Design of MEMS thermal actuator by CNN and PSO

This paper reports a design scheme for MEMS actuator by integrating deep learning and intelligent optimization algorithm, which can quickly and accurately obtain structures with expected properties. Taking the U-shaped thermal actuator as an example, the shape of the actuator can be described as a one-dimensional array. The trained convolutional neural network (CNN) was used to quickly and accurately determine the output displacement of each candidate design in this study, and the effectiveness of this method was demonstrated in comparison with the commonly used computational tools such as finite element method (FEM). After sufficient training of 6000 datasets, the neural network obtained can be used to perform high-precision and high-speed computations. With the combination of particle swarm optimization (PSO) algorithm, it is possible to obtain the U-shaped thermal actuator with desired output displacement and the calculation time can be reduced to 1/5600 of the conventional FEM, with a maximum accuracy of up to 97.18 %.