Enel leverages Amazon SageMaker for automated large-scale power grid asset management and anomaly detection
Introduction:
This guest post by Mario Namtao Shianti Larcher, Head of Computer Vision at Enel, explores how Enel is using machine learning and computer vision to automate the assessment and anomaly detection process of its large-scale power grid. Enel, a multinational company and the first private network operator in the world, inspects its electricity distribution network using millions of photographs and LiDAR point clouds. By utilizing Amazon SageMaker and other AWS services, Enel has developed a machine learning platform that enables efficient training and inference on these large volumes of data. This post delves into the details of Enel’s ML pipeline, including the use of high-resolution photographs, LiDAR point clouds, and satellite imagery for asset identification and anomaly detection.
Full Article: Enel leverages Amazon SageMaker for automated large-scale power grid asset management and anomaly detection
Enel Implements Machine Learning and Computer Vision for Power Grid Assessment Management and Anomaly Detection
Enel, the multinational energy company with a presence in 32 countries, has embraced machine learning (ML) and computer vision to automate the assessment and anomaly detection of its large-scale power grid. With more than 2.3 million kilometers of distribution network to monitor, Enel previously relied on manual analysis of millions of photographs taken during inspections, a time-consuming and error-prone process.
Enel’s ML platform, known as the ML factory, built on Amazon SageMaker, has revolutionized the way they analyze data. They collect data from three sources: aerial network inspections, high-resolution images, and satellite images. These sources provide accurate and detailed information about the power grid, including infrastructure, assets, and anomalies.
Analyzing High-Resolution Photographs
Enel stores the high-resolution photographs taken during inspections on Amazon Simple Storage Service (Amazon S3). These images are manually labeled and used to train deep learning models for various computer vision tasks. The processing and inference pipeline involve identifying regions of interest in the images, cropping them, identifying assets within the regions, and classifying them based on material or anomalies. Enel leverages PyTorch framework and the latest image classification and object detection architectures for these tasks.
To avoid duplicates and group images of the same pole, a reidentification process is implemented. Enel uses GPU instances and Amazon SageMaker Training jobs to train models efficiently and parallelly. Inference is orchestrated by Step Functions, a state machine that governs multiple SageMaker processing and training jobs.
Extracting Precise Measurements with LiDAR Point Clouds
While high-resolution photographs provide valuable information, precise measurements cannot be extracted from them as they are 2D. Enel overcomes this limitation by leveraging LiDAR point clouds, which are 3D reconstructions of the infrastructure. Enel uses KPConv, a semantic point cloud segmentation algorithm, to assign a class to each point in the cloud. This allows Enel to determine the proximity of vegetation to power lines, measure the tilt of poles, and identify other important details. GPU instances are also used for LiDAR point cloud analysis.
Mapping Vegetation Trends with Satellite Images
Enel recognizes the importance of monitoring vegetation trends to prevent service disruptions caused by tree pruning. Inspecting the power grid with helicopters is costly and infrequent, so Enel integrates satellite image analysis into their solution. They analyze satellite images to identify vegetation, its density, and the types of plants. Enel concluded that the free Sentinel 2 images provided by the Copernicus program offer the best cost-benefit ratio for this use case. Furthermore, Enel uses satellite imagery to identify buildings and detect any discrepancies between their presence and Enel’s power delivery.
In conclusion, Enel’s adoption of machine learning and computer vision has revolutionized its power grid assessment management and anomaly detection processes. Through the use of advanced technologies and AWS services like Amazon SageMaker, Enel has automated the analysis of millions of photographs, extracted precise measurements from LiDAR point clouds, and monitored vegetation trends with satellite images. This transformative approach has not only saved time and cost but also improved the accuracy and efficiency of Enel’s infrastructure management.
Summary: Enel leverages Amazon SageMaker for automated large-scale power grid asset management and anomaly detection
This guest post by Mario Namtao Shianti Larcher, Head of Computer Vision at Enel, explores how Enel has leveraged machine learning and computer vision technologies to automate the analysis of their electricity distribution network. Enel collects data from multiple sources including aerial inspections, high-resolution images, and satellite images. They use Amazon SageMaker to build and train models for tasks such as asset identification, anomaly detection, and vegetation mapping. The post discusses the architecture and steps involved in their ML pipeline and highlights the benefits of using SageMaker for training and inference. Enel’s adoption of these technologies has greatly improved efficiency and accuracy in network monitoring and assessment.
