Compression-ignition (CI) engines are typically optimized for a narrow range of fuels and may fail when operating on fuels with ignition properties outside their intended range. Situations such as fuel shortages or mandates to use cleaner, newly developed fuels to meet environmental standards require the development of CI engines with flex-fuel capability. Reliable operation under these conditions is best achieved by incorporating a sensor that measures ignition properties, such as the derived cetane number (DCN), of liquid fuel prior to injection and provides this information to the engine control system. The objective of this work was to develop such a sensor. Given the need for low size, weight, and power consumption (SWaP) for onboard deployment, optical spectroscopy emerged as a promising solution. This work focused on developing a sensor based on fingerprint-region infrared (IR) absorption spectroscopy operating in the 7 – 14 µm range.
Quantum cascade laser (QCL)-based approaches were identified as the most effective spectral measurement technique considering the SWaP constraints. Both external cavity quantum cascade lasers (ECQCLs) and QCL arrays composed of distributed-feedback quantum cascade lasers (DFBQCLs) were evaluated as potential light sources. The performance of these approaches was assessed based on the accuracy of DCN predictions generated by regression models trained using hydrocarbons and real fuels. The performance was validated through both simulations and experiments.
Simulation results indicated that a QCL sensor performed well when utilizing more than 16 wavelengths, with both ECQCL and QCL-array sensors providing comparable prediction accuracy. Using 32 wavelengths, at least 80% of test samples had prediction errors within 10%.
To validate the simulation results, experiments were conducted using two ECQCLs packaged in a single device, covering a total spectral range of 7.4 – 12.8 µm. Absorption spectra for heptane, ARL-CN30 fuel, and 50 vol.% ARL-CN35/CN40 fuel were measured using the attenuated total reflectance (ATR) sampling technique and used to infer DCN with 32 selected wavelengths.
The measured spectra exhibited baseline variations compared to those in the spectral database used for model development, resulting in poor DCN predictions with an average error of 16.7% across the three test samples. After applying baseline fitting to the measured and reference spectra, the average prediction error was significantly reduced to 7.3%. The errors prior to baseline correction reflected the performance of a QCL-array sensor, which measures a limited number of wavelengths, making baseline correction challenging. In contrast, the reduced errors after baseline fitting demonstrated the advantage of an ECQCL sensor, which provides continuous spectral coverage, facilitating effective baseline correction.
Since an ECQCL’s tuning range includes all available wavelengths without additional SWaP costs, more wavelengths can be used to further enhance prediction performance. Simulation results showed that using 71 wavelengths yielded the best performance, reducing the average DCN prediction error to 4.1%. To further minimize the sensor’s size and complexity, a single ECQCL was considered, and the wavelengths were reselected from the spectral ranges of the two ECQCLs (7.4 – 9.09 µm and 9.09 – 12.8 µm). The 7.4 – 9.09 µm range yielded comparable performance to the full 7.4 – 12.8 µm range, with an average error of 4.3%, whereas the 9.09 – 12.8 µm range resulted in a higher average error of 13.4% due to complex overlapping spectral features. These findings highlight the potential of using a single ECQCL with about 300 cm⁻¹ spectral coverage when the appropriate range is selected.
Since the performance of QCL-array sensors is affected by baseline variations, addressing baseline variations becomes essential. Reduced baseline variation, or improved immunity to them, would also benefit ECQCL sensors in scenarios where baseline fitting is less effective.
To minimize baseline variations, a dual-polarization referencing method was considered. The method uses the perpendicular-polarized signal, which has a lower effective path length, as the reference signal, while the parallel-polarized signal as the sample signal. By eliminating the need for a conventional reference, such as air, this method enables a more rugged sensor design that is less susceptible to baseline drift. The method was validated using an FTIR and demonstrated that absorption spectra could be reproduced under instrumental drift without the need for a conventional reference. However, when validated with ECQCLs, baseline drifts were observed under instrumental drift conditions, likely due to the polarization sensitivity of certain optical elements used in the experiment. Despite this challenge, the dual-polarization method remains valuable, particularly because it has the potential to remove the need for air as a reference. Therefore, further efforts to address baseline drifts and refine the dual-polarization method are recommended to enhance its effectiveness.
Other strategies for minimizing baseline variations include improving sensor ruggedness. However, if baseline variations cannot be eliminated, they must be managed properly. One approach is baseline fitting, which could be feasible for QCL-array sensors if additional wavelengths, particularly those unlikely to exhibit absorption across a wide range of relevant fuels, are measured. Another strategy, either as an alternative or in conjunction to baseline fitting, involves training regression models using spectra with diverse baseline variations to enhance robustness. The choice between QCL-array and ECQCL sensors ultimately depends on how well baseline variations can be managed. Thus, further investigation into baseline mitigation strategies is recommended to inform this decision.