ANN-based prediction for a sustainable decision model on a combined sewer overflow screen: using a conceptual approach.

Combined sewer overflow (CSO) screens are critical components of sewer and drainage networks, separating sewer solids from overflow spills before they reach receiving waters. Selecting suitable and sustainable CSO screening devices, however, remains a complex task. This process has traditionally depended on conventional design calculations, technical guidance from screen manufacturers and precedents from past projects. Inappropriate screen selections have led to adverse effects on water quality and public health, due to insufficient screening capacity, the unpredictable behaviour of sewer solids of varying densities, low trapping efficiency, frequent screen blinding or high equipment failure rates, particularly at unmanned or remote sites. This paper presents a design methodology for screen selection and formulates an input-output relationship model. Using 50 screen project data, a framework has been proposed to construct a predictive model that integrates sustainability criteria, lessons learnt from historical applications and artificial neural network (ANN) techniques. A Levenberg-Marquardt-based ANN was developed and trained to identify optimal selection between 2 categories of screen solutions, encompassing 12 screen types - 3 within non-powered self-cleaning and 9 within the powered screen category. The framework aims to provide an initial proof-of-concept evidence with a supplementary decision-support tool, enabling design engineers to make intelligent, resilient and sustainable choices in screen application.