Retroanálise da propagação decorrente da ruptura da barragem do fundão com diferentes modelos numéricos e hipóteses de simulação

Abstract

Mining is an important economic activity in Brazil, which aggregates value into the National Gross Domestic Product. Ore processing generates tailings that are disposed, most often in reservoirs, which usually are considered the cheapest structures for tailings dewatering. The tailings dams have several construction methods, among them theupstream method, which uses part of the tailings to build the dam. This was the case of Fundão dam, a tailings dam owned and operated by Samarco. On November 5, 2015, the Fundão Dam collapsed. Approximately 32 Mm³ of tailings conveyed to the downstreamvalley, causing deaths, destruction of Bento Rodrigues district, aside to affecting the water bodies and other urban areas downstream to the mouth of the River Doce. The hydraulic simulation of this event was done with the incorporation of the characteristics of the tailings deposited in Fundão, totaling 7 (seven) scenarios ranging from aqueous and hyperconcentrated fluids, uni and bidimensional hydraulic modeling, and HEC-RAS and FLO-2D. When Using HEC-RAS, hyperconcentrated fluid propagation characteristics were set up by adjusting Manning and expansion/contraction coefficients according to the Pseudo-Manning methodology (VENTURA, 2011). Using FLO-2D, simulations wereperformed of aqueous fluids and non-Newtonian fluids, adding the rheological curves of the material released from the Fundão Dam. These rheological curves were obtained through tests performed with rheometers and slump tests. The changes in velocities, maximum depths reached and wave arrival time, which are important parameters in the planning of emergency actions and evacuation, were analyzed for the different assumptions used in the simulations. In this way, the different simulations were evaluated through performance factors, weighting the intrinsic computational effort to simulations with a greater degree of complexity, the need for input data and the outputs obtained with these models.