中文摘要：

随着间歇过程越来越受重视，其过程监控和故障诊断技术也成为研究热点。以核Fisher判别分析为基础，提出了基于核Fisher的正常工况与故障包络面模型，给出了基于该模型的在线故障诊断流程。此方法利用了Fisher判别分析对类别的划分特点，分别针对正常工况数据和各故障类型数据建立包络曲面模型。与多向Fisher判别分析相比，该方法按批次方向将数据展开，能够解决生产周期不一致问题，在线故障诊断时也不需要预报完整的生产轨迹，并且加入核函数来处理复杂的非线性。最后在青霉素发酵过程的仿真平台上对该方法进行测试，与改进多向Fisher判别分析方法进行对比，该方法获得了满意的诊断效果：能够及早诊断出故障的发生，并在有效识别已有故障的同时还具有对新故障的自学习能力。

英文摘要：

With increasing importance of batch processes, its monitoring and fault diagnosis technology has become a research focus. An envelope surface model of normal and fault operation conditions based on kernel Fisher analysis technique was proposed, and also an online fault diagnosis program was presented in this paper. With kernel Fisher discriminant analysis＇s characteristics for classification, envelope surface models for normal data and each fault data were built. Compared with multiway Fisher discriminant analysis （MFDA）, the proposed method unfolded the process data batch-wise, so it could deal with unequal batch length and did not require to predict the whole production trajectory when implementing online fault diagnosis. Also kernel function could deal with nonlinear problems. Finally, the proposed method was tested in the simulation platform of penicillin fermentation process. The simulation results comparing with the improved MFDA showed that the proposed method could diagnose the fault early, and also had self-learning ability for new fault.