Modelling the vibration of a pump working on a pipeline system with artificial neural network and fault detection by using predicted vibration value

In this study, an artificial neural network is modelled for estimating the vibration of pump with the working data of equivalent centrifugal pumps running at different locations of a pipeline system. For this purpose, hourly collected 18 different working data of a pumps running on a real pipeline system which is controlled by SCADA system, which is handled as an experimental setup, are collected. 7 of these working data, which have a higher correlation with target output vibration, are used for training of artificial neural network. In order to observe the effect of the number of data given to ANN for training, data matrixes of 250, 500 and 1000 from each value of 7 data are created by taking into consideration that not to use same data in each matrix. The best neural network model is determined for these pumps and vibration is estimated with 4,95 MAPE value. It is shown that the deviation between the actual vibration and estimated vibration can be used for early fault detection.