Fault Diagnostics and Prognostics

A person looking at a monitor in a lab.
Photographer: David Brohede

The Division of Vehicle Systems is active in fault diagnosis and prognosis in both research and education.

Fault diagnosis is about using observations, such as sensor signals, and a mathematical model of a technical system, to detect and diagnose abnormal system behavior. The purpose is to generate diagnoses, i.e., determine if there is a fault in the system and if so, identify the location of the fault. Prognostics is about predicting, based on observations and models, the remaining useful life, i.e., when an ongoing component degradation will result in lost performance or a failure. Fault diagnosis and prognostics are important components in constructing reliable and safe systems, by e.g.

• optimize usage while maximizing the life of key components, such as batteries,
• monitor complex and automated processes and enable fail-safe control to avoid accidents,
• the development of various services such as predictive planning of service and maintenance to minimize the risk of unplanned stops,
• Guide a technician during troubleshooting to minimize the time in the workshop.

The Division of Vehicle Systems is active in fault diagnosis and forecasting in both research and education. The research activities includes both theoretical aspects and method development for fault diagnosis, mainly in model-based diagnosis of nonlinear systems and structural methods, but also in data-driven prognostics and hybrid fault diagnosis that combines physical models with machine learning.

A large part of the research activities in fault diagnosis and prognostics are conducted in collaboration with various industrial partners and have dealt with several different types of diagnostic applications, such as emissions-related vehicle diagnostics, engines, batteries, mining machines, electric machines, and gas turbines.

Model-based diagnosis of complex systems

The principles of model-based diagnosis are about using a mathematical model of the system to be monitored to calculate residual values to be able to draw conclusions about various fault diagnoses. To be able to make reliable decisions about which faults have occurred, as well as minimize the risk of false alarms and missed detections, it is also important to be able to handle model uncertainties and different types of noise. The ability to construct residuals depends on what redundancy is available in the system, in the form of available sensor signals together with a model of the system. An important question is then what kind of residuals can be constructed, i.e., what detectability and isolability performance can be achieved.

Vehicular Systems have long been active in research in the field of fault diagnosis of both linear and nonlinear systems. One example is the research on the use of structural methods to effectively analyze complex dynamic models for sensor placement and computer-aided design of residuals to detect and isolate different types of errors. The research on structural methods for the analysis and design of diagnostic systems for nonlinear systems, for example modeled using ordinary differential equations or differential-algebraic equations, has, among other things, been compiled in a toolbox (Matlab/Python).

Machine learning for diagnosis and prognostics

Data-driven fault diagnosis is about building models from historical data, such as sensor data from different failure scenarios, to be able to predict and classify when a failure occurs. Data-driven diagnosis is complicated by the fact that it can be difficult to collect representative data from relevant failure scenarios and therefore the choice of data-driven model and analysis of data is important to be able to calculate reliable diagnoses and to be able to predict the remaining useful life.

The Division of Vehicle Systems conducts research on data-driven fault diagnosis and prognostics. The purpose is to estimate the remaining life of a system or component, but also how much uncertainty there is in the estimates to, for example, be able to plan when it is best to carry out service and maintenance.

Education

Every year, the Division of Vehicle Systems offers an advanced course for engineering students 'Diagnosis and Monitoring TSFS06 where we teach methods and theory linked to fault diagnosis. We have also for several years organized and supervised diagnosis-related student projects for engineering students in the Automatic Control Project Course TSRT10. We also examine several degree projects every year linked to fault diagnosis and prognostics, both internal and external projects.

Key References

  • Krysander, Mattias, and Erik Frisk. "Sensor placement for fault diagnosis." IEEE Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans 38.6 (2008): 1398-1410.

  • Krysander, Mattias, Jan Åslund, and Mattias Nyberg. "An efficient algorithm for finding minimal over-constrained subsystems for model-based diagnosis." IEEE Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans 38.1 (2007): 197-206.

Doctoral theses from Vehicle Systems in Fault Diagnosis and Prognostics

Model Based Diagnosis and Supervision of Industrial Gas Turbines
Emil Larsson (2014).
PhD thesis, No. 1603, Linköpings universitet.

Model Based Vehicle Level Diagnosis for Hybrid Electric Vehicles
Christofer Sundström (2014).
PhD thesis, No. 1589, Linköpings universitet.

Methods for Automated Design of Fault Detection and Isolation Systems with Automotive Applications
Carl Svärd (2012).
PhD thesis, No. 1448, Linköpings universitet.

Probabilistic Fault Diagnosis with Automotive Applications
Anna Pernestål (2009).
PhD thesis, No. 1288, Linköpings universitet.

Fault Isolation in Distributed Embedded Systems
Jonas Biteus (2007).
PhD thesis, Dissertation No. 1074, Linköpings universitet.

Design and Analysis of Diagnosis Systems Using Structural Methods
Mattias Krysander (2006).
PhD thesis, Dissertation No. 1033, Linköpings universitet.

Residual Generation for Fault Diagnosis
Erik Frisk (2001).
PhD thesis, Dissertation No. 716, Linköpings universitet.

Model Based Fault Diagnosis: Methods, Theory, and Automotive Engine Applications
Mattias Nyberg (1999).
PhD thesis, Dissertation No. 591, Linköpings universitet.

Researchers

About the division

About the department