Condition Monitoring of a Fan using Neural Networks

摘要

Fans are widely used in various industrial fields and it plays a key role in cooling the machinery. For the machinery to work properly, the fan system should remain in stable and error-free condition. Condition monitoring is introduced as a maintenance tool for the failure diagnosis of a fan system. Some methods used in condition monitoring are vibration monitoring and thermal monitoring. Vibration monitoring method was used in this experiment. A fan system based on Machinery Fault SimulatorTM (MFS) was used to simulate different conditions of a fan in the laboratory. An accelerometer was installed on top of the bearing housing. It was used to detect the vibration signal of a running fan. A data acquisition program designed in LabVIEW was used to record and preprocess the raw vibration signal. The collected data was used to detect the condition of the fan system. Neural Network was used for the fault diagnosis. The raw vibration signal is a one-dimensional time domain series data, while the neural network requires multidimensional features as input data. Therefore, it is important to preprocess the raw vibration signal data. Two different preprocessing methods, time-domain features and Auto Regressive (AR) model features were used to preprocess separately. The neural network model was trained by these two methods respectively. The results show that the AR model gave better features than the time domain features method. The condition monitoring system consisted of the following parts: data acquisition, data storage, data preprocessing and the display of results. Some methods were programmed in Matlab, which were called by Matlab scripts in the LabVIEW software. The hybrid programming method helped to generate an efficient program which provided high accuracy of fault diagnosis.