Image Based Liquid Level And Flow Measurement Technology For Micro-Chemical Processes

Micro-chemical technology can enhance the chemical process to have higher product selectivity, faster speed, lower cost and risk. However, the mini and hermetic micro-chemical equipment raises the difficulty of measuring and controlling micro-chemical process, which consequently limited the popularization of micro-chemical technology. Hence, this paper aims at developing a low cost technology for liquid level and flow rate measurement with image based approach.For the liquid level measurement problem, this paper focuses on the scenario in micro-separator, which is characterized with strenuously movement, possible change in color and temperately disappearance of liquid bound due to bubble and impurity. Hence, the image based method needs to be fast, robust as well as low cost. To achieve these requirements,3algorithms are implemented, including corner point detection, mean shift and particle filter algorithm. Furthermore, the efficiency of particle filter is discussed and enhanced by template update mechanism, histogram calculation optimization and penalization. According to the experiments, the corner point and mean shift are quite sensitive in impurity case while particle filter and enhanced particle filter are more robust. Meanwhile, on the prospect of speed, mean shift has the highest speed while particle filter has the lowest. The enhanced particle filer ranks in the middle position but is just slightly lower than mean shift.For the flow measurement problem, the approach is to calculate the drop volume by contour extraction and fitting, then get the flow rate. First, connected component labeling algorithm is implemented to exact original contour from binary image. Then, a more accurate contour can be obtained by Laplace-Young equation based fitting. Newton-Raphson method, coordinate rotator method and genetic algorithm are used to optimize the parameter of Laplace-Young equation so that the contour can best fit that obtained by connected component labeling algorithm. According to the experiments, genetic algorithm has the best optimization ability while has the slowest speed. Newton-Raphson method’s accuracy is slightly lower than genetic algorithm but is much faster. Coordinate rotator method has the poorest fitting ability and is not faster than Newton-Raphson method.