Improving Steam Turbine Performance with Industrial Internet

Abstract

Most of the global electricity production is done in thermal power plants with steam turbine powered generators. The steam turbine is one of the most critical components in a thermal power plant, and non-optimal performance affects the overall profitability of the plant significantly. Industrial Internet utilizes collective measurement data to help in decision making and optimization of different business processes. It is a rapidly developing technology and solutions based on this technology are gaining a stronger foothold in different, varying markets. The thesis aims to investigate how Industrial Internet could in the best way be utilized to improve the profit-making performance of steam turbines, with the solution being able to be productized as a profitable product into the product lineup of a turbine automation company. The thesis follows a typical product concept development process. First, the need of the market is researched through literary study and with qualitative interviews done to both thermal power plants and turbine automation system sales representatives. Based on the market need, acknowledged capabilities of Industrial Internet, and the existing product offering of the steam turbine automation company, a product concept is developed and suggested through iterative action research. Two main methods of increasing the profit-making performance of steam turbines are found during the market need study, optimizing the usage of the turbine such that the most profitable amounts of process steam, district heat, and electricity are continuously produced, and optimizing the maintenance of the turbine such that maintenance, failure, and condition degradation costs are minimized. Condition monitoring is required for any maintenance optimization to be done, and the process and quality of steam turbine condition monitoring can be improved with better utilization and analysis of data that is available. A product concept automating and adding quality to steam turbine condition monitoring is developed. The concept aims to monitor the behavior of steam turbines by performing cloud based data analysis on both process and vibration data. By utilization of machine learning, digital twin models are formed which allow indication of anomalies and long-term changes in the behavior of chosen measurements or parameters. Tracking of proper process values also allows indication of conditions that are leading to condition degradation of the turbine. The concept works as a tool for a condition monitoring expert to diagnose emerging faults and current turbine condition, based on which they can provide the plant with recommendations of future actions. A product based on the proposed concept would fit well into the product lineup of the turbine automation company. Small changes in the process conditions of a steam turbine can affect the monitored measurement values and parameters significantly. The training data used by machine learning to form the digital twin models should contain the typical process conditions that are faced for the anomaly detection to function properly, and the capability of the system to learn as it is used is essential for successful automation of the condition monitoring process. Further study on how successful a functional concept based on the proposition is in automating and improving the quality of steam turbine condition monitoring should be done to prove the added value coming from the concept. This can be done with a live implementation of the concept that is running preferably at a combined heat and power plant and working as a part of the steam turbine condition monitoring process.