Large-scale data processing platform for laser absorption tomography

Abstract

Abstract Laser absorption tomography (LAT) has been widely employed to capture two/three-dimensional reactive flow-field parameters with a penetrating spatiotemporal resolution. In industrial environments, LAT is generally implemented by measuring multiple, e.g. 30 to more than 100, wavelength modulated laser transmissions at high imaging rates, e.g. tens to thousands of frames per second (fps). A short-period LAT experiment can generate extensive load of data, which require massive computational source and time for data post-processing. In this work, a large-scale data processing platform is designed for industrial LAT. The platform significantly speeds up LAT signal processing by introducing a parallel computing architecture. By identifying the discrepancy between the measured and theoretical spectra, the new platform enables indexing of the laser-beam measurements that are disturbed by harsh-environment noise. Such a scheme facilitates effective removal of noise-distorted beams, which can lead to artefacts in the reconstructed images. The designed platform is validated by a lab-based LAT experiment, which is implemented by processing the laser transmissions of a 32-beam LAT sensor working at 250 fps. To process a 60 s LAT experimental dataset, the parallelism enabled by the platform saves computational time by 40.12% compared to the traditional single-thread approach. The error-detection scheme enables the successful accurate identification of noise-distorted measurements, i.e. 0.59% of overall laser-beam measurements that fall out of the physical model.