Multi-source Vibration Signals Fusion-Based Condition Assessment of Francis Turbine Unit under Complex Operating Conditions

Francis turbine units (FTUs) operate under complex conditions and gradually deteriorate during their service life. Information about operating characteristics is indirectly contained in the monitoring vibration signal of the FTUs. However, the influence of operating conditions on vibration characteristics is usually not considered in vibration data cleaning for FTUs. Furthermore, existing unit condition assessment methods require the construction of a health mapping model between operating condition parameters and unit vibration signals. Analysis of single-source vibration signals alone makes it difficult to comprehensively and reliably evaluate unit condition. In this paper, a multi-source vibration signals fusion-based condition assessment of FTUs under complex operating conditions is proposed. Firstly, the sparse spectrum Gaussian process regression method is introduced to establish a probability fitting model of the operating condition-vibration amplitude, calculate the deviation index between working head, power and the vibration distribution under different operating conditions. And then the K-Means++ clustering method based on the deviation index is designed to achieve anomaly data cleaning. Secondly, taking into account the coupling relationship between the operating condition parameters and the multisource vibration signals, a multisource vibration-operating condition correlation health model based on dual-channel convolutional auto-encoder is proposed. Furthermore, by integrating the degradation information of multi-source vibration signals, a comprehensive degradation index is constructed to comprehensively quantify the degradation state of the Francis turbine unit. Finally, the effectiveness of the proposed method is validated through a set of on-site measured data from a large-scale hydropower station.