Assessing upfront the causes and e ects of failures is an important aspect of system manufacturing. Nowadays, these analyses are performed by a large number of experts. To enable semantic uni cation and easy operationalization of these risk analyses, this paper demonstrates an approach to automatically map the captured information into an ontology and accompanying rules. The approach is demonstrated with a use case to identify anomalies and their causes within a ventilation unit.
A FMEA is performed per system component to get a full view of all the possible failures that can occur, their e ects and their possible causes. This analysis results in an FMEA table, as visualised in Figure 1a. In contrast, a FTA identi es the link between a particular observation and a possible failure, as shown Component
Function
Failure Mode FEafifleucrte I Failure Cause O MCoenthtroodl DRPN Sensors are not recording
Box has lost connection
Data loss (a) (b)
Fig. 1: Example of a (a) FMEA and (b) FTA in Figure 1b. Our mapping approach allows to easily combine the two, by offering methods that automatically translate the FMEA tables and FTA trees to ontologies and accompanying rules that correlate the di erent observations, system components, faults and causes. Data generated by the systems can then be semantically annotated using these ontologies, while a reasoner can be used to automatically derive the correlated anomalies and faults with the generated rules. The full mapping approach is given in Figure 2. It consists of two main parts, domain knowledge transformation and rule generation.
The domain knowledge transformation process is shown on the bottom of
Figure 2. Entries from FMEA tables can be mapped to RDF using a mapping
language, resulting in a domain-speci c ontology with risk analysis information.
As such, anomaly knowledge can be extracted from the FMEA, and the causes
of these anomalies can be derived by following the semantic links.
To enable this, an upper ontology, called Folio1, was designed. It captures and
correlates all application-independent concepts that occur within FMEA tables
and AD methods, such as FailureCause, FailureEffect, Criticality and
DetectionMethod. The Semantic Sensor Network (SSN) ontology2 is used to
describe the various system components. Relationships were de ned in Folio to
correlate the SSN concepts with possible failures and e ects. The Folio ontology
functions as a blueprint for further domain-speci c ontology designs.
To translate the information inside the FMEA table, usually in the CSV format,
to ontological Folio concepts, i.e. RDF, the mapping language RML [
+ RMLEditor Folio Ontology RML Rules
SWRL rules Python
Evaluation RML Mapping Domain-specific
ontology that is created according to the standard FMEA structure. If a new column is added, a new rule can easily be created to map this column to the Folio ontology by using the RMLEditor. Updating or incorporating new information inside the tables can be performed without de ning additional RML rules. Rerunning the RML mapper su ces.
The rule generation process is visualized on the top of Figure 2. Observations generated by the system's sensors will be the main entry point for further analyses of unwanted behaviour. This analysis can be expressed as a FTA tree that determines how new observations should be interpreted in the context of failures and e ects. Classical FTA does not allow complex analysis of observations. Therefore, our approach supports an extension, namely decision fault trees (DFTs), which allow tests on the edges of the tree to get a combination and more complex interpretation of the system observation values. A user interface was designed to build such DFTs. In this editor, descriptions of the observation and failure nodes can be given. These di erent node concepts should align with the concepts de ned in the FMEA. Tests describing the relations between these observations and failures can be added or adapted. To transfer these tests to decisions, a script4 was constructed that derives Semantic Web Rule Language (SWRL) rules from the DFTs. Similarly to the construction of FMEA tables, additions and updates to the trees do not require changes to the scripts but rerunning them su ces.
The ontologies and rules generated by using the full translation approach can be incorporated in a knowledge-based monitoring system to identify anomalies and their causes continuously. The monitoring system can semantically annotate new observations using the domain-speci c ontologies, generated by mapping the FMEA tables. A reasoner can process the generated SWRL rules and links de ned in the ontologies to determine whether failures are occurring and what their possible causes are. 3
The Healthbox 3.0 is a ventilation unit for residential buildings, o ered by Renson Ventilation NV, which controls the air ow 24/7 to enhance indoor air quality. 4 Script: https://github.com/IBCNServices/Folio-Ontology/blob/master/swr builder.py It extracts polluted air from di erent rooms in a house. One central fan is responsible for creating the requested air ow. The Healthbox is equipped with a series of sensors of which the observations can be used to detect possible failures. When for instance the requested air ow could not be achieved, the valves could be malfunctioning, or the fan speed can be too low. The described approach was used to automatically generate ontologies and rules that capture the knowledge of the Renson experts for the detection of ventilation anomalies and their causes. Renson performed a FMEA analysis for each component of the Healthbox to get a full view on all the possible failures, their e ects and causes. The demo will show how the designed RML rules, see the previous section, can be used to easily map the contents of this FMEA to a Ventilation ontology, which is an extension of the SSN en Folio ontologies. The user will also be able to add new rows and column to the FMEA. It will be demonstrated how the RMLEditor can be used to map the new columns to the Folio ontology in a user-friendly manner. It will be shown that the new rows automatically get transformed to the ventilation ontology without requiring to adapt the mappings.
The designed tree editor was used by experts to construct DFTs that link the system observations to possible failures of the Healthbox. The example DFT describes a possibly malfunctioning CO2 sensor because the measured values are not in the acceptable range provided by the Renson experts. This sensor is important for the correct functioning of the valves, enabling the unit to ventilate when needed. It will be shown that this tree editor automatically exports JSON les that can be translated to SWRL rules by using the script detailed in the previous section. The user will be able to adapt or construct DFTs during the demonstration with the user-friendly editor. It will be shown that the script is also able to translate these new or adapted DFTs to SWRL rules. The demonstration ends by combining the Ventilation ontology and SWRL rules in Protege with some generated sensor observation instances. The reasoner is executed to illustrate that the rules extract the correct failures from these observations and link them to the possible causes through the ontology 5. Acknowledgment: This research is part of the imec ICON project Dyversify, co-funded by imec, VLAIO, Renson Ventilation NV, Televic Rail & Cumul.io. 5 The whole demo is also available on https://youtu.be/S3pe47Sn2Qs.