Strip Surface Defect Diagnosis Method Based on Extreme Learning Machine with Different Excitation Functions.

Strip surface defects seriously impress the appearance of products, also lead to a decrease in wear resistance, corrosion resistance and the increase of fatigue strength of strip products, thus greatly reducing the service life of strip products. Therefore, a strip surface defect diagnosis algorithm by using on extreme learning machine (ELM) with variable excitation functions was proposed. Based on the neural network principle and the learning method of ELM, the seven excitation functions (Sigmoid, Sin, Hardlim, ReLu, Tanh, Morlet and Arctan) were used to establish the strip surface defect diagnosis model respectively. The neuron numbers of the hidden layer of ELM were analyzed experimentally. The simulation experiment results for the strip surface defect data set in UCI show that the ELM neural network can effectively diagnose the strip surface defect accurately. [ABSTRACT FROM AUTHOR]