Automatic analysis of UAS-based thermal images to detect leakages in district heating systems

ABSTRACT

The mostly subterranean nature of district heating system pipelines makes pinpointing any occurring leakages a challenge. Airborne thermography offers a means for widespread monitoring, allowing thermal anomalies to be identified within multitudes of infrared images. This paper details a program developed to automate the entire image analysis process using mainly open-source software. Thermal images are acquired via unmanned aircraft system (UAS), pre-processed, and georeferenced individually or combined to orthomosaics. The search space is minimized to areas around the pipelines. Regions of interest are determined by image segmentation via tailored triangle histogram thresholding. The majority of resulting false alarms are removed by comparison with characteristic traits and results classified by their severity. The algorithm is applied to images newly acquired in Germany as part of a case study. The implemented methodology allows for a reduction of between 92 and 99% of thermal anomalies to a manageable amount of potential leakages for network operators to view. The use of orthomosaiking software in this context, though helpful in coalescing data, is found to lack robustness, precision and therefore reliability. Despite some limitations, the developed program is able to confidently detect and categorize leakages of varying severity and can be used directly by network operators. Future research will focus on further data pre-processing to eliminate thermal drift and remove the remaining false alarms, which mostly pertain to common urban features.

KEYWORDS:

• District heating systems
• image processing
• leakage detection
• remote thermography
• software development
• unmanned aircraft systems

Acknowledgements

The thermal images for the case study in Germany were acquired in collaboration with the Air Bavarian GmbH and Munich’s municipal utilities company Stadtwerke München.

Disclosure statement

No potential conflict of interest was reported by the author(s).

CRediT author statement

Elena Vollmer: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Writing – Original Draft, Writing – Review & Editing, Visualization.

Rebekka Volk: Conceptualization, Data Curation, Writing – Review & Editing, Supervision.

Frank Schultmann: Writing – Review & Editing, Supervision. All authors have agreed to the published version of the manuscript.

Notes

1. Orthomosaics are georectified mosaics, meaning mosaics that are correctly mapped and have had their inherent distortions removed (ArcGIS Pro 2021a).

2. Preliminary testing showed saliency maps and LoG blob detectors to be ineffective or require the setting of multitudes of dataset-dependent parameters.

3. Heipke and T¨odter (2020) define 2 m/s as the maximum wind speed yet exceeded that limit frequently during their own image acquisition.

4. Heipke and T¨odter (2020) name the following as unproblematic: fibre cement, plastic, protective, steel, and flexible pipes, independent of overhead or underground placement.

5. For instance faulty building insulation, streetlamps, people, vehicles, and metal surfaces.

6. Metal objects, for instance, are depicted with much colder temperatures than are actually measured on their surface (Gade and Moeslund 2014).

7. Latitude and longitude are translated from the geographical coordinate system (unit: degrees, minutes and seconds) into easting and northing values in a Cartesian CRS (unit: metres).

8. Ordinarily, complex angle transformations would be necessary to convert the recorded navigational attitude yaw, pitch, and roll to the photogrammetric equivalents’ omega, phi, and kappa used in georeferencing (Bäumker and Heimes 2002). These can be omitted here because the implemented 3-axis gimbal (SZ DJI Technology Co. Ltd. 2018b) stabilizes the UA, ensuring negligibly small roll and pitch angles (Prof. Grant Petty, pers. comm., April 14, 2021). Owing to nadir UA alignment, the yaw angle is directly assumed to be $\kappa$.

9. An 8° difference is experimentally determined as the threshold for the given UA Matrice 600 Pro (SZ DJI Technology Co. Ltd. 2018a).

10. In the case of a UA’s flight height, a conservative threshold is chosen at 10 pixels.

11. This form is based upon Sledz, Unger, and Heipke (2020)’s observation that an elliptical shape best depicts leakages. Berg, Ahlberg, and Felsberg (2016) consider form as an important factor, but only take circularity into account as a feature for distinguishing leakages. However, as Friman et al. (2014) state, a leakage is not always perfectly round and may also be partially covered, making an ellipse the more obvious choice.

12. Berg, Ahlberg, and Felsberg (2016) propose a similar method of comparing the intensity of a hotspot and its surroundings. However, using temperature data has the advantage of enabling postliminary leakage categorization with empirical values.

13. Sledz, Unger, and Heipke (2020) define a delta of 5°C already sufficient to indicate a leakage.

14. Munich’s municipal utilities company SWM states some manholes may indicate DHS leakages and should therefore not be disregarded completed, although the majority are marked as false alarms in Table 3.

15. In mosaics approximately 85% and in individual images 96% of thermal hotspots.

16. 7% within mosaics and 10% within single images.

17. In the Universal Transverse Mercator (UTM) projection, zone ‘32N’ (denoting European territory), and European Terrestrial Reference System 1989 (ETRS89) map datum.

Additional information

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.