Parametric design-based multi-objective optimisation for high-pressure turbine disc

Mass and radial deformation are of great importance for a high-pressure turbine disc (HPTD). However, computational cost of computer-aided engineering (CAE) is too high to optimise the mutually restricted objectives. A parameterisation-based method is proposed to speed the optimisation process of HPTD: 'body-flange'-based parametric template is used to generate CAE samples; noise-based virtual samples are implemented to enlarge the training set, a cost-effective neural network is used as fitness function of non-dominated sorting genetic algorithm-II for optimisation whose initial population is the combination of different sample sets. Experiment results show that the proposed data-driven framework reduces the engineering difficulty of multi-objective optimisation, and it has high popularisation value for optimisation of other complex products.