[2A1] Interpretable fusion methodology of health indices and multi-dimensional analogue on complex industrial turbine cavitation condition monitoring
Y Fu¹, D Wang¹ and Z Peng¹,²
¹Shanghai Jiao Tong University, China
²Ningxia University, China
Employing health indices (HIs) to depict machine conditions holds paramount importance in forestalling machine malfunctions and subsequent calamities. Fusion and interpretation of the main contributions of HIs extracted from diverse types of sensors are still challenging to machine condition monitoring with the aim of potential fault prediction and maintenance schedule optimisation. With elements of statistical learning for classification and interpretable weights association, an interpretable fusion methodology of multi-sensory information and HIs is proposed in the sequence of multi-dimensional extension, a theoretical framework for statistical learning, an essence of how HIs are fused to realise machine condition monitoring and a reason why the proposed fusion methodology is interpretable. This approach physically proves and illustrates positive and negative weights of HIs, showcasing its ability to identify industrial turbine cavitation statuses from multiple statuses with greater complexity and decide sensors/channels that are more sensitive and sufficient in condition monitoring. The interpretability of the proposed framework allows timely decision-making and effective countermeasures to be taken when cavitation conditions deteriorate.
