Physics-Driven AI Correction in Laser Absorption Sensing Quantification

TL;DR

This work tackles reliability gaps in laser absorption spectroscopy (LAS) quantification by introducing SPEC, a dual-mode framework that combines a conventional ML estimator with a Physics-driven Anomaly Detection (PAD) and a network-based Correction mode. The estimation mode provides an initial state, while PAD assesses physical plausibility through reconstruction and feasible errors, triggering a correction process when needed. The correction mode employs a Hybrid Surrogate Error Model and a Greedy Ensemble Search to iteratively minimize the estimated error, guided by online data and a configurable forward model; reconfigurability is achieved by updating the PAD configuration rather than retraining the estimator. Empirical results show SPEC delivers substantial reliability gains across inside and outside training distributions, as well as under measurement inconsistencies and varied spectral conditions, outperforming other neural-optimization approaches in both efficiency and accuracy. The framework thus offers a practical, physics-informed, and adaptable solution for robust LAS quantification in real-world deployments.

Abstract

Laser absorption spectroscopy (LAS) quantification is a popular tool used in measuring temperature and concentration of gases. It has low error tolerance, whereas current ML-based solutions cannot guarantee their measure reliability. In this work, we propose a new framework, SPEC, to address this issue. In addition to the conventional ML estimator-based estimation mode, SPEC also includes a Physics-driven Anomaly Detection module (PAD) to assess the error of the estimation. And a Correction mode is designed to correct the unreliable estimation. The correction mode is a network-based optimization algorithm, which uses the guidance of error to iteratively correct the estimation. A hybrid surrogate error model is proposed to estimate the error distribution, which contains an ensemble of networks to simulate reconstruction error, and true feasible error computation. A greedy ensemble search is proposed to find the optimal correction robustly and efficiently from the gradient guidance of surrogate model. The proposed SPEC is validated on the test scenarios which are outside the training distribution. The results show that SPEC can significantly improve the estimation quality, and the correction mode outperforms current network-based optimization algorithms. In addition, SPEC has the reconfigurability, which can be easily adapted to different quantification tasks via changing PAD without retraining the ML estimator.

Figures (11)

• Figure 1: The Workflow of SPEC: ML estimator $G$ gives first estimation, which is fed to physics-driven Anomaly Detection (PAD) Module $A$ to calculate actual error $e$. If $e > \epsilon$, correction mode is activated. The error estimated by hybrid surrogate error model is used to guide the optimization of candidate dataset $X_\textmd{C}$. The state $\hat{\mathbf{x}}^* \in X_\textmd{C}$ leading to minimal estimated error is feed to PAD for evaluation. The process is terminated till $e(\hat{\mathbf{x}}^*) \leq \epsilon$ or the iteration budget $T$ is exhausted.
• Figure 2: CO$_2$ absorption spectra at 600 K (top) and 2000 K (bottom), with a mole fraction of 0.07.
• Figure 3: The distribution of the ID test set.
• Figure 4: The performance of existed ML estimator (estimation mode) on the ID test set. The upper panel shows the comparison of the temperature estimation of ML estimator $G$ and the ground truth. The middle panel shows the comparison of the concentration estimation of ML estimator $G$ and the ground truth. The lower panel shows the overall error calculated via PAD.
• Figure 5: The distribution of the OoD test set.