Runtime Tracking-Based Replication of On-Chip Embedded Software Using Transfer Function Learning for Dust Particle Sensing Systems

Abstract

A digital twin is a widely used method that uses digitized simulations of the real-world characteristics because it is effective in predicting results at a low cost. In digital twin analysis, the transfer function between the input and output data is an important research subject. In this study, we intend to investigate the application of the digital twin method to dust particle sensing. A high-performance multi-channel reference dust particle sensor provides particle count as well as particulate matter information, whereas a lightweight embedded test device only provides a particle count. The particulate matter acquisition algorithm for a reference device is unknown and complex. Instead of that, we propose a simple method to calculate the transfer function using singular-value decomposition. In the experimental results, using singular-value decomposition, the predicted particulate matter of the test device was similar to that of the reference device. The obtained transfer function shows similar measurement results of the two dust particle sensor devices, confirming that particulate matter environmental information can be digitized even with low-power and lightweight sensor-embedded devices. In addition, the power consumption of the test device was approximately ten times lower than that of the reference device.