Regresar

Structural Health Monitoring of Offshore Jacket Platforms via Transformers

Abstract:

The goal of this project is to monitor the structural health of jacket-type platforms for offshore wind turbines. The methodology is based on vibration-response-only accelerometer measurement and a transformer-based framework for multivariate time series. The original transformers paper proposed an architecture applied to a natural language processing task, meanwhile later works approached the use of transformers for forecasting, missing value imputation, and classification of time series. In general, the transformers based on attention mechanisms demonstrate being superior in terms of quality and performance on many sequential tasks in comparison to other architectures. Similar results are expected with time series data. Thus, this work proposes to use transformers for the classification of different structural types of damage in jacket-type wind turbines. The methodology follows the next steps: (i) accelerometer data is acquired, (ii) data is cleaned and wrangled into time series, (iii) a transformer-based framework classifies different damage scenarios. In a down-scaled experimental laboratory structure, the method is validated. The results demonstrate the feasibility of the proposed methodology.