pdf download - Software and Computer Technology - TU Delft

1.5 Outline of the Thesis Introduction
This work explores the possible benefits of automated fault diagnosis of the most recent Cardio-
Vascular X-Ray Systems at Philips Medical Systems. In particular, the problem that is encountered
in the work for this thesis is to improve fault diagnosis of Philips Cardio-Vascular X-Ray Systems
with respect to higher dependability. An improved fault diagnosis approach is able to increase
dependability of the system, because timely and accurate identification of root causes of failures
enables system recovery. Fault diagnosis is only successful when produced diagnoses agree with
reality, so the result could be used to recover the system. Accuracy is the extent to which diagnoses
agree with reality, and is considered to be the most important criterion for diagnostic performance.
This thesis presents a proof-of-concept of the model-based approach to fault diagnosis, aimed
at Philips Cardio-Vascular X-Ray Systems. In order to achieve this, the aim of the project is to
1. uncover the drawbacks of today’s practice to fault diagnosis.
2. show which automated techniques to fault diagnosis could possibly increase diagnostic performance.
3. motivate the choice for the model-based approach by criteria for diagnostic performance.
4. apply the model-based approach to an example system. This is a subsystem of the Philips
Cardio-Vascular X-Ray System. This case study elicits modeling issues that typically occur
when MBD is applied in the industrial domain.
5. study the applicability of LYDIA to model specific dynamics, as well as the use of the accompanying
tool set in a real-life scenario.
The modeling issues that are discussed in the case study include a discussion of an approach
to modeling, the use of entropy to estimate the diagnostic performance of a model, and the use
of entropy to determine next best measurements. Beyond the scope of this thesis is to deal with
multiple and dependent faults.
In this thesis it is shown that the current practice at PMS is not optimal with respect to criteria
for diagnostic performance, and that model-based diagnosis is likely to improve the fault diagnosis
process at PMS. However, it is not possible to conclude that MBD improves fault diagnosis
with respect to higher dependability of Philips Cardio-Vascular X-Ray Systems. The experimental
methodology that should be used to establish this conclusion requires that the diagnostic accuracy
can be determined for both the current practice as the proposed MBD approach. Despite the fact
that it is possible at PMS to know what observations caused a failure, it is impossible to determine
the adjudged diagnosis that explains these observations. Consequently, it is not possible to conclude
that MBD improves fault diagnosis with respect to higher dependability of Philips Cardio-Vascular
X-Ray Systems (see Chapter 6).
It is shown that entropy is a valuable metric to show the improvement of different MBD implementations.
This way, it is established that an MBD approach is able to achieve the same accuracy
and entropy as fault diagnosis applied by experts. Furthermore, entropy is also able to suggest
various alternative MBD implementations that lead to higher accuracy.
1.5 Outline of the Thesis
The outline of the thesis is as follows. In the next chapter, the current approach at PMS to fault
diagnosis is reviewed and analyzed by means of an example. The final section of Chapter 2 presents
6