Machine-Learning (ML) Enabled Reliable Multi-Modal Sensor Operation for Rocket Propulsion Systems

Alphacore and its Research Partner, Arizona State University, will develop a framework for self-calibrating sensors, backed by artificial intelligence with in-field calibration capabilities. In Phase I we will prove the feasibility of our approach by modeling MEMS and electronics-based pressure, temperature, strain, and acoustics sensors, designing electrical tests to correlate with physical characteristics, and designing a hardened parametrizable self-test IP. In Phase II we will fabricate test and prototype circuits that implement and validate the work done in Phase I, as well as extend the concepts developed in Phase I to other types of sensors.Phase I of this program will target capacitive pressure sensors, electronics-based temperature sensors, and a MEMS based acoustic sensors. In developing the self-test IP, Alphacore will make every effort to accommodate a large portion of the commercially available collection of MEMS sensors. The self-test IP specifics will be determined based on research on commercially available devices.This project will develop methodologies for 2-tier calibration of sensor-based machine learning systems. The goal of sensor front-end calibration is to maintain highest level of sensor performance throughout the operation. To this end, the sensor hardware is monitored and calibrated continuously in real-time based on the readings built-in self-test monitors. These monitors are implemented as electrical excitation units with an area overhead less than 5% and negligible performance impact. The monitors can be used for extracting sensor performance as well as determining an error model to calibrate the software. Sensor hardware calibration can be in terms of changing bias conditions or determining sensor offset and sensitivity that converts the voltage/current reading back to the value of the physical stimulus, i.e., pressure or acceleration. nbsp;