DeepXtorm combines CFD and machine learning (ML) as a modeling platform (CFD-ML) developed from physical model data and CFD simulations over of 160,000 urban drainage clarifier configurations, loadings, hydrodynamics and particulate matter (PM) granulometry. A novel augmentation of CFD with ML models is developed and trained to create surrogate clarification models. For a clarifier, this CFD-ML platform facilitates (1) analysis, (2) design optimization, and (3) optimization of clarifier retrofits to minimize cost for a required level of clarification. Results with CFD-ML benchmarking indicate that: (a) historical models based on residence time (RT) are not accurate or generalizable for clarifier PM separation, (b) RT models are agnostic to geometrics, hydrodynamics and PM granulometry; and do not reproduce PM separation, (c) trained ML models provide high predictive capability (± 15%) for PM separation. Dynamic similitude analysis indicates clarification is primarily a function of the Hazen number and clarifier horizontal to vertical aspect ratio. With common presumptive guidance of 80% for PM separation, a Pareto frontier analysis with the CFD-ML model generates significant economic benefit for planning/design/retrofits. CFD-ML demonstrate enlarging clarifier dimensions (increasing RT) to address impaired behavior results in exponential cost increases, irrespective of infrastructure adjacency conflicts. CFD is facilitated by high performance computing and coupled in DeepXtorm with ML algorithms to optimize clarifier geometrics (or retrofits) to achieve a required performance. Results, benchmarked with monitoring and cost data for a full-scale operational clarifier, demonstrate that optimization provides cost reduction of at least 10X compared to presumptive guidance based on RT requirements.
