Robust algorithms for quantifying noisy signals of optical fiber probes employed in industrial-scale practical bubbly flows

Optical fiber probing is widely employed in bubble/droplet measurements of gas–liquid two-phase flows. We have developed several types of optical fiber probes with very high S/N ratios and high detection performances for the gas and liquid phases in order to measure the bubble/droplet properties. Recently, demand for more accurate and reliable measurement for the practical purposes of probing is growing in industry. Many previous researches on the optical fiber probes resulted in effectively processing raw signals obtained under a laboratory condition. However in industrial settings of the optical fiber probing, the raw signals include a lot of inevitable noise that is very difficult to be eliminated and depresses the measurement accuracy to applying the conventional signal processing algorithms. To eliminate the noise, either the probe hardware or software or both must be improved. We have endeavored to solve this challenge by new software algorithms. The main purpose of the present study was to develop robust algorithms to analyze the noisy signals even containing strong pre-signals and overshoots. We developed two new types of algorithms, based on a histogram analysis and a median-filtering analysis, respectively, which exhibited their best performance in combination with each other. First, the improved performance of the algorithms was verified in comparison with that of a conventional algorithm of the min–max method. The basic thresholds for determination of gas and liquid phases in noisy signals could not be derived using the min–max method but were consistently obtained using the two new algorithms. Second, they were used to analyze the signals of a four-tip optical-fiber probe employed in bubble measurements in a large-diameter, high-concentration and multi-dispersed bubbly column. The new algorithms demonstrated robustness, and the results of the processing showed satisfactory agreement with the results obtained from high-speed visualizations.