A Study on Fault Diagnosis of Valve Cooling Equipment Based on Parallel Voiceprint Recognition Algorithm of AMFCC and DNN

Aiming at the lack of online monitoring means for valve cooling equipment, its internal faults often rely on the judgment of operation and maintenance experience, the degree of digital operation and maintenance is low and other problems. This article proposes the use of AMFCC (Attention Mel Frequency Cepstral Coefficient) and DNN parallel algorithms to recognize and determine the operating status of valve cooling equipment by utilizing the voiceprint information generated during operation. Firstly, the sound data of the valve cooling equipment under different working conditions are collected and pre-processed by frame splitting and window adding. Then, by adding attention mechanism to the Mel cepstral coefficient (MFCC) feature extraction method, an AMFCC algorithm model is built to extract voiceprint features from the data. Finally, TensorFlow is used to process the data in parallel and combined with deep neural network (DNN) for acoustic feature recognition to realize the fault diagnosis of valve-cooled equipment. The research results indicate that the AMFCC and DNN algorithm models have F1 scores of over 90% for identifying five types of faults in valve cooling equipment through voiceprint data, which is consistent with the characteristic of smooth changes in sound energy of valve cooling equipment.