Determination of sea ice bending strength calculation parameters based on neural network and numerical study on ice-breaking with a cone
Here,to establish relations between physical parameters of cohesive element and bending strength of sea ice,a neural network-based regression model was proposed for determining calculation parameters of sea ice bending strength.Firstly,numerical simulation was performed for 3-point bending tests of sea ice based on finite element method and cohesive element method,and the effectiveness of methods was verified.Then,5 affecting parameters were selected,and Latin hypercube sampling algorithm was used to generate 427 samples.The corresponding sea ice bending strength was obtained with numerical simulation to construct the dataset of neural network.Furthermore,a multi-layer perceptron neural network was used to train all samples'prediction results,and obtain a regression model for predicting sea ice bending strength.Consequently,a numerical model of layer ice with bending strength similar to that of sea-ice tests was constructed.Considering effects of fluid buoyancy and drag on ice fragmentation,numerical simulation and analyses were performed for cone-layer ice interaction under different parameters.The results showed that standard deviation and peak values of layer ice longitudinal force exerted on cone increase with increase in collision velocity,cone waterline plane diameter and cone angle.
finite element methodcohesive element methodsea ice bending strengthneural networkstructure-ice layer interaction
万方数据
维普
