Revolutionizing Utility Industry: Duke Energy Harnesses Machine Learning for Automated Inspection of Wooden Poles

As the world undergoes a rapid digital transformation, most industries are integrating Machine Learning (ML) into their processes, aiming to automate time-consuming monotonous tasks and foster efficiency. The utility industry has not been left behind. Today, this industry pioneers an exploration of automation powered by ML to carry out routine inspections, primarily focusing on wooden utility poles.

Stepping into the spotlight to illustrate this advanced intervention is Duke Energy, a Fortune 150 company. In collaboration with the Amazon Web Services (AWS) Machine Learning Solutions Lab (MLSL), Duke Energy has embarked on an innovative journey to harness ML for automated utility pole inspection.

Duke Energy, one of the largest utility companies in the U.S., serves approximately 7.7 million customers across six states. It maintains an electrical grid that consists of a comprehensive network of electrical poles. Periodic inspection of these wooden utility poles is critical to prevent catastrophic failures that can result in fires, blackouts, and severe economic repercussions. However, this inspection process is often hampered by numerous challenges such as remote access, weather conditions, and the sheer number of poles.

The manual visual inspection method that has been predominant till now involves significant labor and time input. Assessors must physically examine each pole for potential faults or signs of degradation. This technique incurs hefty costs and bears a substantial risk element as it involves working from high altitudes and possible encounters with unstable poles or live wires.

Duke Energy, known for its foray into artificial intelligence in pursuit of operational efficiencies, steps in to navigate these challenges with AI. The company’s historical involvement with AI includes a diverse range of applications like forecasting, preventive maintenance, and grid optimization, among others.

The partnership with MLSL seeks to leverage ML to automate the process of wood pole-related issue identification from high-resolution aerial images. This implementation aims to enhance grid resiliency, ensure regulatory compliance, significantly reduce inspection costs, and curtail greenhouse gas emissions resulting from traditional inspection methods. Most importantly, this will enhance safety by minimizing human involvement in precarious conditions.

Designing robust models for anomaly detection presented certain challenges. The data features vary in size, shape, texture, and color, not forgetting the diverse structural conditions of the poles. Effectively harnessing the power of ML meant initiating numerous data preprocessing techniques, including image enhancement, noise reduction, and normalization, among others.

The success of ML implementation required thorough evaluation. Here, a set of key metrics was established to measure the model’s performance on criteria like recall, precision, and F1-score. The primary objective was to achieve a high true positive rate while maintaining a low false positive rate.

The trailblazing investment and experimentation shown by Duke Energy and AWS MLSL speak volumes about the boundless potential of Machine Learning technologies in revolutionizing the utility industry. Their efforts echo the anticipation of a future where AI and ML aren’t just jargon but authentic tools shaping business operation and strategy. Stay tuned for the specifics on the performance details of their model in the forthcoming article.