Digital Twin Integration for Real-time Monitoring and Predictive Maintenance in Cement Rotary Kilns

Abstract

The development of robust Digital Twin (DT) frameworks has become increasingly important in enhancing the real-time monitoring, fault detection, and predictive maintenance of industrial processes. In this research, a DT model is implemented for a cement rotary kiln to address the critical challenges of operational reliability, performance optimization, and minimizing unplanned downtimes. The main focus of this study is on integrating dynamic first-principle models and probabilistic techniques to enable accurate state estimations, temperature profiling, and fault detection. The methodology involved constructing a first-principle thermal models to simulate the temperature profiles of the kiln alongside the coating estimation. The Extended Kalman Filter (EKF) was employed to iteratively refine state estimations, addressing uncertainties in heat transfer processes. Additionally, the Recursive Least Squares (RLS) algorithm was integrated to further enhance the coating thickness estimation by compensating for uncertainties in radial heat transfer. The burning zone was subdivided into six sections, with updated temperature profiles from the EKF and RLS improving the coating thickness estimates. Known and unknown faults were detected using a combination of EKF and the Online Approximation in Discrete Time (OLAD) system, which facilitated real-time fault learning and prognostics. The results from the Simulink implementation demonstrate that the DT framework provides a strong foundation for real-time monitoring in rotary kilns. The integration of EKF and RLS for state estimation and coating thickness monitoring, combined with OLAD for fault detection and Time to Failure (TTF) estimation, showed promising potential to enhance both fault isolation and system reliability. The ability to detect unknown faults and provide accurate TTF estimates paves the way for more effective maintenance strategies. In conclusion, this research contributes significantly to predictive maintenance in rotary kilns by presenting a novel approach for real-time fault learning, temperature-driven coating thickness monitoring, and fault prognostics. Future work will focus on expanding the model to include mechanical behaviours and refining predictive maintenance strategies for industrial applications.