Far-field identification of ice loads on ship structures by radial basis function neural network

Abstract

The ice load is an environmental load that dominates the ice-resistant design of polar ship structures, and field measurement of ice-induced strains is an important approach to obtain ice loads. Constrained by narrow spaces and watertight structures in the ship-ice interaction area, the monitoring area where strain sensors are mounted sometimes has to be arranged separately from the loading area, which makes the classical influence coefficient matrix method no longer applicable. The far-field identification method for ice loads, based on the radial basis function neural network (RBFNN) with a simple structure and fast convergence, provides an alternative way to solve this problem. The potential mapping relationship between ice loads and far-field ice-induced strains is effectively established based on a generalised RBFNN. The influence of noise on the identification results of the ice loads on the bow shoulder of RV Xue Long 2 is thoroughly studied with finite element numerical analysis and effectively eliminated by noise injection learning algorithm. Under weak strain response and low signal-to-noise ratio, the accurate results of the subsequent scale model test indicate the possibility of engineering application. According to the rationality analysis of the identification results based on full-scale data, the applicability of the RBFNN method to field measurements is verified.