Abstract Bending, typically done using a CNC press-brake machine, is one of the important forming processes involved in the manufacture of sheet metal parts. To manufacture any sheet metal part, an operator needs to determine the appropriate tools, their positions on the machine and the bend sequence to be followed. Process planning for sheet metal bending aims to minimize the manual effort by automatically determining a near-optimal bend sequence for a given part. A near-optimal bend sequence is the order of processing the bends in the blank such that it requires less number of tools, gives good gage positions and makes it easy for the operator to handle. Determining such a sequence is hard because it is a combinatorial optimization problem. Literature reveals the need for a good process planning software that combines an efficient collision detection algorithm capable of quickly determining the collision condition while folding each bend and an effective search algorithm that can reduce the search space and determine the near optimal bend sequence. When both collision detection and search algorithms are efficient, the process planning time for sheet metal parts can be reduced considerably. In the present work, the sequencing at the features level is followed to reduce the process planning effort. Relevant features are extracted first from STL format of the part and tool models. Next, the collision detection strategies are investigated considering bounding volume hierarchies involving oriented-bounding box (OBB) and axis-aligned bounding box (AABB) methods. A new hierarchy considering minimum oriented bounding box (MOB) is proposed in the present work. The collision detection is done through box–box and triangle–triangle intersection tests. It is demonstrated that though the MOB hierarchy is more efficient in minimizing the number of collision tests, the AABB hierarchy is superior in terms of computation time and is the ideal choice for collision detection in sheet metal bending. Further, a two-stage algorithm is described that allows for the quick identification of a nearoptimal bend sequence for a given part and set of tools. In the first stage, a Bend Feasibility Matrix is constructed to map the entire search space by taking a geometric approach to the problem. The matrix helps to quickly establish whether the part can be manufactured using the given set of tools. The second stage uses best first search (graph) algorithm to identify the bend sequence. During search, infeasible sequences are never evaluated and expensive collision tests are not done since the necessary computations are already done in the first stage. Performance of the proposed algorithm is compared with that of genetic algorithm and it is demonstrated that the best first search algorithm is better than genetic algorithm (GA) to solve the bend sequencing problem. The developed software is tested with 10 sheet metal parts taken from literature and industry case studies. These parts have bend numbers ranging from 4 to 14 and have different complexity. It is shown that the proposed approach gives near optimal sequence in less than a minute and has a potential for online processing.