Engineering data boosting for machine learning in reinforced polymer composites

• Main Author: Chong, Yong Lim
• Other Authors: Sunil Chandrakant Joshi
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Engineering; Machine learning; Composites
• Online Access: https://hdl.handle.net/10356/176802
• Institution: Nanyang Technological University

Reinforced Polymer Composites (RPC) are used in various sectors of engineering, such as aircrafts. Due to composites possesses a combination of properties that’s unmatched by traditional materials such as strength, deformation, and elasticity etc. Since there are so many variations of RPC, that would mean uniquely different process to create each one, thus making predictive modelling expensive. To reduce the cost of analysing data, newer ways to create predictive modelling should be used. One new technique that is becoming more prominent in Industry 4.0 is the use of machine learning. Machine learning helps to improve data analysing efficiency by giving an ample amount of data to a model or algorithm which will be “trained” to make prediction or classifications. Since it is expensive to get data just from experiments, data augmentation is used. A new data augmentation technique called COMPOSITES is used instead. It is a 10-step technique that uses mathematical formulas and the physics of the material to modify the dataset instead. Firstly, this report will be talking about the COMPOSITE method which will be used on the following datasets, the first dataset is from Paraloid EXL 2314 where the Flexural Strength is plotted against the weight percentage. It originally had 9 points where after 5 boosts, it had 357 points. The second dataset is the Pristine samples for which Absorbed Energy is plotted against the Impact Energy. It had 18 points and after 5 boosts, it had 1035 points. The third dataset is on the Tensile Dataset and the fourth dataset is the fatigue. The R squared value will be used to see if there is any improvement after the COMPOSITES method is used. There is also a new step added to removing outliers which has shown substantial improvement in the results. Secondly, the report will be presenting about the machine learning aspect. Linear regression will be used. Thirdly, the report will be discussing on the coding and future work that could be done.