This paper presents a machine learning framework for the inverse design of V-beam thermal sensors. The objective is to determine the optimal sensor geometry, including beam inclination angle, beam length, and beam width, to achieve a target displacement under a specified temperature. Additionally, the design aims to minimize both the structural volume and the mechanical stress the sensor must withstand. This problem is ill-posed, as multiple geometric configurations can yield the same displacement, rendering direct regression methods ineffective.
We document a series of five exploratory trials that progressively revealed the nature of the problem, culminating in a two-phase solution: a neural network forward model trained to map geometry and material constants to sensor responses, followed by a gradient-descent inverse optimization over the frozen forward model, minimizing stress and volume simultaneously. The proposed pipeline utilizes a dataset of 3000 samples and achieves a Mean Absolute Percentage Error (MAPE) of 4.76% for predicting displacement, with over 70% of predictions having a MAPE under 5%.
Blogger's Review: This research showcases the potential of machine learning in sensor design, offering effective modeling and optimization methods to tackle complex design challenges. The gradient descent inverse optimization provides a novel approach for multi-variable design, highlighting the practicality of data-driven methods in engineering applications.