Short name: OptiPLANT
Long name: Event correlation and root-cause analysis for optimised industrial predictive maintenance
Company: Intellia ICT
Call: F4Fp-08 (see call details)
Proposal number: F4Fp-08-M17
SUMMARY REMARKS & TESTBEDS
Predictive maintenance and fault diagnosis in industrial environments is a multidisciplinary field involving several steps across the data processing chain. The OptiPLANT experiment focuses on conducting a detailed analysis of monitored parameters, measurements and configurations based on event correlation and root cause analysis methods. By taking advantage of Fed4FIRE+ software and hardware abstractions, the experiment targets to reveal best practices and adaptive strategies for the diagnosis of system failures or other types of data abnormalities by achieving optimised monitoring and extracting useful knowledge about system attributes as well as contextual information. In this experiment, we validate cloud-based event handling algorithms including event detection, correlation, prediction and filtering. We will study and unveil the rationale behind each considered step and we focus on an event correlation algorithm based on a variable-order Markov model. The proposed theory is applied on big datasets (~250GBs)coming from our end-users (i.e., maritime industrial environment)and is validated through extensive experimentation with real sensor streams originating from large-scale sensor networks deployed in a maritime fleet of ships. The experiment will take place over the Virtual Wall testbed taking advantage of the unique characteristics of the IMEC infrastructure and the Tengu platform for data analytics. The ultimate innovation goal of OptiPLANT is to deliver an open event correlation scheme that could provide explanations on the performance of the industrial system and the production line. This tool will be developed in the context of the OptiPLANT experiment and will be based on the outcomes that will be concluded during our experimentation with Fed4FIRE+platform. The module will be parameterized and tested under different installations, datasets and settings.