M1-2.4 - CNN-based classification of scale deposits and damages in drainage pipes using a mobile inspection rover

Event

23. ITG/GMA-Fachtagung Sensoren und Messsysteme 2026
2026-06-09 - 2026-06-10
Nürnberg

Band

Vorträge

Chapter

Energie- und Umwelttechnik

Author(s)

V. Putz, C. Hofer, S. Fleischanderl, L. B. Hörmann, J. Schröck, W. Hofer - Linz Center of Mechatronics GmbH, Linz (Austria), H. Hölzl, T. Schachinger, V. M. Batka - ÖBB-Infrastruktur AG, Wien (Austria)

Pages

351 - 358

DOI

10.5162/sensoren2026/M1-2.4

ISBN

978-3-910600-11-9

Price

free

Abstract

A convolutional neural networks (CNN)-based approach for automated classification of scale deposits and structural damage in drainage pipes of railway tunnels is presented and validated using data collected by a mobile inspection rover - a waterproof, battery-powered robot equipped with cameras that record video data for offline analysis. For evaluation, two deep learning approaches are considered: image classification using VGG₁₆ and object detection using YOLOv₁₁. Both models were trained and validated using data acquired during field tests in Austrian railway tunnels. While VGG₁₆ enables simple frame-wise classification, YOLOv₁₁ provides detailed spatial information and allows parallel detection of defects and various severity levels of scaling per frame. Results showed reliable detection of severe defects, with some confusion between adjacent severity levels. To support planning of drainage pipes maintenance, a post-processing method was introduced that transformed frame-wise detections into a time-series based indicator implying the condition of pipe segments. The approach demonstrated the potential of combining autonomous inspection with machine learning in the future for condition-based and predictive maintenance of tunnel drainage systems.