Speaker
Description
Temperature-induced variability poses significant challenges to the performance of MEMS sensors, leading to changes in sensor sensitivity and zero-point drift caused by the thermal expansion or contraction of materials. These effects directly compromise the accuracy and reliability of measurements. This study numerically investigates the impacts of temperature variability on two types of MEMS sensors: (i) an electrostatically actuated MEMS sensor and (ii) an environmentally excited MEMS accelerometer. The finite element model of the first sensor is developed based on fundamental principles of structural mechanics, governed by partial differential equations. In contrast, the second sensor is modeled as a mass-spring-damper system, representative of typical MEMS accelerometers. After simulating the sensor outputs under varying temperature conditions, machine learning-aided data normalization methods are introduced to mitigate and eliminate temperature effects from the sensor measurements. These normalization methods are based on the concepts of statistical learning and deep learning, employing unsupervised learning strategies to model the underlying data characterization without the need for explicit temperature recordings. The proposed methods ensure robust correction of temperature-induced drifts, which can enhance the reliability and accuracy of MEMS sensor outputs under fluctuating thermal conditions.
