Paradigm Shift in Infrastructure Inspection Technology: Leveraging High-performance Imaging and Advanced AI Analytics to Inspect Road Infrastructure

TL;DR

ROVAI presents a paradigm-shifting end-to-end framework that fuses high-throughput X-ray computed tomography with advanced AI analytics to inspect road infrastructure at unprecedented scale. By leveraging the combined power of SPring-8 XCT imaging and supercomputers Fugaku and Frontier, it achieves rapid, memory-efficient 3D reconstruction and 3D segmentation using a foundation Vision Transformer trained with self-supervised MAE data and simulated labels, enhanced by Symmetrical Adaptive Patching. Key contributions include a high-throughput 3D XCT pipeline with group-row partitioning, a 3D segmentation foundation model with in-situ MAE pretraining, and a memory-resident fused XCT+AI pipeline that minimizes storage I/O via bitmap masks and in-memory inference. The results demonstrate near-peak scalability on full-system hardware, 60 GB/s I/O throughput, 95% segmentation accuracy on 8K volumes, and practical impact for automated, data-driven road maintenance and broader infrastructure health monitoring.

Abstract

Effective road infrastructure management is crucial for modern society. Traditional manual inspection techniques remain constrained by cost, efficiency, and scalability, while camera and laser imaging methods fail to capture subsurface defects critical for long-term structural integrity. This paper introduces ROVAI, an end-to-end framework that integrates high-resolution X-ray computed tomography imaging and advanced AI-driven analytics, aiming to transform road infrastructure inspection technologies. By leveraging the computational power of world-leading supercomputers, Fugaku and Frontier, and SoTA synchrotron facility (Spring-8), ROVAI enables scalable and high-throughput processing of massive 3D tomographic datasets. Our approach overcomes key challenges, such as the high memory requirements of vision models, the lack of labeled training data, and storage I/O bottlenecks. This seamless integration of imaging and AI analytics facilitates automated defect detection, material composition analysis, and lifespan prediction. Experimental results demonstrate the effectiveness of ROVAI in real-world scenarios, setting a new standard for intelligent, data-driven infrastructure management.

Figures (15)

• Figure 1: This work transforms infrastructure inspection towards a technology based on high-performance imaging and AI analytics. We use three flagship facilities in a coordinated manner: a) RIKEN SPring-8 synchrotron (experimental facility): providing high-quality XCT scans of infrastructure specimens in a fast turnaround, while managing high-volume data transfer to other facilities, b) RIKEN Fugaku supercomputer (computational facility): a large-scale system with 158,976 compute nodes, offering a balanced design for data movement (network and storage) and capable of processing dozens of high-resolution volume reconstructions fused with AI analytics, and c) ORNL Frontier supercomputer (computational facility): a powerful system equipped with 37,888 GPUs, enabling more effective large-scale and efficient pre-training of foundational vision models.
• Figure 1: Parameters used in ROVAI. The upper section lists acquisition-specific parameters, the lower section defines parallelization settings.
• Figure 2: Overview of ROVAI: an end-to-end high-performance imaging and AI-driven analysis framework for road infrastructure inspection.
• Figure 3: Illustration of how a single specimen image is reconstructed using 3-dimensional partitioning and parallelism.
• Figure 4: Normal scan versus offset scan XCT. Offset scan is favorable for acquisition efficiency. Offset scan yields a more imbalanced workload in comparison to normal scan.