The modern transportation is rapidly undergoing a major shift towards electrification to improve sustainability and reduce its carbon footprint. The electrification progress includes heavy duty commercial vehicles, creating opportunities to introduce novel technologies in the development process. Compared to conventional trucks propelled by internal combustion engines, battery electric trucks have more complex thermal management systems that need to operate optimally to fulfill the requirements on vehicle range and driver cabin comfort. The development of efficient thermal management systems is becoming ever-more model-based ranging from detailed physics-based models to reduced order modeling techniques. Methods of developing reduced order models for virtual sensing could reduce costs for truck testing and increase efficiency of the development of the thermal system. The coolant flow in the system is of special interest, since controlling the flow in the system plays a significant role in thermal management. This project investigates how qualitative data can be generated from the physics-based GTSuite model and used to build two different deep machine learning models to estimate the flow at various points in the cooling system. The results indicate that the developed methods work well for estimating the flow in this system and that the accuracy is close to simulated measurements. This report concludes that this approach provides an effective method for creating virtual sensors with satisfactory accuracy. The report further proposes a method to determine the actuator settings for a requested coolant flow by integrating a feedback loop with a PID controller and utilizing the virtual sensors in a Simulink environment.