Seiji Hashimoto, Ph.D
	Professor, Gunma University (Japan)
A Deep Learning-Based Trustworthy Defect Diagnosis Using Multi-Sensor Information for Safety-Critical Manufacturing Infrastructure
This keynote presents a trustworthy deep learning–based defect diagnosis framework for safety-critical manufacturing infrastructure. In automotive press production, early detection of rare defects is crucial, yet challenging when anomalies show no clear visual cues and require interpretability and human oversight. We develop an autoencoder trained solely on normal data to detect subtle, concurrent anomalies across multiple sensors. Grad-CAM–based visual explanations enhance transparency and operational governance, providing interpretable insights into model decisions. The framework strengthens the foundations of trustworthy AI by offering a validated and interpretable defect diagnosis approach tailored for emerging manufacturing environments.