论文部分内容阅读
研究了一种用模糊集表示火箭发动机故障模式的神经网络分类器。模糊集是由模糊超立方体聚集形成的集合体，模糊超立方体是一个极小点和极大点构成的用隶属函数表示的ｎ维方盒。极小点和极大点的确定用包含有扩张与收缩阶段的模糊极小极大学习算法实现，这种算法能在一次循环学习中形成非线性模式边界，无需对已知故障模式重新训练就可融合新样本和精炼已存在的故障模式。模糊集用于故障模式分类自然地提供了故障更高水平分类的有用信息。液体火箭发动机故障分类的数值仿真解释了模糊极小极大神经网络的优越性能。
A neural network classifier is presented to represent the failure modes of rocket engines with fuzzy sets. Fuzzy sets are aggregates formed by the aggregation of fuzzy hypercubes, which are n-dimensional square boxes represented by membership functions consisting of minima and maxima. The determination of the minimum point and the maximum point is realized by a fuzzy minimax learning algorithm that includes the expansion and contraction phases. This algorithm can form the non-linear mode boundary in one cycle of learning without the need of retraining the known failure mode Fusion of new samples and refinement of existing failure modes. Fuzzy sets used for fault pattern classification naturally provide useful information for higher level fault classification. The numerical simulation of liquid rocket engine fault classification explains the superior performance of fuzzy minimax neural network.