汽轮发电机组故障诊断的自学习

由于对若干故障的机理缺乏认识，诊断系统不会覆盖所有可能发生的故障，对汽轮发电机组故障诊断全面分析，研究了振动信号的分形特征，在实验的基础上，提出了一种诊断系统获取新故障样本的方法。

Study on Self-learning for Vibration Fault Diagnosis System of Turbogenerator Unit

For a given diagnosis system, its diagnosis ability lies on the knowledge capacity. It is incapable to detect a new fault condition if no priori knowledge is given. We divide the conventional networks into several sub-nets, which is responsible for one specific fault class. The vibration series has obvious fractal feature. It can reflect the essential characteristics of new fault. When the new fault is taken on, a new sub-net is increased and trained with the sample. If other samples are identified as this new class according to proximity, it has been verified experimentally these fractal dimensions of one class are distributed approximately around a definite value that can represents the dimension of the standard sample for the novel fault. Based on non-linear theorem, the approach of identifying new fault and self-learning for diagnosing is put forward.