Embedding diagnosability of complex industrial systems into the design process using a model-based methodology
conference paper
There is a constant increase of the market expectations on the capabilities of industrial high-tech systems. To meet these expectations, designers of such systems have to explore complex solutions that ensure both functionality and maximum up-time. We describe a methodology that supports the designers in this task. Specifically, we introduce a model-based approach that computes both the diagnosability of a system
and the set of hypothetical sensors needed in order to find the root cause of any of the system’s failures. The methodology starts at design time, by creating behavioural models for the replaceable parts of the system. These models specify both the expected behaviour and possible Failure Modes (FMs) of the replaceable parts. Using these models, the system design is composed, with the individual replaceable part behaviours defining the system’s behaviour. To create these models we use a domain-specific language that generates a Bayesian Network that computes the failure symptoms, i.e., readings on a given sensor configuration, for every FM in the system. Finally, we perform the diagnosability analysis by determining FMs for which the symptoms are equal, causing them to be unidentifiable. For the unidentifiable FMs, we
compute a set of hypothetical sensors needed to ensure full diagnosability and the corresponding sensor readings to differentiate between the failures. This information is then used by the designer to make system design trade-offs. We illustrate our approach on two sub-systems of a high-tech machine.
and the set of hypothetical sensors needed in order to find the root cause of any of the system’s failures. The methodology starts at design time, by creating behavioural models for the replaceable parts of the system. These models specify both the expected behaviour and possible Failure Modes (FMs) of the replaceable parts. Using these models, the system design is composed, with the individual replaceable part behaviours defining the system’s behaviour. To create these models we use a domain-specific language that generates a Bayesian Network that computes the failure symptoms, i.e., readings on a given sensor configuration, for every FM in the system. Finally, we perform the diagnosability analysis by determining FMs for which the symptoms are equal, causing them to be unidentifiable. For the unidentifiable FMs, we
compute a set of hypothetical sensors needed to ensure full diagnosability and the corresponding sensor readings to differentiate between the failures. This information is then used by the designer to make system design trade-offs. We illustrate our approach on two sub-systems of a high-tech machine.
TNO Identifier
981809
ISBN
978-1-936263-34-9
Source title
Proceedings of the 6th European Conference of the Prognostics and Health Management Society 2021
Editor(s)
King, S.
Fink , O.
Fink , O.
Collation
9 p.