分析了利用支持向量机进行模式分类的原理 ,指出最优分类面上的样本相对于两类误判而言是等概率的而非等风险的。针对机械故障诊断中两类错误分类具有不同损失的特点 ,说明直接将支持向量机应用于该领域存在不足。在此基础上 ,提出了诊断可信度函数 ,并在特征空间中 ,对最优分类面进行重新设计 ,使其符合等损失的要求。文中算例说明了该方法在实际应用中的可行性。 The principle of pattern classification using support vector machines is analyzed. It is pointed out that the samples on the optimal classification surface are equi-probabilities rather than equal risks relative to the two kinds of miscarriage of justice. Aiming at the characteristics that the two kinds of error classification have different losses in the mechanical fault diagnosis, it shows that there are some shortcomings in applying SVM to this field directly. On this basis, the diagnostic credibility function is proposed, and in the feature space, the optimal classification surface is redesigned to meet the requirements of loss. An example is given to illustrate the feasibility of this method in practical application.