Adaptive Multi-Model Fault Diagnosis of Dynamic Systems for Motion Tracking

For most real-world systems, the exact description of possible faults is unknown, making these faults difficult to detect, and even more difficult to identify. The most promising way is to use multiple hypotheses for faults to find the best fitting fault model by comparing system measurements with t...

Full description

Saved in:
Bibliographic Details
Main Authors: Annalena Daniels, Tommaso Benciolini, Dirk Wollherr, Marion Leibold
Format: Article
Language:English
Published: IEEE 2024-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10815966/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1846099109112446976
author Annalena Daniels
Tommaso Benciolini
Dirk Wollherr
Marion Leibold
author_facet Annalena Daniels
Tommaso Benciolini
Dirk Wollherr
Marion Leibold
author_sort Annalena Daniels
collection DOAJ
description For most real-world systems, the exact description of possible faults is unknown, making these faults difficult to detect, and even more difficult to identify. The most promising way is to use multiple hypotheses for faults to find the best fitting fault model by comparing system measurements with the predictions of the multi-model algorithm. However, this may lead to the need for infinite hypotheses. We propose a novel multi-model approach that considers a small number of different models with a known macro-structure and unknown parameters, combining system identification with simultaneous fault diagnosis. The unknown parameters in the models are estimated using a maximum likelihood approach. The fitted models are then used in an interacting multiple model algorithm to determine the most likely model that best describes the system behavior at any moment in time. An overfitting problem emerging from short data sequences is discussed, and two solutions are introduced. First, a regularization term in the probability estimation is suggested to penalize frequent parameter changes that signal possible overfitting. Second, an algorithm with a shifted data set is presented. The effectiveness of the algorithms is demonstrated on a motion tracking problem where the different fault hypotheses represent the macro-behavior of a moving object, and the real system switches between different modes. In a comparison, the proposed algorithms are the only ones that reliably identify the defined faults. They can be easily adapted to other fault diagnosis problems.
format Article
id doaj-art-b1b29c902eac417a890097574d4ec977
institution Kabale University
issn 2169-3536
language English
publishDate 2024-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-b1b29c902eac417a890097574d4ec9772025-01-01T00:01:48ZengIEEEIEEE Access2169-35362024-01-011219754019755610.1109/ACCESS.2024.352281110815966Adaptive Multi-Model Fault Diagnosis of Dynamic Systems for Motion TrackingAnnalena Daniels0https://orcid.org/0000-0002-3680-7610Tommaso Benciolini1https://orcid.org/0000-0001-5394-947XDirk Wollherr2https://orcid.org/0000-0003-2810-6790Marion Leibold3https://orcid.org/0000-0002-2802-5600Chair of Automatic Control Engineering, Technical University of Munich, Munich, GermanyChair of Automatic Control Engineering, Technical University of Munich, Munich, GermanyChair of Automatic Control Engineering, Technical University of Munich, Munich, GermanyChair of Automatic Control Engineering, Technical University of Munich, Munich, GermanyFor most real-world systems, the exact description of possible faults is unknown, making these faults difficult to detect, and even more difficult to identify. The most promising way is to use multiple hypotheses for faults to find the best fitting fault model by comparing system measurements with the predictions of the multi-model algorithm. However, this may lead to the need for infinite hypotheses. We propose a novel multi-model approach that considers a small number of different models with a known macro-structure and unknown parameters, combining system identification with simultaneous fault diagnosis. The unknown parameters in the models are estimated using a maximum likelihood approach. The fitted models are then used in an interacting multiple model algorithm to determine the most likely model that best describes the system behavior at any moment in time. An overfitting problem emerging from short data sequences is discussed, and two solutions are introduced. First, a regularization term in the probability estimation is suggested to penalize frequent parameter changes that signal possible overfitting. Second, an algorithm with a shifted data set is presented. The effectiveness of the algorithms is demonstrated on a motion tracking problem where the different fault hypotheses represent the macro-behavior of a moving object, and the real system switches between different modes. In a comparison, the proposed algorithms are the only ones that reliably identify the defined faults. They can be easily adapted to other fault diagnosis problems.https://ieeexplore.ieee.org/document/10815966/Parameter estimationfault diagnosismultiple model algorithmmaximum likelihood estimationinteracting multiple model algorithm
spellingShingle Annalena Daniels
Tommaso Benciolini
Dirk Wollherr
Marion Leibold
Adaptive Multi-Model Fault Diagnosis of Dynamic Systems for Motion Tracking
IEEE Access
Parameter estimation
fault diagnosis
multiple model algorithm
maximum likelihood estimation
interacting multiple model algorithm
title Adaptive Multi-Model Fault Diagnosis of Dynamic Systems for Motion Tracking
title_full Adaptive Multi-Model Fault Diagnosis of Dynamic Systems for Motion Tracking
title_fullStr Adaptive Multi-Model Fault Diagnosis of Dynamic Systems for Motion Tracking
title_full_unstemmed Adaptive Multi-Model Fault Diagnosis of Dynamic Systems for Motion Tracking
title_short Adaptive Multi-Model Fault Diagnosis of Dynamic Systems for Motion Tracking
title_sort adaptive multi model fault diagnosis of dynamic systems for motion tracking
topic Parameter estimation
fault diagnosis
multiple model algorithm
maximum likelihood estimation
interacting multiple model algorithm
url https://ieeexplore.ieee.org/document/10815966/
work_keys_str_mv AT annalenadaniels adaptivemultimodelfaultdiagnosisofdynamicsystemsformotiontracking
AT tommasobenciolini adaptivemultimodelfaultdiagnosisofdynamicsystemsformotiontracking
AT dirkwollherr adaptivemultimodelfaultdiagnosisofdynamicsystemsformotiontracking
AT marionleibold adaptivemultimodelfaultdiagnosisofdynamicsystemsformotiontracking