Evidence Boundaries and Scale-Up for Bernoulli-Driven Water Separation: Lifecycle Evidence and Accountable Decisions

Abstract

Scaling a Bernoulli-driven separation mechanism from a laboratory component to a managed water system requires more than transferring a promising performance value. This review maps the evidence boundary from channel geometry and hydraulic measurements to manufacturing variation, maintenance, water-quality outcomes, and end-of-life decisions. AI can support that translation through surrogate modelling, domain-shift detection, and lifecycle prediction, but only when training data, assumptions, and use constraints remain attached to every output. The synthesis contrasts interpolation within a tested hydraulic regime with extrapolation across equipment size, feed composition, and operating environment. It recommends a digital evidence thread linking samples, sensor histories, model versions, uncertainty estimates, operator interventions, and downstream consequences. Accountability is assigned at each transition: model developers define validation limits, engineers approve scale-up criteria, and operators retain authority when evidence becomes discordant. The review does not infer plant readiness from mechanism-level studies or generate new empirical results. It instead provides a lifecycle governance framework for deciding which AI-supported conclusions are transferable, which remain provisional, and what additional evidence is required before scale-up.