Global flowsheet optimization for reductive dimethoxymethane production using data-driven thermodynamic models

Abstract

The absence of accurate thermodynamic models for reductive dimethoxymethane (DMM) synthesis has limited the design of corresponding processes to approximate calculations only. To enable a more reliable process design, we measure liquid equilibrium densities and fit parameters for the PCP-SAFT equation of state (EOS). This EOS is highly accurate for systems at high pressures and therefore suitable for the high pressure reactor and the flash unit for gas recycling. As the resulting flowsheet optimization problem is nonconvex, we use our deterministic global solver MAiNGO to solve the problem. To improve computational tractability, we approximate process models that require the PCP-SAFT EOS with artificial neural networks and Gaussian processes. Finally, the so-called reduced-space problem formulation and a hybrid of the McCormick and the auxiliary variable method enable convergence within 5.8 CPUh. At the optimal operating conditions, an exergy efficiency of 91.9% is achieved for a reactor pressure of 120 bar.