Machine Learning Algorithms Applied to the Inference of the Flow Rate in a Non-intrusive Thermal Flow meter

This work presents a non-intrusive thermal flow measurement system for small diameters and flow rates. Such a system employs the principle of forced convection. It has a set of six K-type thermocouple type temperature sensors, and a heating resistor installed on the outer surface of a pipe to obtain the thermal distribution as a combined effect of the heating, generated by the resistor, and of the cooling, caused by the internal flow. From the profile of the thermal distribution in the pipe, the inference of the flow rate is investigated by the linear regression model and, later, in a similar way, with different machine learning algorithms such as K nearest neighbors, Decision trees, Random Forest, AdaBoost, and Gradient Boosting. The results showed that it is possible to develop a non-intrusive meter capable of inferring the flow rate with a maximum error of less than 2%.