An Innovative Long-Path Off-Axis DOAS System: High Light Throughput Design and Dedicated Spectral Processing Algorithm for Cl2 Monitoring

This paper develops a novel long-path off-axis DOAS (LP-OA-DOAS) system integrated with an automatic spectral processing and concentration inversion algorithm. An asymmetric off-axis parabolic configuration is proposed to avoid central occlusion of coaxial structures. 1:1 comparative ray-tracing models are established for coaxial and off-axis setups to evaluate their optical performance. Ray-tracing simulations suggest that the off-axis layout offers potential advantages over the coaxial structure in terms of luminous flux transmission efficiency and alignment tolerance. On this basis, an LP-OA-DOAS system is developed. For spectral processing, a physics-informed neural network-based algorithm, termed DOAS-Cl 2 -PINN (Differential Optical Absorption Spectroscopy-based Physics-Informed Neural Network for Cl 2 inversion), is proposed to extract broad-band Cl 2 absorption features from spectra affected by environmental noise. CBAM (Convolutional Block Attention Module) is employed to adaptively enhance the representation of informative spectral features, while physical prior constraints based on the Beer-Lambert law are incorporated to improve the robustness of concentration inversion, physical consistency and enhance model interpretability. Controlled experiments were conducted using a gas cell with an effective Cl 2 absorption length of 133.86 mm positioned along a 200 m geometric open optical path. The proposed method accurately retrieved the gas-cell Cl 2 concentration, achieving an RMSE of 17.71 ppm and an R 2 of 0.99. Compared with the conventional least squares (LS) method used in DOAS analysis, the proposed method improved retrieval accuracy by 17.64% and consistently outperformed baseline models, including SVR, MLP, and 1D-CNN. Robustness analysis under varying environmental conditions further demonstrated stable retrieval performance under spectral disturbances and environmental variations. The calculated method detection limit of the LP-OA-DOAS system was 7.387 ppm·m, and measurement variations remained within ±5%. This detection limit corresponds to an equivalent minimum detectable concentration of 36.93 ppb, assuming a uniform Cl 2 distribution over the entire 200 m path. A six-month deployment in a liquid chlorine filling workshop demonstrated stable, fully automatic monitoring without false alarms under leakage-free operating conditions. Taken together, the controlled experimental results and field deployment support the system’s Cl 2 detection and concentration retrieval capabilities, as well as its operational stability. Its detection performance in actual industrial Cl 2 leakage scenarios warrants further validation under appropriate conditions.