A Clustering Method And A Neural Network Based On Firefly Algorithm In Diagnostic Analysis

Fault diagnosis of the automotive transmission is a pattern recognition process, but it hasalways been a difficulty in the field of intelligent diagnosis as transmission structure iscomplex, operating conditions change frequently and the background noise is large. As twocommon intelligent diagnostic methods, C-means clustering and BP neural network both havethe local extreme problems resulting in poor diagnostic accuracy and stability.This paper introduces a new firefly clustering method and uses it in mechanical faultidentification. Verify the validity of the method by UCI data and bearing data, and the resultsshow that the firefly algorithm with a global optimization capacity can solve C-meansclustering’s local minima problems efficiently.This paper presents a novel firefly neural network, of which parameters are selectedautomatically. UCI data and bearing data show that the proposed method is better than BPneural network, genetic neural network and particle swarm neural network on the trainingefficiency and fault identification rate. Based on previous studies, make an improvement ofthe firefly algorithm with a movement by partition, and propose a firefly variable structureneural network algorithm (IFAVNN), which can determine the number of nodes in the hiddenlayer automatically in the training process. The effectiveness of the proposed method isverified by the bearing experiment.Carry out the automotive transmission gear failure experiment on a transimission test rig.Use Firefly clustering and IFAVNN to diagnoise the gear samples of three types on singlecondition and different conditions, and analyse the performance of proposed methods used inthe diagnosis process of the transmission. In contrast with other methods, the validity of theproposed methods is verified.