Model-based force and state estimation in experimental ice-induced vibrations by means of Kalman filtering

摘要

Bottom-founded structures deployed in ice-choked waters may experience severe ice-induced vibrations. A direct monitoring of the level-ice forces requires installation and use of load panels. This is often cumbersome and costly. Indirect measurements interpreted by inverse techniques are therefore favourable since sensors for response measurements are easier to install, less expensive and provide information as to the structural motion. In this paper, the level-ice forces exerted on a scale model of a compliant bottom-founded structure are identified from non-collocated strain and acceleration measurements by means of a joint input-state estimation algorithm. The algorithm allows for uncertainty in the model equations, can be applied to full-scale structures and reconstructs forces without any prior assumptions on their dynamic evolution. The identification is performed employing two different finite element models. One is entirely based on the blueprints of the structure. The other is tuned to reproduce the measured first natural frequency more accurately. Results are presented for two different excitation scenarios characterized by the ice failure process and ice velocity. These scenarios are known as the intermittent crushing and the continuous brittle crushing regimes. The accuracy of the identified forces is assessed by comparing them with those obtained by a frequency domain deconvolution, on the basis of experimentally obtained frequency response functions. The results show successful identification of the level-ice forces for both the intermittent and continuous brittle crushing regimes even when significant modelling errors are present. The response (displacements) identified in conjunction with the forces is also compared to those measured during the experiment. Here the estimated response is found to be sensitive to the modelling errors in the blueprint model. Simple tuning of the model, however, enables high accuracy response estimation.