Monday, April 6, 2020
KOMIPO Predicts Turbine Failures at Thermal Power Plants with AI
An AI-based Turbine Diagnostics Solution
KOMIPO Predicts Turbine Failures at Thermal Power Plants with AI
  • By Jung Min-hee
  • February 27, 2020, 10:32
Share articles

The turbine of Unit 1 of Shin-Boryeong Thermal Power Plant in Boryeong, South Chungcheong Province

Korea Midland Power (KOMIPO) and an artificial intelligence (AI) startup have introduced the world's first artificial intelligence (AI) program that can predict failures by diagnosing the turbine operation state of a thermal power plant.

KOMIPO recently began to run a deep learning-based turbine diagnostic system at Unit 1 of the Shin-Boryeong thermal power plant located in Boryeong, South Chungcheong Province. OnePredict, an AI-based industrial equipment diagnostics solution provider, supplied GuardiOne Turbine, a product optimized for turbine equipment diagnostics.

Shin-Boryeong Thermal Power Plant consists of Units 1 and 2, and was built for the first time in Korea by using 1,000MW ultra-supercritical power generation equipment. It is characterized by higher efficiency and lower carbon dioxide emissions than other coal-fired power generation systems. The turbine of Unit 1 turns 60 times per second and 3,600 times per minute.

However, 1,000MW turbines were difficult to operate because the soundness problem of last stage blades (LSB) has been unsolved for 10 years. The U.S. Electric Power Research Institute (EPRI) has worked on this issue for a long time, but has not found a way to diagnose the state of turbine blades. An LSB is part of the blades of a steam turbine, the plant's key facility. If a blade is damaged, the whole power plant is stopped. If it breaks, other equipment will be damaged and the plant will suffer a significant loss.

In response, KOMIPO and OnePredict have developed a deep-learning based blade diagnostic algorithm through cooperative research. They will jointly apply for a patent on the technology.

GuardiOne Turbine predicts device failures by using vibration data captured by dozens of sensors on rotors and blades which are turning parts of turbines. Its analysis incorporates deep learning algorithms with device experts’ knowledge and experience. GuardiOne Turbine combines the two methods to derive turbine health indicators and automatically categorize failure probabilities and types and display them on a dashboard screen.

KOMIPO is considering applying the AI-based predictive diagnostics solution to other 1000MW turbines.