Enhancing Power Grid Inspections with Machine Learning

TL;DR

This work demonstrates the effectiveness of 3D semantic segmentation for automating power grid inspections using LiDAR point clouds from the TS40K dataset. It benchmarks transformer-based methods, analyzes the impact of class weighting, noise handling, ground removal, and color/normal features, and introduces an inspection tool that flags uncertain predictions for manual review to maintain reliability. Key findings include high IoU for power lines (e.g., up to $IoU=0.9553$ under noise-aware training) and substantial gains from removing ground points, with TS-RGB providing limited color-based improvements. The proposed framework promises substantial gains in inspection efficiency and safety, while acknowledging remaining challenges like extreme class imbalance and noise in real-world data, and outlining avenues for multimodal and real-time extensions.

Abstract

Ensuring the safety and reliability of power grids is critical as global energy demands continue to rise. Traditional inspection methods, such as manual observations or helicopter surveys, are resource-intensive and lack scalability. This paper explores the use of 3D computer vision to automate power grid inspections, utilizing the TS40K dataset -- a high-density, annotated collection of 3D LiDAR point clouds. By concentrating on 3D semantic segmentation, our approach addresses challenges like class imbalance and noisy data to enhance the detection of critical grid components such as power lines and towers. The benchmark results indicate significant performance improvements, with IoU scores reaching 95.53% for the detection of power lines using transformer-based models. Our findings illustrate the potential for integrating ML into grid maintenance workflows, increasing efficiency and enabling proactive risk management strategies.

Figures (4)

• Figure 1: The TS40K dataset is derived from raw 3D scans illustrated in Figure \ref{['fig:teaser-raw-sample']} and processed into three different sample types: (1) Tower-radius focuses on the towers that support power lines and its environment (Fig. \ref{['fig:teaser-tower-radius']}). (2) Power-line samples have power lines as their main focus in the 3D scenes (Fig. \ref{['fig:teaser-power-line']}). (3) No-tower samples represent rural terrain where the transmission system is located, excluding supporting towers but potentially including power lines (Fig. \ref{['fig:teaser-no-ts']}). In Figures \ref{['fig:density1']} and \ref{['fig:density2']}, we showcase the semantic class densities of the TS40K dataset. Figure \ref{['fig:density1']} illustrates the class density for each of the sample types and Figure \ref{['fig:density2']} shows the overall class density in the TS40K train and test sets.
• Figure 2: Qualitative results showcasing the performance of Point Transformer V3 (PTV3) wu2023ptv3 on the TS40K dataset. While PTV3 is not the highest mIoU performing model, it consistently achieves the highest segmentation performance in crucial inspection elements, namely supporting towers and power lines. Thus, it is particularly well-suited for tasks prioritizing the accurate detection of these components in power grid inspections.
• Figure 3: Visualization of the TS-RGB dataset. TS-RGB is an augmented version of the TS40K dataset, incorporating RGB channels to improve 3D semantic segmentation in power grid environments. Covering approximately 8,000 kilometers of transmission network, it includes over 1,295 million points collected using LiDAR sensors. Ground points are automatically removed by heuristics and annotations exclude differentiation between low and medium vegetation.
• Figure 4: Qualitative results illustrating the performance of Point Transformer V2 (PTV2) wu2022point on the TS-RGB dataset, utilizing point coordinates as the sole input features. Table \ref{['tab:labelec_xyz']} demonstrates that PTV2 achieves the highest mean IoU and outperforms all other models in tower segmentation across all experiments in this configuration. These results underscore PTV2's effectiveness and reliability for semantic segmentation tasks, particularly in high-stakes applications such as power grid inspections.