15th European Conference on Turbomachinery Fluid dynamics & Thermodynamics

Paper ID:

ETC2023-160

Main Topic:

Hydraulics Machine

Authors

Hanbing Ma  - University of Stuttgart, Germany
Oliver Kirschner - University of Stuttgart, Germany
Stefan Riedelbauch - University of Stuttgart, Germany

Abstract

Nowadays, pump units are installed in many plants for a wide variety of applications. Due to the long operating time, wear and tear such as erosion, abrasion and corrosion are inevitable. The monitoring of operating points and a subsequent evaluation of the condition of the pump may support the decision for required maintenance. In this paper, convolutional neural networks (CNN) are implemented to predict the operating points of pump units. A test rig with a standard centrifugal water pump and measurement system is designed. With the proposed preprocessing method, the data collected from accelerometers, microphones, and structure-borne sound sensor are used as input for the neural network. The head and volume flow of the pump, which characterize the pump performance, is selected as output variables. The influence of the location of sensors and the performance of different signals or signal combinations are investigated. A systematic comparison is performed, and the results are evaluated through the performance of operating points estimation.







Download it! Paper is available from journal web site