Flow measurement is a key factor in process management, particularly in the nuclear and industrial sectors. However, current measurement methods require complex installations, especially in environments with strict regulations, such as in the nuclear sector. To address these challenges, the CEA has developed an innovative method for measuring flow in non-isothermal fluids, based on the analysis of thermal fluctuations. This technique, which uses two temperature sensors installed upstream and downstream on the pipeline, is simple to implement and involves minimal constraints. The temperature variations are carried by the flow from one sensor to the other, and by comparing the signals recorded by these sensors, it is possible to calculate the thermal transit time between them, which allows the flow velocity, and consequently, the flow rate, to be determined. The goal of this thesis is to optimize this method by enhancing its reliability. To achieve this, the propagation of thermal noise within the flow will be studied, and both the type and placement of the sensors will be optimized. This work will be carried out within the Core and Circuit Thermohydraulics Laboratory in collaboration with the Instrumentation, System and Method Laboratory, which has state-of-the-art experimental equipment. Numerical simulations will complement the experimental work to validate the obtained results. In parallel, artificial intelligence approaches will be explored to improve the processing of thermal signals. By the end of the thesis, the doctoral candidate will have acquired extensive skills in experimental and numerical work and will be able to leverage these in future endeavors.