Riassunto analitico
This thesis presents a comprehensive framework for the generation, homogenization, and predictive modelling of custom-defined Triply Periodic Minimal Surface (TPMS)-based unit cells. These are valued for their exceptional mechanical properties and lightweight nature. These characteristics make TPMS structures optimal for their utilization in aerospace, automotive, and biomedical engineering applications. The objective of this work is to advance the design and optimization of TPMS structures through a comprehensive and systematic approach.
The framework begins with the generation of four fundamental TPMS cells in the nTop software environment - namely Primitive, Gyroid, Diamond, and I-WP - and their combination using specific percentage weights and thickening values starting from their 3D base surface. Additionally, custom-defined hybrid cells are created not only by combining these fundamental TPMS cells but also by employing goniometric functions, which generate unique unit cell geometries. The goniometric approach involves mathematical functions to define the surface geometry of TPMS structures, thereby affording precise control over cell parameters and enabling the creation of complex and optimized designs.
Subsequently, the unit cells are transformed into finite element (FE) models through a detailed meshing process. The FE models undergo a homogenization procedure, the objective of which is to extract their effective elastic mechanical properties. This process entails the conversion of the continuous TPMS geometry into a discrete model that can be analysed to outline the key mechanical properties. A variety of homogenization tools are employed and compared in order to ensure the accuracy and reliability of the obtained properties.
To streamline and enhance the efficiency of the homogenization process, a Python script based on a Design of Experiment (DOE) scheme has been developed for the purpose of automating the procedure. The automation process generates a comprehensive dataset, which is a crucial input for training Machine Learning (ML) models. The mechanical properties obtained from the homogenization process are systematically compiled into a Comma-Separated Values (CSV) summary table, thereby facilitating efficient data handling and analysis.
An extensive data collection phase is conducted to enable the training of an Artificial Neural Network (ANN), which is designed to predict the mechanical properties of the custom-defined unit cells. The trained ANN is subjected to rigorous validation and testing, which demonstrate its high predictive accuracy and generalization capabilities. This predictive modelling approach significantly enhances the design process, enabling rapid and reliable projections of the mechanical behaviour of novel TPMS-based structures.
This research makes a substantial contribution to the advancement of lightweight, high-performance materials, providing a robust and efficient framework for the design and optimization of TPMS structures. The integration of goniometric functions in the generation of unit cells, coupled with automation and predictive modelling techniques, not only enhances the efficiency of TPMS design but also opens new avenues for their application in various engineering fields. The outcomes of this work pave the way for future innovations in the development and application of TPMS structures, reinforcing their potential in creating advanced materials with superior mechanical properties.
|
