Multi-Species Sensing Using CEAS and DNN in Shock Tube Kinetics
Item Type Conference Paper;Poster
Authors Mhanna, Mhanna;Sy, Mohamed;Elkhazraji, Ali;Farooq, Aamir Citation Mhanna, M., Sy, M., Elkhazraji, A., & Farooq, A. (2022). Multi- Species Sensing Using CEAS and DNN in Shock Tube Kinetics.
Imaging and Applied Optics Congress 2022 (3D, AOA, COSI, ISA, PcAOP). https://doi.org/10.1364/3d.2022.jtu2a.4
Eprint version Post-print
DOI 10.1364/3d.2022.jtu2a.4
Publisher Optica Publishing Group
Rights This is an accepted manuscript version of a paper before final publisher editing and formatting. Archived with thanks to Optica Publishing Group.
Download date 2023-12-03 17:31:27
Link to Item http://hdl.handle.net/10754/685231
Multi-Species Sensing Using CEAS and DNN in Shock Tube Kinetics
Mhanna Mhanna, Mohamed Sy, Ali Elkhazraji, Aamir Farooq
King Abdullah University of Science and Technology (KAUST), Physical Science and Engineering Division, Thuwal, 23955-6900, Saudi Arabia Author e-mail address: [email protected]
Abstract: Simultaneous multi-species sensing in shock tube experiments are performed using a single laser source. A tunable DFB-ICL emitting near 3.3 µm, off-axis CEAS, and DNN enabled sensitive and selective measurements of target species. © 2022 The Author(s)
Abstract: Simultaneous multi-species sensing in shock tube experiments are performed using a single laser source.
A tunable DFB-ICL emitting near 3.3 µm, off-axis CEAS, and DNN enabled sensitive and selective measurements of target species. © 2022 The Author(s)
1. Introduction
Measurements of target species at high temperatures have witnessed remarkable progress due to utilizing laser sensors which provide non-intrusive, high-temporal and -spatial resolution sensing. While laser absorption spectroscopy (LAS) has been successful in quantifying single species time histories in shock tube kinetics, multi-species detection is still a major challenge. Such a capability can play a major role in studying flow physics problems, such as reaction kinetics at elevated temperatures which involves several evolving species. For example, polycyclic aromatic hydrocarbons can be formed through pyrolysis/oxidation of aromatic components of distillate fuels. To minimize harmful emissions, it is crucial to understand the detailed chemistry of such components. Gudiyella and Brezinsky [1]
chose propylbenzene as the surrogate fuel for alkyllbenzenes so that both the aliphatic nature of its side chain (C3H7) and the aromatic nature of its ring (C6H5) can be studied. In this paper, we report a diagnostic based on a mid-IR laser source and cavity-enhanced absorption spectroscopy (CEAS) to develop a sensor for time-resolved speciation of propylbenzene pyrolysis in a shock tube. We employed deep neural networks (DNNs) to distinguish the similar broadband absorbance spectra of evolving species.
2. Sensor Description
We simulated the pyrolysis of propylbenzene (C9H12) using the chemical kinetic model proposed by Darcy et al. [2].
Simulations were performed using CHEMKIN-PRO in a homogenous batch reactor with constant volume assumption over T = 1000 – 1300 K, P = 1 atm, 𝜒𝐶9𝐻12 = 0.1 – 0.2%. Among the top 20 consumed/produced species during the first 2 ms, benzene, toluene, ethylbenzene, ethylene, styrene, and propylbenzene were selected as target species for our absorption sensor. This is based on room-temperature absorption cross-sections near 3.3 µm of these species in the PNNL database [3], as high-temperature data are unavailable in literature. In this wavelength region, the remaining species are considered as non-absorbers due to their negligible absorption cross sections and/or mole fractions. A CEAS setup was implemented to amplify the absorbance of target species at low mole fractions. Figure 1 shows the optical setup of the designed sensor. A DFB-ICL (Nanoplus) emitting near 3.3 µm is co-aligned with a 670 nm red laser (Thorlabs) to facilitate optical cavity alignment. Two ZnSe mirrors of 99.25 ± 0.3 % nominal reflectivity were used to form a cavity in the low-pressure shock tube facility at KAUST, which has been described in a previous work [4]. A 10 kHz ramp signal was fed to scan the laser over 3038.6 – 3039.8 cm-1. The laser was aligned in an off-axis cavity to significantly suppress spurious coupling noise compared to an on-axis cavity [5]. The transmitted signal was collected via a focusing lens onto a DC-coupled, TE-cooled photodetector (Vigo Systems). Scan time was converted to wavenumbers via a 7.62 cm germanium etalon.
Fig. 1. Optical schematic of the sensor setup
3. Results and Discussion
It is essential to measure the mirror reflectivity as the nominal reflectivity provided by the manufacturer can have high uncertainty. For that, experiments were performed on known room temperature mixtures of benzene, toluene, ethylbenzene, ethylene, styrene, and propylbenzene in argon, and later verified at shock heated temperatures. Using the measured CEAS absorbance and simulated single-pass absorbance, the mirror reflectivity was obtained to be 𝑅 = 1 − 𝐴𝑠𝑝
𝑒𝐴𝐶𝐸𝐴𝑆−1= 98.98 ± 0.012 %, within a relative error of 0.27% as compared to the nominal reflectivity. This results in an absorbance gain of ~ 100, enabling low mole fraction detection of target species. Absorption cross- sections of these species are unavailable in literature at high temperatures, and thus were measured over 1000 – 1250 K. Figure 2 shows that the measured cross-sections of the target species at 1200 K are similarly broad, which makes it challenging to distinguish their contributions in a cumulative absorbance spectra particularly in the presence of measurement noise. To overcome this challenge, a denoising process was applied on the measured total absorbance spectrum based on 10 deep neural networks (DNN) models trained and tested with various signal to noise ratios (SNR) with a random 80/20 split [6], containing six hidden layers with 256, 256, 128, 64, 32, 32, and 16 nodes, respectively.
Each model resulted in a composite absorbance vector which was fed into a multidimensional linear regression (MLR) model to split total absorbance into contributions from reference target species. This resulted in 10 slightly different solutions, where the best one was selected based on lowest p-value criterion. The obtained solution was used to calculate instantaneous concentrations of target species during the pyrolysis of 0.163% propylbenzene in argon at 1200 K and 1 atm. A comparison between mole fraction time histories of target species obtained by our sensor and CHEMKIN PRO simulations using the reference model [2] over the first 2 ms is shown in Fig. 3. The significant differences in the pyrolysis rate and formed amounts in the experimental results indicate that the chemical kinetic mechanism needs to be improved. Such a sensor can thus be highly valuable for multi-species detection in complex reaction kinetics systems.
Fig. 2. Measured absorption cross sections of benzene, toluene, ethylbenzene, ethylene, styrene, and propylbenzene at T = 1200 K
and P = 1 atm.
Fig. 3. Real-time measured and simulated target species mole fractions during the pyrolysis of 0.163% propylbenzene in argon at T
= 1200 K and P = 1 atm.
4. Conclusions
A laser sensor based on cavity enhanced absorption spectroscopy has been developed to measure low concentrations of benzene, toluene, ethylbenzene, ethylene, styrene, and propylbenzene. The high mirror reflectivity of the cavity amplified measured absorbance by two orders of magnitude to enable low mole fraction measurements. High- temperature absorption cross-sections of all target species were measured. The sensor can selectively and simultaneously measure all target species via multidimensional linear regression (MLR) after a denoising process using deep neural networks (DNN). Finally, the sensor was applied for multi-speciation of propylbenzene pyrolysis in a shock tube.
5. References
[1] Gudiyella, S., and K. Brezinsky. "The high pressure study of n-propylbenzene pyrolysis." Proceedings of the Combustion Institute 34.1 (2013): 1767-1774.
[2]Darcy, Daniel, et al. "A high-pressure rapid compression machine study of n-propylbenzene ignition." Combustion and flame 161.1 (2014):
65-74.
[3] Sharpe, Steven W., et al. "Gas-phase databases for quantitative infrared spectroscopy." Applied spectroscopy 58.12 (2004): 1452-1461.
[4] Shakfa, Mohammad Khaled, et al. "A mid-infrared diagnostic for benzene using a tunable difference-frequency-generation laser." Proceedings of the Combustion Institute 38.1 (2021): 1787-1796.
[5] Mhanna, Mhanna, et al. "Cavity-Enhanced Measurements of Benzene for Environmental Monitoring." IEEE Sensors Journal 21.3 (2020):
3849-3859.
[6] Haider, Syed Kamran, et al. "Performance enhancement in P300 ERP single trial by machine learning adaptive denoising mechanism." IEEE Networking Letters 1.1 (2018): 26-29.