V. Hanagandi, A. Metcalf, D. Landay
Optimal Solutions, Inc.,
United States
Keywords: infrared spectroscopy, machine learning, jet fuel characterization, aviation chemicals analysis, portable sensors
Summary:
Contaminated aviation fuel can cause significant damage ranging from fuel system corrosion, clogging, instrumentation failures, and even constricting the fuel supply to engines causing them to stall during flight. Our innovation addresses this critical market need. The goal is to develop a sensor platform which enables near real-time quality assessment of aviation fuels and chemicals. Our solution will lower the cost and cycle-time in fuel analysis and increase the adoption of rigorous fuel analysis techniques, thereby reducing delays and risks in military and civilian air travel. We demonstrate 1) a customized, compact, and low-cost sensor suite for fuel contamination analysis, 2) data processing algorithms, and 3) communication software – all aimed at showcasing the platform’s ability to take readings from fuel samples in the field and transmit them for analysis and interaction with remote technician(s). Our team is comprised of accomplished individuals in the field of portable spectroscopy, optics, fuel analysis, and data analysis. We are an emerging leader in portable sensing and are presently leveraging this expertise in several active SBIR projects. Our product features 1) a sensor suite (which is comprised detectors and sampling accessories), 2) data processing software, and 3) communication interfaces. The sensor is custom-built for the purpose of aviation fuel analysis, the data processing software will leverage cutting-edge machine learning (ML) approaches, and the communication interfaces will eventually comply with DOD IT standards. The sensor will be accessible to the field technicians via an iPhone interface hence providing high portability and secure connectivity. A custom-built portable sensor suite is at the heart of our product. Our unique design features a wide spectral range starting from the UV-vis range and extending into the Mid-infrared (MIR) region. Our platform is scalable to incorporate additional sensors and it also includes an innovative common sampling port, which will be used to present the sample to the detectors. Given the unique and customized sensors we aim to build, the spectral data will span a wide range of wavelengths. Traditional chemometric techniques struggle to accurately capture the complex relationships between the absorption spectral data provided by the sensor and the attributes of the analyte. Machine learning (ML) algorithms provide powerful methods to yield highly accurate and robust sensor calibration models. Our ML-based calibration models will augment the remote technicians’ established models, which are typically based on commercially available chemometric software. This approach will increase the speed and accuracy of analysis. Using over a hundred samples obtained from fuel testing laboratories, we are able to develop our sensor platform. We have achieved good classification of fuels and quantification of properties using our portable sensors and ML. Our results demonstrate the significant benefits of using nonlinear ML techniques to give indications to the end users when fuels are fit-for-use and when they are contaminated and need to be quarantined. The attached PDF file presents a schematic of our innovation. This work was funded by a grant from USAF.