Neural network open loop control system for wave soldering

Abstract

This paper describes the development of neural networks for the prediction of (1) printed circuit card surface temperature during a wave soldering process, and (2) the quality level of the circuit card assembly soldered connections. Using a combination of production data and design of experiment data, a set of hierarchically connected neural networks were developed and validated. These networks predict thermal behavior of a printed circuit card assembly at various points in the solder process based on process settings and circuit card design data. Then these predictions are used as inputs, together with other parameters, to estimate the quality of solder connections. The system can be used to decrease the number of solder connection defects, reduce set-up and preparation time between lots, and lead to consistent, repeatable process settings without trial production runs or operator tuning efforts. For the wave soldering process studied, this is especially important since the batch size is quite small, quality demands are stringent, and process settings are changed frequently.