Research on Classification of Carbon Content in Pr-Nd Alloys Based on MFCC and BP Neural Network

Abstract

The carbon content in rare earth praseodymium neodymium alloy is a key indicator to define whether the alloys are qualified. In order to improve the measure efficiency of the carbon content in alloys and reduce the cost of quality inspection, this paper proposes a soft measurement method of carbon content in praseodymium-neodymium alloys based on the extracted Mel Frequency Cepstrum Coefficient (MFCC) features. and BP neural network. A three-layer structure of BP neural network was established according to the input and output requirements to classify the MFCC features of different carbon content alloys. Lastly, a number of 539 signals from praseodymium neodymium alloys with varied carbon content were used to verify the neural network and the accuracy of classification reached up to 97.44%. This paper also introduced the receiver operating characteristic curve (ROC) to comprehensively evaluate the trained BP neural network model. The ROC showed that the probability of identifying a negative sample as a positive sample was 2.56% and the area under the ROC curve line accounted for 85.5%, which indicated that the proposed non-destructive testing (NDT) method was effective and had potential for practical application.