Abstract
The application of fibre-reinforced thermoset material systems has been established in the aerospace industry, e.g. primary structure on commercial aircrafts. However, there is an increasing interest in thermoplastic-based material systems due to their potential for fast forming, weldability, their inherently superior fatigue performance, and excellent fire/smoke/toxicity (FST) properties. Current repair techniques for thermoset panels are adhesive bonding and mechanical fastening. However, these techniques are limited when applied to thermoplastic composites as mechanical fastening leads to stress concentrations and localized delamination which is worse for thermoplastic composites. In this paper, the optimisation of dissimilar material was carried out using a hybrid genetic algorithm - artificial neural network (GA-ANN) model. Due to the complexity of the ultrasonic welding (USW) process, Bayesian optimisation is adapted to determine the most suitable ANN architecture to develop a robust model. The predictive model is developed to map the relationship between welding energy, vibration amplitude, and welding force to the corresponding Lap Shear Strength (LSS). The model was trained on 27 experiments using the leave-one-out cross-validation method to measure the model's ability to generalise. To evaluate the optimised joint performance, The bonded joints were tested to determine the tensile load carrying capability, and their failure modes were analyzed with the primary aim to develop an efficient repair joining methodology.
Original language | English |
---|---|
Publication status | Published - 2023 |
Event | 23rd International Conference on Composite Materials, ICCM 2023 - Belfast, United Kingdom Duration: 30 Jul 2023 → 4 Aug 2023 |
Conference
Conference | 23rd International Conference on Composite Materials, ICCM 2023 |
---|---|
Country/Territory | United Kingdom |
City | Belfast |
Period | 30/07/23 → 4/08/23 |
Keywords
- Dissimilar materials
- Machine learning
- Ultrasonic welding