Modern gas turbines use lean premixed combustors to meet stringent NOx regulations. Lean premixed combustors, however, have a narrow operating window and the performance is susceptible to changes in ambient conditions and operating variables. For example, the combustor may suffer from lean blowout issues when the ambient temperature drops in the winter. For the safe operation and optimized performance of the gas turbine, combustion tuning is periodically conducted with seasonal changes and maintenance cycles, which is usually done by tuning several key operating variables, such as the overall fuel flow, the fuel flow split between the combustor stages, and the compressor discharge air flow. Traditionally, the prerequisite for successful combustion tuning is the mapping between operating performance (e.g., emissions, combustion stability) and operating variables (e.g., fuel flow and air flow), which is obtained by full-scale combustion tests and field validation data. The traditional approach is both time consuming and cost intensive, given the complexity of identifying the link between the multiple performance objective parameters and the multiple operating input variables of the highly nonlinear system, especially when these performance parameters have mutual impacts. For example, increasing the fuel split in the non-premixed stage may improve the stability performance while it results in more NOx emissions.

Recent advances in data science suggest that machine learning can be a good fit for the optimization of the multi-objective, multi-input, highly nonlinear combustion system. In this paper, we present a cross-disciplinary study that combines machine learning with physics analysis for the combustion tuning of an F-class gas turbine. An artificial-neural-network-based model is developed to predict the performance (output) of the combustor, including NOx emissions, combustion instability (dynamic pressure), and mechanical vibration (acceleration). The features (input) of the model are identified by analyzing the key operating variables that impact the combustor performance, such as the overall fuel flow, the fuel flow split between the pilot nozzle, the non-premixed nozzle, and the premixed nozzle, and the inlet guide vane angle, et al. Field data from an F-class gas turbine power plant are used to train and validate the model. The trained model is able to predict the combustor performance at a fairly high accuracy within the space of the input variable. The model is then used in combination with the multiple objective particle swarm optimization method and the genetic algorithm to optimize the performance of the combustor. Results show that the machine-learning and data-driven method offers a promising alternative for combustion tuning at a low cost and fast turn-around.


Combustion Tuning for a Gas Turbine Power Plant Using Data-Driven and Machine Learning Approach

Division / Sponsor:
International Gas Turbine Institute (IGTI)

Paper Proceedings: https://asmedigitalcollection.asme.org/GT/GT2020/volume/84058

Turbo Expo - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.