Optimization of production planning in printed circuit board assembly

Abstract

Many complicated planning problems arise in Printed Circuit Board (PCB) assembly. This research focuses on two high-level planning problems, i.e., the Component Allocation Problem (CAP) and the Multi-Line Scheduling Problem (MLSP), both of which are important for improving production efficiency in PCB assembly. For a PCB job (batch) to be processed by an assembly line, the component allocation problem is investigated, which is to allocate the component placements required by the PCB to the placement machines in the line, so that the line cycle time is minimized. The problem is intertwined with the lower-level machine optimization problems (feeder arrangement and placement sequencing), which determine the process (placement) time of each machine. Considering the great computational complexity, a decomposed solution strategy is proposed. This strategy relies on a regression-based placement time estimator, which can estimate the placement time of each machine accurately without solving the machine optimization problems. Based on this estimator, a specific genetic algorithm is developed. Experimental tests show that the proposed genetic algorithm can solve the problem both effectively and efficiently. Compared with the existing software provided by the machine vendor, the line cycle time is reduced. For a set of PCB jobs to be produced by multiple assembly lines, the multi-line scheduling problem is investigated, which is to assign the PCB jobs to the lines and sequence the jobs in each line, so that the sum of weighted tardiness and weighted makespan is minimized. A mixed integer linear programming model for the problem is established. Line-dependent cycle times, different due dates of the jobs, sequence-dependent setup times, and precedence constraints are considered so that the model is realistic and applicable. Experimental tests show that exact solutions can not be obtained for realistic-sized problem instances. A specific genetic algorithm is developed for solving the problem. Due to the complexity of the problem, a new replacement strategy is proposed to improve the performance of the algorithm. Experimental tests show that the genetic algorithm can solve the problem both effectively and efficiently. A study of a real case is conducted and illustrates the applicability and usefulness of the method.