While implementations of infeasible interior-point methods remain the state-of-the-art in nonlinear programming, there are serious limitations in their use within the framework of MINLP due to lack of warm-start and infeasibility detection capabilities. We present a primal-dual penalty approach that allows interior-point methods to have such capabilities, and remains flexible enough to accommodate changing bounds, additional constraints, and additional variables in the nonlinear subproblems.