J. Bosarge, T. Summe
Forward Edge-AI, Inc.,
United States
Keywords: Raman spectroscopy, artificial intelligence, federated learning, fuel contamination detection, field diagnostics, dual-use technology, low cost
Summary:
Objective: This Small Business Innovation Research (SBIR) Direct-to-Phase II effort (Contract FA810024C0003) seeks to develop and validate the Blaise™ AI-augmented Raman spectrometer, a handheld, low-cost device capable of detecting and quantifying hydraulic fluid and polyalphaolefin (PAO) contamination in jet fuel. The objective is to reduce F-35 and broader Air Force sustainment costs by enabling on-site fuel quality analysis, mitigating the need for costly laboratory testing, and minimizing aircraft downtime associated with fuel defueling and maintenance. The project aims to achieve ASTM certification for standardized fuel contamination detection and pave the way for dual-use commercial applications. Methods: Forward Edge-AI, Inc. is advancing a miniaturized Raman spectrometer integrating AI-based spectral analysis, deep neural networks (DNN), and federated learning for decentralized model improvement. The device uses an 808 nm VCSEL laser with a Raspberry Pi infrared-specific detector and a dichroic mirror to enhance stability, signal clarity, and portability. The optical system was optimized through iterative design, 3D modeling, and additive manufacturing. Concurrently, a DNN is being trained using Raman spectra generated by Sam Houston State University, including controlled contamination levels down to 0.0625 %, across multiple jet-fuel types and additives to ensure model generalization. Testing procedures follow ASTM D1655 as a benchmark, with comparative validation using Gas Chromatography-Mass Spectrometry (GC/MS) and Fourier Transform Infrared Spectroscopy (FTIR). A Surface-Enhanced Raman Spectroscopy (SERS) substrate and emerging microfluidic chip architecture are incorporated to improve sensitivity and reduce cost. Iterative user evaluations inform interface design for intuitive field operation, compliant with MIL-STD-810D environmental requirements. Future interlaboratory studies (ILS) and inverse computational spectra generation (ICSG) methods will benchmark thermal stability predictions against ASTM D3241. Federated learning through Microsoft’s Project Florida will enable secure, distributed model updates while preserving data integrity. Results / Expected Results: Prototypes have demonstrated reliable Raman spectra capture and accurate classification of hydraulic and PAO contaminants, achieving 94–96 % detection accuracy in laboratory testing. Optical redesign, adding a dichroic mirror, input slit, and laser dump, has increased stability and reduced fluorescence, culminating in a patent application filed December 2024. The modular sample holder now accommodates both direct and SERS-enhanced measurements. AI compression techniques (TensorFlow Lite) are reducing model size by nearly 90 %, supporting real-time inference on edge devices such as Android phones. Upcoming milestones include MIL-STD-810D certification, ASTM interlaboratory validation, and production of 100 commercial prototypes for DoD and aviation industry evaluation. Significance and Impact: Blaise introduces a transformational capability for real-time aviation fuel quality assurance, offering analytical precision comparable to laboratory spectrometers priced between $30,000 and $80,000, at a projected unit cost of $800 and per-sample cost under $0.50. By enabling rapid detection of contamination that causes 30–45 days of aircraft downtime and 80 man-hours of rework per incident, the system supports Department of the Air Force sustainment cost-reduction goals and mission readiness. Beyond military aviation, this standardized, AI-driven Raman platform holds dual-use potential across commercial aviation, energy, and industrial fluid diagnostics, establishing a scalable foundation for AI-assisted spectroscopic analysis and ASTM-certified field testing.