A New Deep Learning Method for Temperature Determination from CPP Fs-Cars Spectra

Coherent anti-Stokes Raman Scattering (CARS) is a laser-based diagnostic dedicated to the measurement of temperature and major species concentrations in reactive flows. In the femtosecond excitation regime (∼100 fs), CARS enables instantaneous measurements at high repetition rates (i.e., 1–10 kHz), which makes it suitable for studying the temporal evolution of turbulent phenomena inherent to fast fuel mixing characteristics and subsequent dynamic processes inside modern propulsion systems. Among femtosecond CARS strategies, Chirped Probe Pulse CARS (CPP fs-CARS) process with a genetic algorithm has proved to provide reliable measurements of temperature, but suffers from long computation time. A new post-processing approach based on a Convolutional Neural Network (CNN) is proposed to perform temperature measurements from CARS spectra in quasi-real time, while maintaining good measurement accuracy. This method significantly reduces the processing time and paves the way for fast and efficient analysis of a large set of experimental spectra recorded at high repetition rates, thereby improving the practical usefulness of CPP fs-CARS in turbulent reactive conditions.Novelty and significance statement: This study presents a novel data processing strategy for the Chirped Probe Pulse Femtosecond anti-Stokes Raman Spectroscopy (CPP fs-CARS) diagnostic applied to reactive flows, which, to the best of your knowledge, has never been reported in this configuration. Although CPP fs-CARS enables temperature measurements in a single acquisition at high repetition rate, its widespread use has been limited by the substantial computational demands of usual spectral fitting with genetic algorithms. The proposed alternative, based on a Convolutional Neural Network (CNN), overcomes this limitation by enabling near real-time temperature estimation while maintaining a good accuracy required for combustion diagnostics. This advance significantly increases the practical value of CPP fs-CARS for time-domain studies of turbulent and unstable combustion.