials offering promising avenues for advancing monitoring and maintenance practices in various engineering applications.