Addressing Uncertainty on Machine Learning Models for Long-Period Fiber Grating Signal Conditioning Using Monte Carlo Method

The massive adoption of machine learning (ML) and artificial intelligence models in the field of instrumentation and measurement has raised several doubts concerning the validity of their response and the methodology for estimating their errors. In this study, we revisit ML models that were used to interrogate long-period fiber grating (LPFG) sensors. We used these models to present a comprehensive analysis of the uncertainty propagation through the ML-based optical fiber sensor signal conditioning. The uncertainty propagation was studied using the Monte Carlo method. The results showed that the proposed models were capable to damp some optoelectronic noises, do not induce systematic errors under noise, and thus, the noise-damping effect of the ML models does not impact the interrogator’s resolution. Moreover, we hope that this work serves as a methodological framework for the evaluation of the uncertainty of ML-based optical sensor interrogators.