A surface mount technology (SMT) pick-and-place machine is a key device used in printed circuit board (PCB) assembly. An SMT pick-and-place machine is usually expensive and is the bottleneck of the PCB assembly line. Therefore, improving the throughput of the machine is significant. The optimization problem of SMT pick-and-place machines is challenging because hundreds or thousands of electronic components, of different types and sizes, must be placed onto specific locations of the PCBs with appropriate nozzles equipped on gantry heads.

The objective of this research is to find optimal nozzle and feeder setups and component pick-and-place sequences to minimize the assembly cycle time per PCB for SMT collect-and-place machines, considering important manufacturing constraints. Because the optimization problem is machine specific, this research focuses on the collect-and-place machines, which have dual gantries, revolving heads, and unmovable feeder banks and placement tables.

The research begins by investigating a single-gantry's pick-and-place operations and proposing an overall mathematical model of the problem. In the single-gantry mode of placement operations, a dual-gantry machine can be treated as two single-gantry machines and each gantry places PCBs independently. Through several assumptions and generalizations, the single-gantry optimization problem is modeled as an integer linear program, which is an integration of a quadratic assignment problem (QAP) and a capacitated vehicle routing problem (CVRP). Because QAP and CVRP are well known NP-complete problems, the single-gantry optimization problem is at least NP-complete. This means the problem is computationally intensive. To be specific, the number of solutions could grow exponentially with the problem size. A typical problem has around 3 million decision variables given that 300 electronic components are to be placed. Therefore, heuristic and metaheuristic approaches are proposed to solve the problem. To solve the single-gantry optimization problem, four population based metaheuristic algorithms are developed based on special operators for the integer encoded solutions. Computational results are presented to compare the performances of these four algorithms.

Next, the dual-gantry join-mode placement operations are studied. Two gantries can alternately place electronic components on the same PCB in the join-mode placement. Compared to the single-gantry mode, the join-mode placement is more complicated because an allocation of the locations on the PCB to gantries should be solved. A nonlinear integer programming model is developed to describe the optimization problem. Multiple types of nozzles and the matchings between component types and nozzle types are captured in the model. For the nozzle setup and component allocation subproblems, different heuristic rules are proposed and compared in numerical experiments. To solve the whole dual-gantry optimization problem, a multi-phase planning (MPP) strategy and an adaptive clustering genetic algorithm (ACGA) are developed. The proposed approaches are compared to existing algorithms in literature using twenty industrial PCB samples. Experimental results show that the improved MPP method can efficiently yield good solutions within a minute, while the ACGA method produces less cycle time results than MPP but costs more computation time. Therefore, the improved MPP method is suitable to a low-volume high-mix production and ACGA is suitable to a high-volume low-mix production.