Optimized Design of Graphene Waveguide Composite Structure Pressure Sensor Based on Machine Learning

Aiming for high sensitivity and satisfying transmission loss, intelligent optimization methods involving machine learning are worthwhile to apply for optical pressure sensor development since machine learning is versatile enough to correlate the complex and nonlinear relationship between device performance and sensor structural parameters. This article proposes a three-stage framework to conduct the optimal design of an optical pressure sensor facing multiple optimization objectives. First, a dataset consisting of 1600 samples with combinations of varying waveguide lengths and core widths is used to train and establish a prediction model, which also acts as the objective function for subsequent optimization. Second, a genetic algorithm is applied to find out the Pareto front within the sample space according to optimization objectives. Third, by inverse design based on the Pareto-optimal set, the multiobjective optimization for a sensor with optimal structural parameters is achieved. Experimental studies show that the optimized design of the graphene waveguide composite structure pressure sensor based on machine learning gives a sensitivity increase of 63.7% and a transmission loss reduction of 18.5% compared to similar sensors designed in previous work. This effectively verifies the feasibility of the proposed optimization design method in this article. Without a large number of sensor fabrications and trial-and-error experiments, the reverse design of sensors with various performance requirements could be achieved automatically, which provides a new approach to improve the performance of sensors in the future.