The flame propagation in different spark-ignited engines was recorded with various imaging systems via endoscopic and full optical access. Images for hundreds of consecutive combustion cycles were analysed together with pressure-derived heat release rates and mass fraction burnt (MFB) to obtain relationships between engine output and the physical properties associated with flame propagation. Combined crank-angle resolved imaging of flame propagation and cycle-resolved measurement of flow-field was performed to improve the understanding of the effects of bulk in-cylinder flow on spark and turbulent flame propagation behaviours. 

The first part of the thesis describes the development of an algorithm with automatic dynamic thresholding to detect the line-of-sight projected flame boundary despite artifacts caused by the spark and the large dynamic range in image brightness across each time series. The unsupervised and computationally inexpensive algorithm segments a sequence of such images into three levels which correspond to spark, flame, and background. The main idea is to exploit the images' correlation in time to predict a suitable binarization threshold from the previous image, and the threshold is then corrected based on the now estimated foreground. The algorithm is adapted with a noise reduction model based on Fast Fourier Transformation (FFT) to minimize the existing periodic pattern noise from the high-speed CMOS detector.  The algorithm was compared with two standard segmentation methods from the literature. Also, the robustness of the scheme was examined with a set of raw data images captured from various imaging systems and engines through endoscopic and full optical access.

The second part of the thesis compares selected imaging systems used to visualize flame propagation in engine cylinder via endoscopic and full optical access. This work investigates the image quality achievable with a large-aperture endoscope system and high-speed cameras in terms of detecting the premixed flame boundary in spark-ignited engines by chemiluminescence. The imaging systems compared here were cinematography with a CMOS camera, both with and without an intensifier, the latter variation being used in a four-cylinder automotive engine as well as in a single-cylinder motorcycle engine. A fundamental problem in evaluating the systems’ efficacy is that it is not clear what constitutes “correct” detection of the flame boundary. To help clarify this question, the endoscopic results are compared among each other and to a “best-case scenario”, which was unintensified high-speed imaging with a large-aperture commercial camera lens in an engine with full optical access.

Finally, stroboscopic particle image velocimetry (PIV) and Mie-scattering imaging were conducted to map the velocity vectors of flow-field and to obtain 2-D planar details of the flame front on the vertical tumble plane that could not be identified by chemiluminescence imaging due to the projected line-of-sight nature of the latter technique. Combined high-speed flame chemiluminescence and double-frame PIV imaging were also conducted to study the relationship between flow field and turbulent flame growth.