中文摘要：

针对装配过程尺寸偏差的小样本检测条件，提出了基于条件独立性检验的结构学习算法，结合柔性装配偏差关系模型，推导了贝叶斯网络子节点的先验条件概率，将小数据集与先验概率融合并获得贝叶斯网络参数，实现了装配偏差影响因素的贝叶斯网络建模，并用于某车型侧围装配过程的偏差源诊断．结果表明，所提出的偏差源诊断方法具有较高的准确性．

英文摘要：

Based on the small samples collected in the assembly process, a new approach based on Bayesian networks was proposed for the variation-source diagnsosis. The conditional independence testing algorithm was proposed to obtain the structure of Bayesian networks. After the prior conditonal probabilities are cal- culated based on the mapping of the variation simulation model, posterior conditonal probabilities are updated by incorporating the small sample data. The results of the body side case show the method is effective and acurate for fixture fault diangosis.