An Improved Fuzzy Neural Network and Its Application in Machine Fault Diagnosis

This paper presents an improved fuzzy neural network (IFNN) for pattern recognition. The IFNN consists of several sub-networks, which represent different patterns. Each sub-network distinguishes a particular pattern from others, and each pattern corresponds to the certain inputs. In IFNN, an empirical formula tested many times is used to calculate the number of nodes in the hidden layer, and the learning algorithm with 3 self-adjustable coefficients is utilized to improve the learning efficiency of the training process. After that the ultimate outputs represent the degree of the state data belonging to the specific pattern. The performance of the IFNN was tested and verified by an example of machine fault diagnosis, and another issue about knowledge discovery was put forward.