Maintworld 1/2020
ROTATING EQUIPMENT SERVICES: A COMPREHENSIVE, WORRY-FREE PACKAGE // SELF-INFLICTED RELIABILITY PROBLEMS OF ROTATING MACHINERY // VIEWING MAINTENANCE AS A SYSTEM TO OPTIMIZE PERFORMANCE
ROTATING EQUIPMENT SERVICES: A COMPREHENSIVE, WORRY-FREE PACKAGE // SELF-INFLICTED RELIABILITY PROBLEMS OF ROTATING MACHINERY // VIEWING MAINTENANCE AS A SYSTEM TO OPTIMIZE PERFORMANCE
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
TECHNOLOGY<br />
Industrial AI’s capacity to analyze very large amounts of<br />
high-dimensional data can change the current maintenance<br />
paradigm and shift from preventive maintenance systems to<br />
new levels. The key challenge, however, is operationalizing predictive<br />
maintenance, and this is much more than connecting<br />
assets to an AI platform, streaming data, and analyzing those<br />
data. By integrating conventional data such as vibration, current<br />
or temperature with unconventional additional data, such<br />
as audio and image data, including relatively cheap transducers<br />
such as microphones and cameras, Industrial AI can enhance<br />
or even replace more traditional methods. AI’s ability to predict<br />
failures and allow planned interventions can be used to<br />
reduce downtime and operating costs while improving production<br />
yield. For example, AI can extend the life of an asset beyond<br />
what is possible using traditional analytics techniques by<br />
combining data information from designer and manufacturer,<br />
maintenance history, and Internet of Things (IoT) sensor data<br />
from end users, such as anomaly detection in engine-vibration<br />
data, images and video of engine condition. This information<br />
fusion during the lifecycle of the asset is called product lifecycle<br />
management (PLM).<br />
Explainable AI in Maintenance<br />
Advances in AI for maintenance analytics are often tied to<br />
advances in statistical techniques. These tend to be extremely<br />
complex, leveraging vast amounts of data and complex algorithms<br />
to identify patterns and make predictions. This<br />
complexity, coupled with the statistical nature of the relationships<br />
between input data that the asset provides, makes them<br />
difficult to understand, even for expert users, including the<br />
system developers, Figure 4. This makes explainability a major<br />
concern.<br />
Figure 4: Engineers and data scientist must co-create the AI<br />
solution for maintenance together<br />
While increasing the explainability of AI systems can be<br />
beneficial for many reasons, there are challenges in implementing<br />
explainable AI. Different users require different<br />
forms of explanation in different contexts, and different contexts<br />
give rise to different needs. To understand how an AI<br />
system works in the maintenance domain, users might wish to<br />
know which data the system is using, the provenance of those<br />
data, and why they were selected; how the model and prediction<br />
work, and which factors influence a maintenance decision;<br />
and why a particular output is obtained. To understand what<br />
type of explanation is necessary, careful stakeholder engage-<br />
ment and well-thought-out system design are both necessary<br />
as can be seen in figure 5.<br />
Figure 5: Architecture of explainable AI systems for Maintenance<br />
Decisions<br />
There are various approaches to creating interpretable systems.<br />
Some AI is interpretable by design; these systems tend to<br />
be kept relatively simple. An issue with them is that they cannot<br />
get as much customization from vast amounts of data as<br />
more complex techniques, such as deep learning. This creates a<br />
performance-accuracy trade-off in some settings, and the systems<br />
might not be desirable for those applications where high<br />
accuracy is prized. In other words, maintainers must accept<br />
more black boxes.<br />
In some AI systems – especially those using personal data or<br />
those where proprietary information is at stake – the demand<br />
for explainability may interact with concerns about privacy. In<br />
areas such as healthcare and finance, for example, an AI system<br />
might be analyzing sensitive personal data to make a decision<br />
or recommendation. In determining the type of explainability<br />
that is desirable in these cases, organizations using AI will<br />
need to take into account the extent to which different forms<br />
of transparency might result in the release of sensitive insights<br />
about individuals or expose vulnerable groups to harm.<br />
In the area of maintenance, when the AI recommends a<br />
maintenance decision, decision makers need to understand<br />
the underlying reason. Maintenance analytics developers need<br />
to understand what fault features in the input data are guiding<br />
the algorithm before accepting auto-generated diagnosis<br />
reports, and the maintenance engineer needs to understand<br />
which abnormal phenomena are captured by the inference algorithm<br />
before following the repair recommendations.<br />
One of the proposed benefits of increasing the explainability<br />
of AI systems is increased trust in the system. If maintainers<br />
understand what led to an AI-generated decision or recommendation,<br />
they will be more confident in its outputs. But the<br />
link between explanations and trust is complex. If a system<br />
produces convincing but misleading explanations, users might<br />
develop a false sense of confidence or understanding. They<br />
might have too much confidence in the effectiveness or safety<br />
of systems, without such confidence being justified. Explanations<br />
might help increase trust in the short term, but they do<br />
not necessarily help create systems that generate trustworthy<br />
outputs or ensure that those deploying the system make trustworthy<br />
claims about its capabilities.<br />
REFERENCES:<br />
Galar, Diego, Pasquale Daponte, and Uday Kumar. Handbook of Industry 4.0 and SMART Systems. CRC Press, 2019.<br />
Galar, Diego. Artificial intelligence tools: decision support systems in condition monitoring and diagnosis. Crc Press, 2015.<br />
Galar, Diego, Uday Kumar, and Dammika Seneviratne. Robots, Drones, UAVs and UGVs for Operation and Maintenance. CRC Press, <strong>2020</strong>.<br />
50 maintworld 1/<strong>2020</strong>