Corrective variational mode decomposition to detect multiple oscillations in process control systems

Abstract

Monitoring oscillatory behavior in industrial control systems is essential to ensure process safety and enhance productivity, but unfortunately, current decomposition-based monitoring methods struggle to extract and accurately detect multiple oscillations. This struggle is primarily due to the level of noise susceptibility in current monitoring methods and the intermittent nature of multiple oscillations in industrial control systems. As a potential solution, variational mode decomposition (VMD) shows promising advantages in processing non-stationary industrial signals with a significant amount of interference. Nevertheless, improperly configuring the number of modes in the VMD can lead to mode mixing and degrade the oscillation extraction accuracy. To overcome this challenge, we propose a corrective VMD approach that automatically adjusts the mode number to create a robust and automated framework for monitoring multiple oscillations. Our framework excels in detecting oscillatory behavior and quantifying the number of oscillations, even in the presence of noisy, intermittent, and irregular disturbances. To validate its effectiveness and practicality, we applied the framework to both a benchmark industrial dataset and a self-constructed industrial dataset, comparing its performance against state-of-the-art oscillation detection methods. The framework demonstrated superior accuracy, achieving 93.90% in detecting oscillations and 85.37% in quantifying the number of oscillations within the benchmark dataset, with similarly excellent results observed in the self-constructed industrial dataset.