Senior Principal Applications Engineer at MKS Instruments, specializing in near-infrared and infrared process analytical applications and data modeling.
The Methane Number (MN) is a measure of the resistance of fuel gases to engine knock. The MN can be described on an engine by comparing the fuel to pure methane, which has an MN of one hundred and pure hydrogen which has a MN of zero. Several models and calculators have been developed to calculate an estimated me MN based on a thorough compositional analysis.
Operators of large scale gas engines can improve efficiency and reduce maintenance by tuning and operating engines under optimum conditions. These large-scale engines, typically ranging from 1 MW to 20 MW per engine (depending on bore size and cylinder configuration), are typically used in ships or power plants and it has been shown that the ability to measure the MN in real-time allows for application of control strategies to avoid knocking and optimize engine performance with regards to both efficiency and emissions.
A fast, rugged spectroscopic sensor based on the MKS Tunable Filter Spectroscopy (TFS) platform has been evaluated for use in measuring MN. The TFS analyzer operates in the Near Infrared spectral region which has been used extensively for the analysis of hydrocarbon composition and performance properties. MKS P5 TFSTM analyzer is currently widely used in the natural gas processing and distribution industry and uses two filters, one to scan the spectral range containing features due to hydrocarbon vibrations, and a second filter to capture a spectral range where CO2, CO vibrations occur.
Spectra of gas blends were collected and the MN was calculated using an industry standard calculator. The calibration spectra included varying concentrations of the gases typically found in natural gas and liquified natural gas (C1-C5 + carbon dioxide, carbon monoxide and nitrogen). A Partial Least Squares Regression (PLSR) model was developed to correlate the spectral response to the MN. A strong correlation was obtained, although some nonlinearity was observed. In order to avoid the need to run many gas blends to calibrate a sensor an alternative approach to calibration was developed where spectra of pure gases (Methane through Pentane + CO2) are collected and then a synthetic calibration set is calculated to cover the expected composition ranges. The MN was calculated for each spectrum and modeled directly against the spectra using PLSR.
We will present two sets of data, one based on gas mixture spectra, and a second set of results based on mathematical mixtures of the pure gases. Initial results demonstrate our ability to meet the generally required accuracy of +/- 2 in MN. Challenges in handling non-linearities, integrating external data, dealing with multiple MN standards and implementing the models developed on the embedded platform will be discussed.