Data-Driven Acoustics Modelling of Additively Manufactured Lattice Structures

ABSTRACT

With the rapid advancements in additive manufacturing technologies, there has been an increasing interest in the design and modelling of 3D-printed lattice structures for acoustic applications. Lattice structures can be generally classified into many different categories, three of which are strut lattices, plate lattices, and triply periodic minimal surface (TPMS) lattices. In this seminar, the relationships between structural geometry and sound absorption properties for these three categories of lattice structures will be discussed. The process of developing semi-empirical, data-driven mathematical models relating geometry to property will also be presented. These models were trained using experiment measurements of additively manufactured samples and boast high degrees of modelling accuracies, while being computationally friendly as compared to numerical simulation methods. We will also explore the utility of these mathematical models to design various heterogeneous lattice structures obtained by stacking or overlaying of different unit cells. This seminar is suitable for those who are interested in data-driven design of lattice structures and acoustic metamaterials for practical applications.

Flow Past a Bluff Body: High-fidelity Simulation and Intelligent Flow Control

ABSTRACT

Flow past a bluff body represents one of the most fundamental and widely studied phenomena in fluid mechanics, ubiquitous in nature and engineering applications. To resolve the intricate spatiotemporal evolution of flow structures and uncover the underlying physics, a high-fidelity numerical framework is presented that leverages the lattice Boltzmann method, integrating the immersed boundary method, large eddy simulation, a nonuniform mesh partitioning technique, together with the GPU parallel technique for ultrafast computations. Based on this solver, we demonstrate intelligent flow control strategies targeting drag reduction, vibration suppression, noise mitigation, and hydrodynamic stealth. A deep reinforcement learning (DRL) approach is implemented to design the closed-loop control system. For instance, in flow past a circular cylinder, windward-suction-leeward-blowing actuators dynamically regulate near-wall flow behavior, while velocity sensor arrays supply real-time feedback. After 1,000 training episodes, the DRL agent derives an optimal control policy that eliminates wake signatures and achieves hydrodynamic stealth. In summary, this talk delivers a high-fidelity, computationally efficient framework for bluff body flow analysis, providing scalable solutions with affordable computational time and hardware cost, enabling practical applications of intelligent closed-loop flow control in complex fluid systems.