The relevance of the research is caused by the need of profitability improving in oil and gas sector with automated control of petroleum-containing fluid separation, particularly, energy cost reduction without tank oil quality loss. The automated control can be built on the basis of mathematical models that can be obtained with natural experiment. The main aim of the research is to develop adaptive automated information and measurement system for experimental petroleum-containing fluid separation dynamics estimation with different water-in-oil emulsion layer stability and interbed diffusion for data collection for mathematical model development. Object: technological process of petroleum-containing fluid separation in the context of lab bench on the gravity type separation basis. Methods: natural experiment, computer vision, convolutional neural networks, machine learning, digital image segmentation problem, volume ration calculation, transient processes, physics and chemistry experiment automation. Results. The authors have developed an adaptive information and measurement system on computer vision and convolutional neural networks basis, which allows estimating petroleum-containing fluid separation dynamics with different water-in-oil emulsion layer stability and sharpness of layers borders. The basis of functioning of adaptive information measurement system is software, that allows performing experiment considering different qualitative and quantitative conditions. The paper considers the algorithm for discretization period with respect to reaction length. The system was tested on lab bench and layers volume ratio was calculated in real-time.