A controller, process, and glass manufacturing apparatus can be configured to minimize defects. Embodiments can be adapted to control injection of nitrogen and/or argon and a mixture of nitrogen and hydrogen or nitrogen, hydrogen, and argon during glass float manufacturing to facilitate a pre-selected hydrogen concentration within a tin bath furnace while also minimizing glass surface defects that can be caused from tin condensation and tin bath impurity concentrations. Empirical use data can also be collected and provided to a pre-defined machine learning element of a host device to update a pre-defined control scheme of a controller for adapting the operational condition set points or other target values to account for furnace operation history and performance.