An Industry Scale Digital-Twin for Machine Fault Diagnosis and Prognostics

Digital Twins promise a new capability for machine monitoring through a virtual real-time replica of a real physical system that can be used to improve operational efficiencies, identify faults and quantify the effects of different operating scenarios. However, few examples exist of true Digital Twins that have been deployed in real world, full-scale industrial operations, where these cyber-physical systems adopt standardized and recognized software platforms and sensing technologies. This paper describes the practicalities developed to manage an industry-scale digital twin for machine condition monitoring through the integration of intelligent algorithms for automated fault diagnostics and future fault state prognostics. Built in collaboration between Ailsa Reliability Solution Ltd and The University of Strathclyde, The Fan Skid Digital Twin system integrates an industry scale electro-mechanical fan assembly with statistical fault models through the use of commercially available cloud-based infrastructure. A standardized sensing and telemetry philosophy have been adopted alongside the implementation of generic control and actuation systems to provide a robust approach to bi-directional interactions. Automated fault diagnosis is performed using an Expert Bayesian Network approach elicited from domain knowledge. Subsequently, a statistical approach to modeling the fault and degradation pathway is adopted, which is constrained by well understood mechanical engineering principles. Furthermore, a rudimentary approach to prognosis uncertainty quantification is introduced, based upon extrapolation of industry standard sensor equipment error margins, to provide best- and worst-case future fault scenario characteristics. Finally, the architecture demonstrates close-to real-time feedback control capability to prevent the onset of worsening faults.

Keywords: Digital Twin, Fan, Condition Monitoring, Diagnostics, Prognostics, Uncertainty.