关键词: 贝叶斯网, 工艺质量建模, 多品种小批量, 机理知识, 机器学习, 印刷线路板

Abstract: To solve the problem of high-dimension and sparse data faced by process quality modeling in multi-type & small-batch manufacturing mode, a modeling method based on Bayesian Network (BN) was proposed. The BN structure was constructed on the basis of manufacturing process mechanism knowledge and elimination method was used to ensure integrity and simplicity of the model structure. Then a set of balanced data was achieved from experiments organized by Design of Experiment (DOE), and Conditional Probability Table (CPT) of BN-model was obtained by machine learning with this data set. The CPT was updated in application with the data obtained from production line to model the effect of uncertain factors during process. This modeling method was testified in Printed Circuit Board (PCB) micro-hole drilling process and compared to other methods. The result verified that this modeling method was applicable in multi-type & small-batch manufacturing mode with higher modeling precision.

Key words: Bayesian networks, process quality modeling, multi-type & small-batch, mechanics knowledge, machine learning, printed circuit board