Composite wind turbine blade covers are usually manufactured with the VARI technique, due to its low cost and high manufacturing rate for large composite parts. Whilst FE simulations are extensively used to assess the resin filling process of the manufacturing, it is necessary to use surrogate models developed with ML techniques to replace the time-consuming FE simulations to enable real-time control during manufacturing and to allow engineers to explore a vast design space efficiently in optimisation.
A FE approach is used to simulate the resin filling process in the VARI manufacturing of a wind turbine composite blade cover. Four ML methods, DT, RF, Support Vector Machine (SVM) and ANN, are developed with the simulated FE data, for fast prediction of the manufacturing quality of the blade cover. Optimal ML models are obtained by tuning the hyperparameters of the ML models to achieve the best fit.
Both ANN and SVM models seem to be the best ones for prediction of the manufacturing quality. The feature importance assessment indicates that the resin filling temperature is the most important parameter to influence the resin filling process and final manufacturing quality. Further investigations should focus on the effects of the temperature with considerations of temperature distribution and change in the cover.
ML models, once developed with an appropriate ML algorithm, can be used to replace time-consuming FE simulations to predict product manufacturing quality at any condition in a very short time to enable real-time control during manufacturing or optimisation in a broader design space.
Temperature looks like the most influential parameter to affect the cover manufacturing quality, and further investigations should focus on the effects of the temperature with considerations of the temperature distribution and change in the cover.
Implementation of the optimised manufacturing processes will significantly improve the blade cover quality thus reduce the scrap rate, contributing to sustainable development of the wind farm industry and the world economy.
Application of various ML methods for prediction of the manufacturing quality of a wind turbine blade cover is explored and the best optimal ML model is obtained through comparison of the modelling performance of the models. The focus for optimisation of the manufacturing process is identified. It is demonstrated that a surrogate model such as the ANN can be used to predict the manufacturing quality in a much shorter time, compared to the conventional FE simulation.
