Abstract

A significant problem encountered during the manufacturing process of thermoset composite structures is the distortion of their shape from their CAD-nominal geometry. Shape distortions can be attributed to the residual stresses which are imposed within the structure during its manufacture. The choice of a suitable material model to simulate process induced distortions is important in order to achieve a right first time approach in the design of new moulds. This work investigates the ability of the Cure Hardening Instantaneously Linear Elastic (CHILE) model and a linear viscoelastic material model to predict process induced distortions of an aerospace composite frame. The CHILE material model is widely used to simulate shape distortions of composite structures due to its simplicity, its fast calculation times and its enhanced accuracy relative to the use of pure elastic models. However, it cannot predict any stress relaxation during curing. The linear viscoelastic material model investigated employs an innovative constitutive update scheme for simulating anisotropic, thermo rheologically simple, viscoelastic solids. The implementation of a viscoelastic material model allows the simulation of all the time dependent factors which affect shape distortions like the cure time, heating and cooling rates and stress relaxation. The modelling of the viscoelastic behaviour of the resin is done with the use of a generalized Maxwell model. A novel methodology is applied for shifting the relaxation times of the composite based on its temperature and degree of cure. Both material models were coupled with a cure kinetics model and a chemical shrinkage model in order to capture the multi-physics phenomena that take place during the curing process. The constitutive equation of the material models is implemented using the UMAT subroutine of the ABAQUS FEA software. As a reference for assessing the accuracy of the developed simulation processes and boundary conditions, measurement data which come from the 3D scanning of the manufactured frame is used. It is shown that the viscoelastic model more accurately predicts the measured distortions due to its ability to account for stress relaxation. Further steps to increase the accuracy of the simulation processes are proposed.