Description
A Master of Science thesis in Mechanical Engineering by Mohamed Shendy entitled, “Machine Learning Assisted Approach to Design Latti”, submitted in March 2023. Thesis advisor is Dr. Maen Alkhader and thesis co-advisor is Dr. Bassam Abu-Nabah. Soft copy is available (Thesis, Completion Certificate, Approval Signatures, and AUS Archives Consent Form).
Abstract
Lattice-based metamaterials belong to the phononic crystals class of materials which are known for their ability to interact with, direct, and block elastic waves. These properties made lattice-based metamaterials appealing in wave guiding, noise filtering, and vibration isolation applications. However, capitalizing on the full potential of lattice-based materials in isolation and filtering applications has been hindered by the lack of systematic and efficient design methodologies capable of producing a lattice with pre-set band gap characteristics. Existing design methodologies utilize timeconsuming iterative computational schemes and often move towards geometrically complex lattices whose fabrication requires expensive additive manufacturing techniques. This work proposes an artificial intelligent-assisted design methodology that integrates sinusoidal perturbations and the easy-to-fabricate double-wall hexagonal lattice. In the proposed approach, sinusoidal perturbations with different frequencies and amplitudes are superposed on the double-wall hexagonal lattice to increase the number and bandwidth of its band gaps. Finite element analysis is used to determine the band gaps in the perturbed lattices. By using five perturbation frequencies, five amplitudes, and six lattice porosities, the perturbed lattices delivered a band gap at each frequency in the range of 0 to 1000kHz. Machine learning, namely deep neural networks, is used to model the relationships among the perturbation parameters, lattice porosity, and the corresponding band gap characteristics. Three parallel neural network models are developed. These predict the maximum number of band gaps and the width and centroid of the band gap with maximum bandwidth. Results showed that the developed neural network models had an average accuracy of 90%. The developed neural network models constitute the core of the proposed design methodology. They are used to determine the coarse design parameters (i.e., porosity and perturbation parameters) required to realize prescribed band gap characteristics. The coarse design parameters are subsequently refined using finite element analysis. This approach accelerates the design process and eliminates the need for time-expensive iterative processes. A case study is presented to demonstrate the efficiency and practicality of the proposed design process.