Novas contribuições na aplicação do filtro de Kalman aumentado para o balanceamento de rotores

Abstract

The increasing complexity and criticality of industrial systems have driven the development of advanced methodologies for monitoring and fault diagnosis in rotating machinery, essential for ensuring the operational reliability of such equipment. In this context, virtual sensing (VS) has emerged as a promising solution to overcome the technical and economic limitations of relying solely on physical sensors. By means of mathematical models and estimation algorithms, virtual sensing enables the inference of non-directly measurable quantities, thereby enhancing monitoring capabilities and optimizing resource utilization. At the same time, among the main challenges in rotordynamics, rotor unbalance stands out as a common fault caused by an asymmetric mass distribution around the rotational axis, which induces undesirable centrifugal forces, compromises dynamic stability, increases vibrations, and accelerates component wear. In this regard, balancing becomes a strategic measure to mitigate failures, extend equipment lifespan, and reduce maintenance costs. Thus, aiming to integrate virtual sensing with effective predictive maintenance practices, this thesis proposes an approach based on the Augmented Kalman Filter (AKF) to estimate, in near real-time, non-directly measurable operational states during machine operation, reducing dependence on physical sensors and optimizing balancing procedures. Furthermore, to enable the application of this methodology across diverse industrial contexts, a dedicated Python API was developed for estimating the correction forces required for flexible rotor balancing. Additionally, considering the inherent uncertainties in modeling and experimental measurements, the Differential Evolution (DE) optimization method was incorporated as an automatic tuning technique for the filter parameters, enhancing the estimator's robustness and reducing the need for specialized technical knowledge. The effectiveness of the proposed approach was validated through experimental studies under various uncertainty conditions, including variations in estimation parameters, model accuracy, and measurement noise, demonstrating its reliability and reinforcing its potential as a significant contribution to advancing monitoring and predictive maintenance practices in rotating machinery.