Optimising reticulation network design using heuristic graph traversal
David Coulton, Laurence Wall
Key words: paste; hydraulics; backfill
Underground reticulation networks for paste backfill systems must hydraulically connect a single paste plant to hundreds or even thousands of stopes. These networks are composed of boreholes linking underground drives and intra-level pipes. The positioning of boreholes and routing of intra-level pipes is a fundamental design decision that directly influences both the operability and cost of the backfill system. The hydraulic route from plant to stope constrains the paste to a feasible rheology window. Rheology is principally controlled by the CPB solids concentration, which in turn affects the geotechnical properties of the cured paste. Suboptimal route design can therefore result in (1) a narrow or even non- existent operating window, making consistent delivery difficult, or (2) reduced solids concentration, which increases binder demand and raises the unit cost of the backfill. In a study conducted by Retic Systems for an underground operation, an algorithmic design approach was developed to optimise reticulation layout and achieved a 13% reduction in binder usage. The process involved converting spatial mine design data into a graph representation and applying a heuristic graph traversal algorithm iteratively to construct routes for each stope. Graph edge costs were defined as a combination of CAPEX (pipe procurement and installation) and OPEX (binder consumption). The traversal algorithm was adapted from A* search: during each frontier update, route hydraulics were evaluated to (1) determine the required pipe class and (2) ensure that line pressures remained within minimum and maximum tolerance limits. The iterative nature of the approach allowed optimization of shared infrastructure between stopes, with core reticulation sections re-evaluated as new connections were added. This enabled minimization of life of mine costs while maintaining hydraulic feasibility across the network.