A Two-stage Parameter Optimization Approach For Stencil Printing Processs

As the key process in printed circuit board assembly industry, surface mounting technology (SMT) manufacturing process generally consisted of three sub-processes: solder printing, component pasting, and reflow soldering. These processes involve a lot of control parameters. Moreover, there exist nonlinear interactive relations among the parameters. Hence, the control of process is extremely complicated. A lot of studies show that the welding problem is one of the prime causes for the cost increase and quality decrease in surface-mounting engineering. And the studies also indicated that about 60% of the welding defects are resulted from the stencil printing process. Aiming at improving the product quality in surface-mounting engineering, researchers have conducted a lot of researches on the stencil printing process.Due to its complexity, no precise mathematical models can be drawn to predict the quality of stencil printing. As a result, experiment design based methods and mathematical model fitting based methods are wildly used to predict and control the performance of stencil printing process.Among the previous studies, the experiment based methods are popularly used because they require little amount of data for optimization. Unfortunately, the optimized results achieved by this method are limited to the combinations of parameter levels in experiment. Neural network successfully overcomes this limitation and becomes another mainstream solution for the stencil printing process optimization problem. However, in order to guarantee an accurate result, neural network requires a huge amount of training data, which is hard to provide for some electronics manufacturing services factories. Meanwhile, the customers continuously require electronic products which are more reliable, light and less costly. This situation brings us some challenges: (1) Quickly and accurately fitting the relation model with small-scale data. (2) Guaranteeing the key quality index the ability to defend random variation. And (3) Determining a reasonable solution by compromising among the multiple quality objectives. Aiming at solving these problems, this paper focuses on the following contents and proposes a two-stage parameter optimization method based on neural network and response surface method.1. A deep study is made on the characteristics of stencil printing process optimization problem, its influence factors and quality indexes.2. Two-stage optimization is conducted to determine the optimal parameters. On the first phase, the rough fitting models are obtained with small-scale data using neural network. The parameter ranges are then reduced based on the models. On the second phase, response surface method is adopted to build the accurate models in the reduced parameter ranges. The optimal parameters is then determined based on the models.3. The concept of confidence level is introduced into the optimization model to guarantee that the optimized result could satisfy the control limit of key quality indexes.4. An interactive desirability approach is adopted. The optimal compromise solution is determined through an interactive process with the cooperated factory.Experiments are conducted to verify the proposed method. The results show that on condition that the key quality index fulfills requirements, the proposed two-stage optimization method improves the solder thickness variation from 52.23 to 45.37, and the language variable for printing quality evaluation from 2.7 to 2.4.