Automated Flame Analysis for Optical Engine Combustion Diagnostics Using Computer Vision Analisi automatizzata della fiamma per la diagnostica di combustione in motori ottici tramite visione artificiale

Optical access internal combustion engines allow direct visualisation of in-cylinder flame propagation, revealing combustion phenomena that pressure-based diagnostics alone cannot capture. Yet the analysis of high-speed combustion videos remains largely manual. Expert operators must inspect thousands of frames per experiment — a slow, subjective process that does not scale to large cycle-to-cycle variability studies. The problem grows harder when multiple fuels are involved, because hydrogen, methane, and their blends produce flames with very different luminosity and shape. This thesis presents an automated image-processing system for detecting and characterising flames in optical engine experiments at CNR Istituto Motori. The system uses a seven-stage pipeline — CLAHE enhancement, Gaussian filtering, Otsu thresholding, morphological operations, region-of-interest masking, area filtering, and fuel-specific geometric validation — with separate, optimised parameter sets for hydrogen (H₂), methane (CH₄), and hydrogen–methane blend (CH₄+H₂) combustion. A data indexing module synchronises optical frames with crank angle position through a dynamic frame counting algorithm, correcting the previous static assumption and capturing 264 total flame frames at 25 frames per combustion sequence (52.8 degrees CAD). An interactive user interface guides parameter selection, cutting configuration time from roughly 45 minutes to about 8 minutes per video. Validation showed clear differences between fuels. Methane flames had high mean brightness (141.7 on a 0–255 scale) and moderate circularity (0.53), while hydrogen flames were dim (46–52) and irregular (circularity 0.29–0.35). Blends fell in between (brightness 83.5, circularity 0.46). Flame area variability ranged from a coefficient of variation of 21.7% for blends to 57.5% for methane, with a combustion duration CV of 18.2%. The backindex system was validated through sequence extraction, confirming correct low–peak–low pressure profiles with a mean combustion duration of 185.9 ± 33.8 degrees crank angle and peak alignment within ±1.4 frames across cycles.