Questions Utilities Don’t have to ask with AI
Months before an engineer at a natural gas or coal power plant can smell smoke emanating from a turbine blade, vibrations and small fluctuations in temperature are causing stresses to the system that could be leading to a potential breakdown. Turbine blade failures could cost a utility as much as $10 million dollars in repair. For decades, utility professionals conducted maintenance based on preset schedules or when operators happened to hear a rattle, see abnormal vibration or smell smoke. This type of maintenance schedule was common practice throughout generation, transmission, and distribution. In the last decade, however, IoT devices in the form of smart sensors have become more common for utilities (Figure 1). While many utilities are still in the planning phase for IoT, those that have deployed sensors are capturing valuable data, which is changing how utilities perform asset maintenance.
Utility Deployment of IoT Devices
Source: “The Autonomous Grid in the Age of the Artificial Intelligence of Things” Zpryme Research 2019
Capturing asset data with IoT sensors on turbine blades, at voltage converters, and at transmission substations is just the first step for utilities to change how they perform maintenance. To truly transform operational efficiency, that data must be analyzed quickly to make predictive and even prescriptive business decisions which is where Artificial Intelligence comes in. Most utilities have not implemented AI software at the same level that they have deployed IoT sensors, but machine learning data analysis is the key to unlocking the holistic understanding of the asset life cycle.
One utility that has made tremendous strides in combining IoT sensors and AI to do predictive maintenance is Duke Energy. Duke’s business goal was to improve efficiency and reduce downtime with condition-based maintenance at the asset & component level. To accomplish that goal, they developed the SmartGen program, which integrated equipment monitoring, advanced pattern recognition algorithms, and an asset health management software system to perform diagnostics. With more than 30,000 sensors on assets measuring a wide variety of inputs, the Advanced Pattern Recognition AI algorithm became the essential component to understanding the asset health.
Source: Duke Energy
The SmartGen program was first deployed in January of 2013. By 2018, Duke had more than 500,000 points of data flowing into 11,000 Advanced Pattern Recognition models. The results to date have been monumental from a financial perspective, with more than $131 million dollars in avoided costs. The fusion of sensor data, machine learning, and advanced algorithms is leading to an Artificial Intelligence of Things (AIoT) that has profound implications for utilities. The ability to make strategic decisions based on data and AI models will lead to better asset-life cycle management, and improved grid operations. From an everyday perspective, engineers won’t have to ask “does this smell funny to you?”
Duke Energy will be discussing their approach to Artificial Intelligence and IoT devices on a webinar with Zpryme. The webinar will focus on AI at utility companies and explore how AI and IoT work together to deliver everything from improved threat detection to better customer engagement. Zpryme will be speaking with Norv Clontz, Duke Energy’s Director of Data Science Innovation, and SAS’s Mike Smith. You can register for free using the link below.
The Autonomous Grid in the Age of Artificial Intelligence
Thursday, January 10th, 2019
