Estimating dynamic behavior of trawl codend based on machine learning models

Understanding the hydrodynamic behaviors and fluttering motions of trawl codends is crucial for improving trawl selectivity and controlling fish escape while retaining desired catch. This study investigated factors like mesh size, twine diameter, codend length, knot direction, and twine material on codend fluttering in a flume tank. Fast Fourier Transform (FFT) visualized the Strouhal number, fluttering amplitude, and drag force evolution. Backpropagation (BP) neural networks were used to predict oscillatory characteristics under various conditions. Results showed that empty codends had a waist-to-end hanging ratio > 0.8, while codends with catch had ratios < 1.2. T₉₀ and polyethylene meshes maintained mesh openings better than T₀ and nylon. The mean drag force coefficient decreased with Reynolds number and then tended to level off slowly. Extreme fluttering amplitudes occurred at Strouhal numbers between 5.41 × 10⁻⁴ and 7.15 × 10⁻⁴. Drag force amplitude increased with mean drag force coefficient for empty codends, but for codends with catch, it increased with smaller mesh sizes, twine diameters, and higher flow velocities. Positional amplitude increased with lower waist-to-end hanging ratio. BP neural network predictions matched experimental results with over 90% accuracy, demonstrating its effectiveness in predicting codend oscillations under varying parameters.