Inlet temperature-dependent performance prediction via a physics-informed neural network and flow mechanism analysis in a multistage dense-phase CO2 pump

Dense-phase CO2 pumps stand out as an effective method for transporting CO2 in Carbon Capture, Utilization, and Storage (CCUS). The pumps' characteristics are highly dependent on the operating temperature, which fluctuates with season and location. This paper studies the impact of inlet temperature on the performance of a multi-stage dense-phase CO2 pump. Numerical simulation, incorporating real gas properties, was performed to obtain flow details. Based on sparse samples, a physics-informed neural network (PINN) framework was introduced and trained to predict the overall performance. Compared to the random forest and traditional fully connected layer neural network, the PINN with an appropriate physical constraint demonstrates superior accuracy. The performance maps indicate that an increase in the inlet temperature leads to a decrease in overall pressure rise and shaft power. Additionally, the pump achieves a maximum polytropic efficiency exceeding 80% with higher inlet temperatures shifting the optimal efficiency point toward lower mass flow rates. A velocity triangle model, coupled with an entropy production theory, was proposed to elucidate the performance changes. It is found that the efficiency is governed by the flow fields. These findings provide insight into control and maintenance of dense-phase CO2 pumps in CCUS applications.