Early Prediction of Lean Blowout from Chemiluminescence Time Series Data

ABSTRACT

Lean combustion systems are usually employed in gas-turbine engines, particularly in land-based applications, to reduce NOx emission. These combustion systems are often susceptible to lean blowout (LBO), which can be detrimental for operation and productivity of gas-turbine engines. An abrupt decrease in the equivalence ratio during a throttling operation, which is often encountered in power plants due to sudden decrease in load demand and also in aircraft engines at the time of landing, may lead to an unexpected LBO. From this perspective, online data-driven algorithms are deemed necessary for early prediction of potential transitions to near-blowout conditions. This procedure would provide the human operator/active controller with an appropriate lead time to alter the operating conditions so that the system can be brought back to a desired stable condition. The paper emulates pertinent conditions of LBO on a laboratory-scale apparatus of swirl-stabilized dump combustor with transient time series of CH* chemiluminescence data, where the objective is early prediction of LBO. The underlying algorithms are constructed based on a well-known statistical learning tool, called Hidden Markov Modeling (HMM), which can be used in the setting of supervised learning to discern near-blowout time series data from stable data. Being solely data-driven, the proposed methodology is model-free; it has been shown to be numerically efficient as well as sensitive to regime changes when the combustion system moves toward or away from LBO.

KEYWORDS:

• Lean blowout
• Hidden Markov modeling
• real-time monitoring

Nomenclature

Table

Acknowledgments

The authors acknowledge SPARC, MHRD, Government of India for supporting collaboration between Pennsylvania State University (PSU) and Jadavpur University (JU) through its Project No. P1065. Also, the financial support from RUSA 2.0 Programme (Reference no. R-11/274/19) of Jadavpur University is gratefully acknowledged. Any opinions, findings, and conclusions in this paper are those of the authors and do not necessarily reflect the views of the sponsoring agencies.

Supplementary material

Supplemental data for this article can be accessed on the publisher’s website

Additional information

Funding

The work reported in this paper has been supported in part by U.S. Air Force Office of Scientific Research (AFOSR) under Grant Nos. FA9550-15-1-0400 and FA9550-18-1-0135 in the area of dynamic data-driven application systems (DDDAS).