Low-Complexity Architecture for Cyber-Physical Systems Model Identification

We propose a low complexity architecture for cyber-physical system (CPS) model identification based on multiple-model adaptive estimation (MMAE) algorithms. The complexity reduction is achieved by reducing the number of multiplications in the filter banks of the MMAE algorithm present in the cyber component of the CPS. The architecture has been implemented using FPGA for 16, 32, 64 filter banks as part of position and velocity estimations of autonomous auto-mobile application. It has been found up to 78% reduction in multiplications is possible, which translates to the reduction of 39% lookup tables, 13% FFs, 27% DSPs, and 43% power reduction when compared with the conventional architecture (without multiplications reduction) at 100MHz operating frequency. Furthermore, the proposed architecture is able to identify accurate model of auto-mobile application just within 510 ns, in the presence of external disturbances and abrupt changes.

I. Introduction

Model identification has numerous cyber-physical system (CPS) based applications [1], [2] including autonomous auto-mobiles, adaptive estimation [3], [4], intelligent adaptive plant control [1], fault detection-isolation [5], [6]. CPS systems are often characterised by high degrees of uncertainty, and hence practical adaptive control is likely to be important for good closed loop response. This brief considers a hardware architecture for the parallel implementation of multiple model adaptive estimation (MMAE) schemes for model identification. MMAE forms a component of Multiple Model Adaptive Control (MMAC) schemes, see [7] for a complete modular analysis. Within MMAC, MMAE algorithms to identity the physical plant so that the control architecture can dynamically switch in appropriate controllers in real time. MMAE based algorithms significantly improve performance compared to contemporary designs [8], [9], particularly in the presence of uncertainties including external disturbances and abrupt changes. However, significant computational demands and massive number of filter banks required by the MMAE algorithm [7] preclude its use in resource constrained applications using embedded computing platforms with low cost or low power requirements. To the best of our knowledge, there is no reported hardware architecture for MMAE algorithms. Therefore, in this brief, we propose for the first time a low complexity hardware architecture for the computationally intensive MMAE algorithm for the CPS model identification.