Optimizing Vortex Shedder for Vortex Flow Meter Using Deep Learning

The vortex flow meter is widely employed within the industry as one of the primary flow measurement devices. It operates on the principle of vortex shedding. One of the primary challenges in the design and development of this flow meter lies in the design of the vortex shedder. The vortex shedder must generate strong vortices to enable accurate sensor detection. Given the necessity for extensive experiments and computational fluid dynamics (CFD) analysis in the design of new bluff bodies, this process is both time-consuming and expensive. To address this issue, we employ machine learning (ML) models for the development of new vortex flow meters. We have developed two ML models that aid in the design of new bluff bodies based on deep learning (DL) architectures. These models were trained on both a fully connected deep neural network (DNN) and a convolutional neural network (CNN) architecture, yielding comparable performance. To facilitate model training, we compiled a comprehensive dataset from existing literature, paying particular attention to the symmetrical profiles of bluff bodies. The dataset exclusively contains point coordinates (X, Y) from the right half of the bluff body s cross-section, this consideration leads to more smooth generated shape. The first model was designed to generate new bluff body geometries based on input parameters such as Reynolds number, Strouhal number, and other geo-metric parameters. The second model focused on predicting the linearity of Strouhal numbers for these generated shapes. The models demonstrated an error rate of approximately 2.80% in the validation phase. These results indicate the potential utility of this approach in the initial phases of bluff body design.