Design of an Algorithm of Fault Diagnosis Based on the Multiple Source Vibration Signals

Abstract

The main application direction of this paper is to diagnose various fault states or normal states of the three kinds of pumps, such as the lubricating oil pump, the centrifugal pump and the hydraulic pump, on the industrial equipment and other equipment. The data processing is performed according to the acceleration signals measured by each sensor in the fault or normal state of the three pumps provided, and the data is divided into a training set and a validation set. The common algorithm of fault diagnosis is adopted, and the one-dimensional convolutional neural network is used as the core to construct the overall framework of fault diagnosis, so as to judge whether the fault is faulty by detecting the single-source vibration signal, and obtain the correct rate of judgment. The two-dimensional convolutional neural network model is first built, and the method of convolutional neural network is used for multi-source information fusion. The vibration signals measured by the acceleration sensors at four different locations are composed of two-dimensional signals and input into the new two-dimensional convolutional neural network. Input the dataset classification into the architecture for architecture training and accuracy analysis, and change the convolutional neural network structure to achieve higher accuracy.