Artificial Intelligence Augmented P(AI)ID Cycle-by-Cycle Controller for Automotive DC-DC Converter Applications Based on AURIX TC4x

Inhalt:
The new generation of automotive microcontrollers like AURIX TC4x allows to execute the artificial intelligence (AI) algorithms on these edge devices deployed in the car. These AI structures are limited in their complexity mainly due to available on-chip memory and execution time of the automotive microcontrollers. To serve real time applications like DC-DC converter with cycle-by-cycle control of 100kHz, the execution time of the AI network must be well below 10us. Due to these restrictions, the feedforward network size is limited to approximately 300 neurons. To reduce the AI network complexity, the existing solution of the DC-DC application use case is enhanced. The proposed approach augments well-known PID based DC-DC controller running 100kHz cycle-by-cycle computation. The PID structure with AI extension is forming the proposed P(AI)ID architecture suitable to run in real time on state-of-the-art automotive microcontroller hardware. The proposed solution is demonstrated using phase-shifted full bridge DC-DC topology controlled by AURIX TC4x microcontroller, which is widely used in automotive applications.