Number Of Physical Constraints Research Articles

A physics-aware neural network for protein-ligand interactions with quantum chemical accuracy.

Quantifying intermolecular interactions with quantum chemistry (QC) is useful for many chemical problems, including understanding the nature of protein-ligand interactions. Unfortunately, QC computations on protein-ligand systems are too computationally expensive for most use cases. The flourishing field of machine-learned (ML) potentials is a promising solution, but it is limited by an inability to easily capture long range, non-local interactions. In this work we develop an atomic-pairwise neural network (AP-Net) specialized for modeling intermolecular interactions. This model benefits from a number of physical constraints, including a two-component equivariant message passing neural network architecture that predicts interaction energies via an intermediate prediction of monomer electron densities. The AP-Net model is trained on a comprehensive dataset composed of paired ligand and protein fragments. This model accurately predicts QC-quality interaction energies of protein-ligand systems at a computational cost reduced by orders of magnitude. Applications of the AP-Net model to molecular crystal structure prediction are explored, as well as limitations in modeling highly polarizable systems.

Surrogate Modelling and Optimization for Complex Liquefied Natural Gas Refrigeration Cycles

In this paper, surrogate modelling and optimization is investigated for use in large scale chemical processes. A novel CryoMan cascade liquefied natural gas (LNG) refrigeration cycle is selected as the case study which has been highlighted for potential use within industry. Given its high nonlinearity and dimensionality (31 input variables and 20 output variables with a number of physical constraints) and short time horizon for real-time decision-making, an time-efficient optimization scheme must be developed to maximize process performance. Therefore, various supervised and unsupervised learning techniques as well as surrogate model structures are explored in order to accurately capture the behaviour of this highly complex and interrelated process flowsheet. Optimal solutions identified by the surrogate models are validated against the rigorous process model. Following from the challenges encountered by artificial neural network based surrogate models, Gaussian processes were adopted and combined with partial least squares to simultaneously reduce dimensionality and capture the nonlinearity of the underlying chemical process. Through this innovative surrogate modelling strategy, overall time to optimize the LNG production process was reduced by orders of magnitude compared to the rigorous model based optimization methodology, hence significantly facilitating the industrial application of this new process.

Dynamic control of a two-stand cold mill

The design of a control system for a two-stand tandem cold rolling mill producing thin brass sheet is developed. The control objective is to minimize output thickness variations resulting from three process disturbances while observing a number of physical constraints. Three manipulated variables are available, and five process measurements can be made if desired. A linearized mathematical model, valid for fluctuations about the steady-state operating point, is derived. The model is multivariable, fully interacting, and contains a time delay representing the travel time of the metal between stands. Linear optimal control theory combined with digital simulation is used to derive the control. The control proposed has a novel structure. In addition to providing acceptable dynamic performance, the control is insensitive to steady-state errors in the process model, and may offer significant advantages over more conventional controls.