| Wave soldering has been a well-established soldering technique for more than twenty years but in recent times, it has been challenged by alternatives such as selective soldering. Till now, wave soldering systems have managed to cut through the competition and hold their position but it may be difficult in the future because the alternatives are claiming faster process and low defect products. Even though the wave soldering systems are modernized today, there are many factors that can affect the yield. Valuable time is wasted in deciding upon the parameters and even after that low defects are not assured. This can cost money and time to the manufacturer which is simply unacceptable in the current competitive environment. Predictive models have been built in the past which help in selecting the best values of the input parameters but such models neglected some factors like board orientation, nozzle angle, wave speed, etc. Even if a model is built, a layman or the system operator may not have the knowledge to understand a complicated mathematical predictive model. There needs to be a medium through which results of such models are conveyed to the operators. The objectives of this research are to first develop a design of experiment (DOE) with various process affecting factors (e.g., wave type, board orientation, and others) as input parameters and the outputs identified as different defects found on the board (e.g., insufficient holefills, shorts, and others). Based on the DOE results, a predictive model for each type of defect will be established. Finally a Windows application to make the predictive models accessible to any operator will be developed.; A fractional factorial design was developed for the experiment. Predictive models were developed based on the results of the experiment. Predictive models were developed using SAS software and neural networks. The models are confined by the parameters used in the experiment but are capable of predicting the defects for boards having similar components although with different quantity. To make the predictive models accessible to the operators, a Windows application is developed using C# programming language and .NET framework, to directly display the output in the form of defects for all combinations of the parameters. Buttons are also made available to highlight the best combination of the parameters for each type of defect. Although developed for a particular electronics manufacturing services provider, this methodology can be applied not only to other EMS providers, but essentially to any enterprise that wants to predict process output based upon process inputs. |
