Affiliation:
1. Institute of Fluid Mechanics and Hydraulic Machinery, University of Stuttgart, 70569 Stuttgart, Germany
Abstract
The head and flow rate of a pump characterize the pump performance, which help determine whether maintenance is needed. In the proposed method, instead of a traditional flowmeter and manometer, the operating points are identified using data collected from accelerometers and microphones. The dataset is created from a test rig consisting of a standard centrifugal water pump and measurement system. After implementing preprocessing techniques and Convolutional Neural Networks (CNNs), the trained models are obtained and evaluated. The influence of the sensor location and the performance of different signals or signal combinations are investigated. The proposed method achieves a mean relative error of 7.23% for flow rate and 2.37% for head with the best model. By employing two data augmentation techniques, performance is further improved, resulting in a mean relative error of 3.55% for flow rate and 1.35% for head with the sliding window technique.
Subject
Mechanical Engineering,Energy Engineering and Power Technology,Aerospace Engineering
Reference16 articles.
1. Fault diagnosis of a centrifugal pump using MLP-GABP and SVM with CWT;Altobi;Eng. Sci. Technol. Int. J.,2019
2. Application of CNN-LSTM in gradual changing fault diagnosis of rod pumping system;He;Math. Probl. Eng.,2019
3. Tang, S., Zhu, Y., Yuan, S., and Li, G. (2020). Intelligent diagnosis towards hydraulic axial piston pump using a novel integrated CNN model. Sensors, 20.
4. Zhao, W., Wang, Z., Lu, C., Ma, J., and Li, L. (2016, January 12–15). Fault diagnosis for centrifugal pumps using deep learning and softmax regression. Proceedings of the 2016 12th World Congress on Intelligent Control and Automation (WCICA), Guilin, China.
5. Wu, J., and Zhang, X. (2022). Convolutional Neural Network Identification of Stall Flow Patterns in Pump–Turbine Runners. Energies, 15.