The latest “ESRA webinar” was given by Prof. Mitra Foularidad, of Ecole Centrale de Marseille on a foundational topic concerning models predictions and uncertainties, on April 6th, 12.00-13.00 CET  (Click here for the slides).  

 Prediction: thorough uncertain models rather than imperfect exact models 

 In presence of data (measurements, observations, simulations), the prediction of a physical phenomenon evolution can be carried out by using statistical models. These models can take into account the underlying physical model or completely neglect this latter but offer a relatively precise prediction with confidence bounds.  These models permit to take into account the uncertainties and randomness related to the parameters, environmental, experimental or usage conditions. The prediction will be proposed with confidence bounds and sensitivity analysis can be carried out. The parameters of these models can be updated based on available new data and an on-line adaptative prediction can be proposed. Statistical models, less expensive in computation time, permit the prediction in cases where deterministic models are not able to give results due to the computation time or due to the complexity of the model in presence of unknown factors and parameters. The relevance of statistical models is essentially linked to the quality and size of available data. We go through some case studies to highlight the interest of these models.

About the speaker: Mitra Fouladirad is Full professor in mathematical modelling for Reliability, Prognosis and Maintenance. After 15 years in University Technology of Troyes focusing in PHM and Maintenance modelling in collaboration with industry she joined in 2021 the Ecole Centrale Marseille in France. Her main focus is in condition-based and predictive maintenance modelling tasking into account  different source of uncertainty. She is one the maintenance technical chairs of ESRA and she participates actively to ESREL since 2007.