liu.seSearch for publications in DiVA
Change search
ReferencesLink to record
Permanent link

Direct link
Bayesian Inference by Combining Training Data and Background Knowledge Expressed as Likelihood Constraints
Linköping University, Department of Electrical Engineering, Vehicular Systems. Linköping University, The Institute of Technology.
Linköping University, Department of Electrical Engineering, Vehicular Systems. Linköping University, The Institute of Technology.
2009 (English)In: International Journal of Approximate Reasoning, ISSN 0888-613XArticle in journal (Other academic) Submitted
Abstract [en]

Bayesian inference, or classification, from data is a powerful method for determining states of process when no detailed physical model of the process exists. However, the performance of Bayesian inference from data is dependent on the amount of training data available. In many real applications the amount of training data is limited, and inference results become insufficient. Thus it is important to take other kinds of information into account in the inference as well. In this paper, we consider a general type of background knowledge that appears in many real applications, for example medical diagnosis, technical diagnosis, and econometrics. We show how it can be expressed as constraints on the likelihoods, and provide detailed description of the computations. The method is applied to a diagnosis example, where it is clearly shown how the integration of background knowledge improves diagnosis when training data is limited.

Place, publisher, year, edition, pages
National Category
Engineering and Technology
URN: urn:nbn:se:liu:diva-51926OAI: diva2:278130
Available from: 2009-11-24 Created: 2009-11-24 Last updated: 2009-11-30Bibliographically approved
In thesis
1. Probabilistic Fault Diagnosis with Automotive Applications
Open this publication in new window or tab >>Probabilistic Fault Diagnosis with Automotive Applications
2009 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

The aim of this thesis is to contribute to improved diagnosis of automotive vehicles. The work is driven by case studies, where problems and challenges are identified. To solve these problems, theoretically sound and general methods are developed. The methods are then applied to the real world systems.

To fulfill performance requirements automotive vehicles are becoming increasingly complex products. This makes them more difficult to diagnose. At the same time, the requirements on the diagnosis itself are steadily increasing. Environmental legislation requires that smaller deviations from specified operation must be detected earlier. More accurate diagnostic methods can be used to reduce maintenance costs and increase uptime. Improved diagnosis can also reduce safety risks related to vehicle operation.

Fault diagnosis is the task of identifying possible faults given current observations from the systems. To do this, the internal relations between observations and faults must be identified. In complex systems, such as automotive vehicles, finding these relations is a most challenging problem due to several sources of uncertainty. Observations from the system are often hidden in considerable levels of noise. The systems are complicated to model both since they are complex and since they are operated in continuously changing surroundings. Furthermore, since faults typically are rare, and sometimes never described, it is often difficult to get hold of enough data to learn the relations from.

Due to the several sources of uncertainty in fault diagnosis of automotive systems, a probabilistic approach is used, both to find the internal relations, and to identify the faults possibly present in the system given the current observations. To do this successfully, all available information is integrated in the computations.

Both on-board and off-board diagnosis are considered. The two tasks may seem different in nature: on-board diagnosis is performed without human integration, while the off-board diagnosis is mainly based on the interactivity with a mechanic. On the other hand, both tasks regard the same vehicle, and information from the on-board diagnosis system may be useful also for off-board diagnosis. The probabilistic methods are general, and it is natural to consider both tasks.

The thesis contributes in three main areas. First, in Paper 1 and 2, methods are developed for combining training data and expert knowledge of different kinds to compute probabilities for faults. These methods are primarily developed with on-board diagnosis in mind, but are also applicable to off-board diagnosis. The methods are general, and can be used not only in diagnosis of technical system, but also in many other applications, including medical diagnosis and econometrics, where both data and expert knowledge are present.

The second area concerns inference in off-board diagnosis and troubleshooting, and the contribution consists in the methods developed in Paper 3 and 4. The methods handle probability computations in systems subject to external interventions, and in particular systems that include both instantaneous and non-instantaneous dependencies. They are based on the theory of Bayesian networks, and include event-driven non-stationary dynamic Bayesian networks (nsDBN) and an efficient inference algorithm for troubleshooting based on static Bayesian networks. The framework of nsDBN event-driven nsDBN is applicable to all kinds of problems concerning inference under external interventions.

The third contribution area is Bayesian learning from data in the diagnosis application. The contribution is the comparison and evaluation of five Bayesian methods for learning in fault diagnosis in Paper 5. The special challenges in diagnosis related to learning from data are considered. It is shown how the five methods should be tailored to be applicable to fault diagnosis problems.

To summarize, the five papers in the thesis have shown how several challenges in automotive diagnosis can be handled by using probabilistic methods. Handling such challenges with probabilistic methods has a great potential. The probabilistic methods provide a framework for utilizing all

information available, also if it is in different forms and. The probabilities computed can be combined with decision theoretic methods to determine the appropriate action after the discovery of reduced system functionality due to faults.

Place, publisher, year, edition, pages
Linköping: Linköping University Electronic Press, 2009. 44 p.
Linköping Studies in Science and Technology. Dissertations, ISSN 0345-7524 ; 1288
National Category
Engineering and Technology
urn:nbn:se:liu:diva-51931 (URN)978-91-7393-493-0 (ISBN)
Public defence
2009-12-18, Visionen, Hus B, Campus Valla, Linköpings universitet, Linköping, 10:15 (English)
Available from: 2009-11-30 Created: 2009-11-24 Last updated: 2010-11-23Bibliographically approved

Open Access in DiVA

No full text

Other links

Link to Ph.D. Thesis

Search in DiVA

By author/editor
Pernestål, AnnaNyberg, Mattias
By organisation
Vehicular SystemsThe Institute of Technology
In the same journal
International Journal of Approximate Reasoning
Engineering and Technology

Search outside of DiVA

GoogleGoogle Scholar
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

Total: 202 hits
ReferencesLink to record
Permanent link

Direct link