Machine Learning-Assisted Optimal Design of a Balloon-Like Single-Mode Fiber and Sagnac Interferometric Composite Pressure Sensor

Fiber-optic pressure sensors are widely used in precise or specialized sensing applications due to their high sensitivity and strong immunity to electromagnetic interference. Traditional optimization methods for fiber-optic sensors primarily rely on theoretical derivation together with trial-and-error through simulations, which are hardly competent to achieve synergistic optimization across multiple performance metrics. To address this issue, this article proposes a machine learning-based optimization design method for a balloon-like single-mode fiber and Sagnac interferometric composite pressure sensor. Support vector regression (SVR) is employed to establish predictive models to infer sensitivity and transmission loss of the optical sensor according to its fiber bending diameter and substrate elastic modulus. These models are then integrated with the nondominated sorting genetic algorithm II (NSGA-II) multiobjective optimization algorithm for Pareto-optimality search. Optimization results indicate that with a bending diameter of 4.0 mm and a substrate made of rubber that gives an elastic modulus of 0.1 GPa, the sensor achieves a sensitivity of 93.4 nm/N and a transmission loss of 13.2 dB, outperforming those equivalents that with substrates made of common materials. Furthermore, cascading the optimized sensing unit with a Sagnac reference sensor enables high-pressure-sensing sensitivity at -518.2 nm/N. This study demonstrates the effectiveness of machine learning in multiobjective collaborative optimization for fiber-optic pressure sensors.