Physics-Informed Machine Learning for Multiphase Emission Measurement

Abstract Physics-informed neural networks (PINNs) have emerged as a promising tool for solving complex partial differential equations (PDEs) prevalent in engineering disciplines. Their effectiveness lies in requiring minimal training data and computational resources compared to traditional computational fluid dynamics (CFD) methods (Raissi et al., 2019). This is particularly advantageous in scenarios like solving the three-dimensional advection and diffusion equation of multiphase emission scenarios, such as the common mix of emission gases, humidity, and pollutant particles. PINNs offer a compelling alternative to established models such as the Gaussian Plume Model in a simplified gas-only emission measurement and monitoring applications (Abdelhak et al., 2021). This study seeks to assess the efficiency and accuracy of PINN-powered numerical solutions for multiphase emission measurement, in comparison to the traditional CFD approaches. Through a thorough comparison of these two numerical techniques, new insights into the capabilities of PINNs in multiphase emission measurement are anticipated. Traditional methods for simulating emission dispersion use computationally demanding CFD simulations that require advanced computing resources (Zhang et al., 2018). Alternatively, PINNs present a less resource-intensive option, efficiently managing the complexities of advection and diffusion equations of multiphase emission mix with reduced computational load (Sun et al., 2020). By leveraging the inherent advantages of PINNs, such as the ability to generalize from limited data and low computational overhead, significant improvements in both efficiency and accuracy in emission measurement tasks are anticipated. Furthermore, the elimination of the need for extensive training data and the reduced computational requirements positions PINNs as an attractive option for real-time applications, where timely and accurate results are paramount (Karniadakis et al., 2021). The study offers a controlled environment for conducting comparative studies between PINN-based solutions and traditional CFD methods. Meticulous simulations and analysis are employed to quantify any performance differences between the two approaches. Additionally, the scalability of PINNs and their applicability to real-world scenarios beyond laboratory settings are explored. In summary, this research advances the field of multiphase emission measurement by evaluating the potential of PINNs as a viable alternative to conventional CFD techniques. Demonstrating their efficiency and accuracy in solving complex PDEs, particularly in dispersion modeling of multiphase emission mixes, the study seeks to catalyze the adoption of PINNs in various engineering applications. The findings aim to inform industry stakeholders and researchers about the benefits of integrating PINN-based solutions into their workflows, potentially revolutionizing emission monitoring practices. The study highlights the advantages of PINNs over CFD in multiphase emission measurement modeling, presenting a new approach to solving complex dispersion equations without relying on traditional simplifications that limit accuracy and applicability.