Method and System for Creating Digital Twin of a Sensor for a Vehicle in Real Time

Modern Vehicles have many sensors equipped with them to give an idea to the control system of the vehicle as to what is happening in and around the vehicle. These sensors are very costly and they are critical in controlling the vehicle. If certain sensors fail, it can lead to the vehicle being stopped mid operation. Hence, it is imperative that the sensors have a reliable fail-safe mechanism so that above mentioned problems can be prevented, this will provide a better customer experience. For some of the sensors the mechanism for diagnosing and handling the failure is a legislative requirement itself.

However, the current failsafe mechanisms for sensors such as limp home, empirical maps, mathematical models etc. have their own drawbacks. This paper explains how a Deep Learning model trained in real time within a vehicle control unit provides a solution which overcomes all of the existing drawbacks. It also describes in detail what is the behavior of the Deep Learning model when it is getting trained in real time, what are the limitations of such a method and describes how to overcome those limitations. These tests are performed on a custom-built Deep Learning model in MATLAB Simulink and tested for 338 hrs of vehicle drive time simulations.