针对在数字电路故障诊断过程中存在的样本不平衡度严重的问题，采用层次式支持向量机实现对其故障诊断，通过考虑各类样本的数据量来构造以支持向量为叶节点的树，该方法可有效地解决样本不平衡所带来的问题，同时能够减少计算SVM分类器的个数，提高了训练和诊断速度及准确率。针对故障样本集不可能覆盖所有故障状态而出现的未知故障状态的问题，提出利用分类概率作为标准来判断样本是否为未知故障，实现了对未知故障的判定和对故障分类器的扩展。实验验证表明，本文提出的方法能够解决此类问题，取得较好的效果。

Considering the problems as unbalance training data and fault classifier extension in fault diagnosis of digital circuit, a new method was proposed and verified for the fault diagnosis. Firstly, exhaustive method was used to stimulus the circuits, and the samples of normal state and fault states were obtained after preprocessing. The hierarchical Support Vector Machine (SVM) using the length of samples as criterion was adopted to diagnose the faults of digital circuits. It can alleviate the problems brought by the unbalance training data，and the training speed and accuracy were improved as well. Since the fault samples can not cover all the faults of the circuit due to the diversity of faults of one circuit,we proposed to use the classification probability as the criterion to judge whether the sample was an unknown fault sample. The results of the experiment shows that the method proposed can solve the problems in digital circuit fault diagnosis.