Audio Fault Analysis for Industrial Equipment Based on Feature Metric Engineering with CNNs

Work in the field of industrial production equipment machine is a mechanical noise used to diagnosis the important feature of the machine working state dimension. Early when the abnormal running condition of mechanical equipment, is often accompanied by abnormal audio signal. But industrial production equipment abnormal audio signal with normal audio signal generated by the class spacing is small, which leads to failure of audio features is difficult to locate. In order to effectively solve this problem, this paper puts forward a kind of engineering and the convolutional neural network model based on feature measure industrial audio abnormal analysis method. First of all, the preprocessing operations industrial audio data set, the audio data standardization, First, the industrial audio dataset is subjected to pre-processing operations to standardize the audio data, and then the features of different domains are extracted for each audio segment and the similarity of the features is measured to construct a feature-classification map. Then, combined with the evaluation indexes in feature metric engineering to select Then, appropriate audio features are selected by combining the evaluation metrics in the feature metric project, and then input into the convolutional neural network to train the classification model, and finally use the model to realize the binary classification of normal/abnormal tones. On a real The experimental results on the industrial audio dataset of real scenes show that the algorithm effectively breaks through the limitation that the general scene oriented audio classification methods are difficult to be applied to the industrial audio field. The classification accuracy reaches 0.967.