Reinforcement Learning based intelligent fan controller

In Automotive world, the major factors that are important during vehicle development are design and testing of hardware and software. Hardware involves all the components, modules and systems in a vehicle and Software involves the control logic for functioning of these components. Generally, software inside Control Unit could be validated in various ways e.g., Software in Loop, Hardware in Loop, Vehicle testing. During software validation, Plant model and Controller may differ in level of complexity and largely depends on timeline, availability of components, etc. Many times, hardware components might be available before control software and there will be a need to test them for their performance, for which a near to accurate control software is required. Intelligent Controller mimicking the original controller can be an option to reduce complexity and improve the quality of testing. This Paper proposes a Reinforcement Learning (RL) based intelligent controller which captures dependency behavior of variables for controlling the FAN actuator in a coolant system which could be extended to other components as well. This controller can either be integrated in parallel to actual controller or be a standalone for a project which includes system and components. In simulation world, this could be advantageous from resource availability point of view, as it could act as an alternative during initial phase of control software development. RL based controller can also help to answer questions related to component behavior to obtain better efficiency by minimizing losses. It will also automatically find action within tolerant ranges of actual component behavior by maximizing rewards and thereby reducing modelling and testing time of components or plant models.