| dc.description.abstract | The H-Q curve serves as a critical parameter in pump operation and design selection for a given system. However, it is important to recognize that changes over time, such as wear and tear on the pump, pipes, and fittings, as well as alterations to the process and external factors, can significantly impact the pump's performance.
Monitoring changes in the head estimate over time can serve as an early indicator of potential issues, allowing operators or maintenance personnel to take corrective action promptly. The assumption here is that the machine learning models generated can be seamlessly integrated into a control system and executed therein, following a reduction to weights and biases matrices. However, it is worth noting that similar integration can also be achieved through first principles modeling or soft sensors.
This thesis utilizes machine learning, specifically simple neural networks, to achieve positive results in analyzing pump systems. By employing these techniques, the thesis successfully identifies H-Q curves within noisy data, laying the foundations upon which condition monitoring systems can be built. These findings highlight the potential of machine learning in enhancing the understanding and management of pump systems, offering opportunities for improved efficiency, reliability, and predictive maintenance strategies. | |
