A Proposal for a Pseudo Strain Sensor for Hull Monitoring Using Artificial Neural Network

Abstract

　Full-scale hull monitoring is a direct way to ensure the structural integrity of the ship during actual operation. Typically, multiple strain sensors are installed on structural members of several cross-sections. The greater the number of sensors, the more accurate the monitoring is, but the more costly to install and maintain them. This paper proposes a pseudo-strain sensor for hull monitoring using an artificial neural network (ANN). It means that the ANN learns the relationships of structural responses among the multiple sensors for a certain period, and after the learning is completed, the representative sensors predict the structural responses at the remaining sensors. First, the proposed method is applied to a full-scale measurement on an 8,600 TEU container ship to investigate the teacher data needed to predict the response of the ship's bottom from sensors on deck. Next, the predicted results by the ANN are compared to a data assimilation method to predict the hull-girder response using the Kalman filter (RKF), and the pros and cons of the ANN and RKF are discussed. Finally, the learned ANN is used to estimate fatigue damages of the ship's bottom members based on Miner's rule. The estimated results show a good agreement with the measured ones and that the ANN is applicable to the development of pseudostrain sensors for hull monitoring.