Failure modes occurrence measurement, one of the necessary steps of FMEA, has a significant impact on the accuracy of it. However, the existing occurrence measurement methods have strong dependence of subjectivity and label data. To solve the above problems, an unsupervised learning based failure modes occurrence measurement method of hydroturbine is proposed in this paper. The self-organizing map is used to establish the baseline model. And then the minimum quantitative error calculated by the baseline model is used to quantitatively evaluate the fault probability and the occurrence can be calculated by this probability. A case of tile burning fault is studied to described the proposed method in detail. The result shows that the proposed methods can efficiently calculate the occurrence.