In this paper, a feasible path-based branch and bound (B&B) algorithm is presented for solving mixed-integer nonlinear programming problems with highly nonconvex nature. The main advantage of this novel algorithm, comparing to the conventional branch and bound algorithms, is that when solving a nonlinear programming (NLP) subproblem at each node, our previously proposed hybrid steady-state and time-relaxation-based optimisation algorithm is employed. This approach allows circumventing complex initialisation procedure and overcoming the convergence difficulties of process simulations. During B&B, the solution from a parent node is used to initialize the NLP subproblems at the child nodes to improve the efficiency of this algorithm. The capability of the proposed algorithm is illustrated by solving a dividing wall column optimisation case for separation of a ternary mixture. The optimal design is obtained in 2, 712 CPU s with TAC 43,344 $ y(1.