pdf download - Software and Computer Technology - TU Delft

Automated Fault Diagnosis
3.1 Overview Techniques
of output observables and structure to pinpoint to one or more suspected components. A wellknown
example of this approach is Pinpoint [7].
Off-line inference: A white box approach in which experts use their system knowledge to deduce
and/or abduce a mapping between symptoms and diagnoses.
MBD: A white box approach in which an automated engine deduces diagnoses from a consistencybased
model and real-life observations. The model defines cause-to-effect relations based on
first principles 2 .
Approach
Characteristic Quadrant Reasoning Offline/Online
PMS-1 1 and 2 Manual deduction
and abduction
Diagnosis
By Whom
Offline Manual Service Innovation,
production and development.
Online Manual Service Engineers
and experts.
PMS-2 1 and 2 Manual deduction
and abduction
PMS-3 3 Manual induction Off-line Manual Service engineers
PMS-4 2 Abduction Off-line Manual Service specialists
Automated Induction (BB) 3 Automated induction
(Online) Automated Some automated
entity (see
Section
Automated Induction (WB) 1 Automated induction
Off-line Inference 1 and 2 Manual deduction
and abduction
MBD 1 (and 2 Automated deduction
and 3)
3.2)
(Online) Automated Some automated
entity (see Section
3.4)
Offline Automated Expert system (see
Section 3.3)
Online Automated MBD engine (see
Section 3.5 and
Chapter 4)
Table 3.1: Evaluation approaches.
Table 3.1 classifies MBD in quadrant 1 of Figure 3.1. However, in MBD it is also possible to
emulate fault diagnosis approaches in quadrant 2 and 3. This is useful, because sometimes it is
not possible to formalize a consistency-based model of the system. Another note to Table 3.1 is
that the approaches that use inductive reasoning need a history of diagnoses before diagnoses can
be inferred realtime. Therefore, the ’online’ predicate of approaches ’Automated Induction (BB)’
and ’Automated Induction (WB)’ is written between parentheses; it only holds after a certain time
period. In the following sections, the approaches of Table 3.1 are shortly discussed by using the
power supply example (as introduced in Section 2.1.2). This is done to provide the reader a more
concrete understanding of these approaches. The evaluation of these approaches that is presented in
the final section of this chapter is based upon own estimation. Therefore, sufficient understanding is
needed to judge this evaluation. Figure 3.3 repeats the figure of the power supply example that was
previously shown in Chapter 2.
2 The model used in MBD can also define symptom-diagnosis mappings that are abductively deduced during the
modeling activity. In other words, MBD is able to explicitly implement a LUC (see Chap. 4).
25