Sbarbaro Daniel, Johansen Tor Arne, Yañez Jorge
Department of Electrical Engineering, Faculty of Engineering, Universidad de Concepcion, Concepción 4070371, Chile.
Center for Autonomous Marine Operations and Systems (AMOS), Department of Engineering Cybernetics, Norwegian University of Science and Technology, 7491 Trondheim, Norway.
Sensors (Basel). 2025 Apr 16;25(8):2513. doi: 10.3390/s25082513.
This study addresses the challenges of real-time spectroscopic sensing in industrial applications, where external factors such as temperature fluctuations, pressure variations, and particle size distribution significantly impact measurement accuracy. Conventional quantitative analytical methods often neglect these dynamic influences, leading to erroneous concentration estimates. To overcome these limitations, we propose an integrated modeling framework that combines a discrete-time process model with a physics-based spectroscopic sensor model, explicitly accounting for the dynamic properties of the system. A key innovation of this work is the development and application of an Adaptive Kalman Filter (AKF) to systematically correct for measurement distortions caused by external disturbances. Unlike conventional filtering techniques, the AKF dynamically adjusts to changing process conditions by leveraging real-time observability analysis, ensuring robustness even in the presence of sensor noise and environmental variability. Furthermore, to address cases where full observability is not achievable, we introduce a reduced-order Adaptive Kalman Filter (rAKF), which optimally estimates concentrations while minimizing computational complexity. A comprehensive series of simulations was conducted to assess the sensitivity of the estimation to variations in external signal type, noise levels, and initial values for parameters and states. The findings of this study demonstrate the superior performance of both AKF and rAKF in comparison to conventional filtering techniques, including the Extended Kalman Filter. The proposed approaches have been shown to enhance the reliability of spectroscopic sensor measurements, enabling more precise real-time estimations that can be used for monitoring and advanced process control strategies in industrial settings.
本研究探讨了工业应用中实时光谱传感面临的挑战，在这些应用中，诸如温度波动、压力变化和粒度分布等外部因素会显著影响测量精度。传统的定量分析方法往往忽略这些动态影响，导致浓度估计错误。为克服这些限制，我们提出了一个集成建模框架，该框架将离散时间过程模型与基于物理的光谱传感器模型相结合，明确考虑了系统的动态特性。这项工作的一个关键创新是开发并应用了自适应卡尔曼滤波器（AKF），以系统地校正由外部干扰引起的测量失真。与传统滤波技术不同，AKF通过利用实时可观测性分析动态调整以适应不断变化的过程条件，即使在存在传感器噪声和环境变化的情况下也能确保鲁棒性。此外，为解决无法实现完全可观测性的情况，我们引入了降阶自适应卡尔曼滤波器（rAKF），它在最小化计算复杂度的同时最优地估计浓度。进行了一系列全面的模拟，以评估估计对外部信号类型、噪声水平以及参数和状态初始值变化的敏感性。本研究结果表明，与包括扩展卡尔曼滤波器在内的传统滤波技术相比，AKF和rAKF均具有卓越性能。所提出的方法已被证明可提高光谱传感器测量的可靠性，实现更精确的实时估计，可用于工业环境中的监测和先进过程控制策略。
