Aller au contenuAller au menuAller à la rechercheAller à la page d'actualités

Laboratoire Angevin de Recherche en Ingénierie des Systèmes

Navigation principale


    Fil d'ariane

    Séminaire LARIS - 11 décembre 2018

    Séminaire LARIS - 11 décembre 2018

    • Partager la page sur les réseaux sociaux
    • Envoyer cette page par mail

      Envoyer par mail

      Séparés par des virgules
    • Imprimer cette page

    Séminaire LARIS - Daniel Jung

    14h00 | salle du conseil - B106 | IstiA - Université d'Angers

    Le 11 décembre 2018

    Combining physical-based models and training data for fault diagnosis

    Daniel Jung a)

    a)Assistant Professor at Linköping University



    Data-driven fault diagnosis uses machine learning and training data from nominal system operation and different fault scenarios to train fault classifiers. However, machine learning methods require large data sets, representing all relevant operation scenarios, to train a reliable model with satisfactory prediction or classification performance.

    In many industrial applications training data is limited or it is not possible to collect data from all relevant cases in advance. One solution is to utilize a physical-based model of the system. Model-based diagnosis try to determine the system condition based on a mathematical model, derived based on physical insight of the system. By using a set of fault detectors, for example residuals, where each detector is designed to monitor a given subset of the system, fault detection and isolation can be performed. In many systems, there are more residual candidates that can be designed than necessary to detect and isolate the faults. However, model uncertainties and sensor noise will have a negative impact on fault detection performance of the different residual candidates, meaning that all residuals are not equally good. Machine learning methods are suitable to select sets of residuals and to design fault detectors with good detection and isolation performance by using models and available training data.

    In this presentation, I will present some of my research where I have investigated how to utilize available models and training data in the diagnosis system design process. I will also present an approach how to combine model-based and data-driven fault diagnosis methods for fault isolation when training data is limited. As a case study, I have used data from an internal combustion engine during transient operation with different faults injected into the system during operation.

    Daniel Jung was born in Linköping, Sweden in 1984. He received a Ph.D. degree in 2015 from Linköping University, Sweden. During 2017 he was a Research Associate at the Center for Automotive Research at The Ohio State University, Columbus, OH, USA.

    Since 2018, he is an Assistant Professor at Linköping University. His current research interests include theory and applications of model-based and data-driven fault diagnosis, and optimal control of hybrid electric vehicles.

    He works with SFD team.