Multimodal data fusion integrates various data sources to generate a single representation appropriate for complex analysis. This study describes a method for detecting and characterizing defects in metals by combining the techniques of phased-array ultrasonic testing (PAUT) with pulsed thermography (PT) using a data fusion coordinate transformation technique to combine the capabilities of the two modalities into a volumetric dataset. PAUT inspection is limited to internal defects, whereas PT inspection is limited to surface and near-surface defects. The data fusion technique combines complementing information from both modalities, allowing one to comprehend defects that would otherwise be invisible using either technique alone. To enhance the defect detection process, a multimodal automatic defect recognition (M-ADR) system that includes a Deep Neural Network (DNN) and a Bi-Planar Medial Axis Transform (Bi-MAT) algorithm was developed. The integrated DNN system with fused volumetric information achieves a remarkable flaw detection accuracy of 91.46%, outperforming conventional DNN models and single-modality inspection techniques. M-ADR allows extraction of precise defect geometries, which sizes the smallest defect of lambda/4.