Free surface vortices in hydraulic intakes can lead to air entrainment, vibrations, and energy loss. This study aimed to optimize an Acoustic Doppler Velocimeter (ADV) for measuring tangential velocities in vortices. The used statistical methods include calculating the median, mode, and arithmetic mean without outlier removal; and removing outliers using truncated mean, Z-Score, box plot, histogram, and a two-standard deviation range. The Machines), and DBSCAN (Density-Based Spatial Clustering of Applications with Noise). Experiments were conducted at Fr = 0.85, 1, and 1.27. In order to determine the data acquisition duration, a long-term experiment was conducted, and based on the output graph, 30 seconds was determined as the appropriate time. The results showed that the mean and median at all Fr provided smooth and physical profiles, while the mode method produced unrealistic fluctuations due to the multimodal nature of the data. The outlier removal methods produced almost overlapping profiles with an RMS deviation of less than 1.5%, indicating the stability of these methods and the appropriate quality of the ADV data. Among the machine learning methods, the Isolation Forest algorithm provided the most accurate and stable results, and its behavior was most consistent with valid statistical methods. Comparison of the results with the Rankine model also showed that the measured velocity profile accurately reproduces the linear growth of the velocity in the core and the drop in the outer regions of the vortex. Overall, this research shows that combining ADV data with robust data processing methods specifically, trimmed mean, Z-score, and Isolation Forest significantly increases the accuracy and reliability of eddy velocity profile determination and can be used in the analysis and design of eddy control systems in reservoirs.

Keywords