=Paper=
{{Paper
|id=Vol-2721/paper516
|storemode=property
|title=Semantic ML for Manufacturing Monitoring at Bosch
|pdfUrl=https://ceur-ws.org/Vol-2721/paper516.pdf
|volume=Vol-2721
|authors=Baifan Zhou,Yulia Svetashova,Tim Pychynski,Evgeny Kharlamov
|dblpUrl=https://dblp.org/rec/conf/semweb/ZhouSPK20
}}
==Semantic ML for Manufacturing Monitoring at Bosch==
https://ceur-ws.org/Vol-2721/paper516.pdf
Semantic ML for Manufacturing Monitoring at Bosch
Baifan Zhou1,2 , Yulia Svetashova1,2 , Tim Pychynski1 , and Evgeny Kharlamov3,4
1
Bosch CR, 2 Karlsruhe Institute of Technology, 3 Bosch Center for AI, 4 University of Oslo
Motivation. Technological advances that come with Industry 4.0, in e.g. sensoring
and communication, unlock unprecedented large volumes of manufacturing data. This
opens new horizons for data-driven methods like Machine Learning (ML) in analysis of
manufacturing processes for a wide range of industries. An important scenario here is
monitoring of manufacturing processes, including e.g. analysing the quality of the man-
ufactured products and predicting the health state of machines and equipment. Consider
an example of welding quality monitoring at Bosch, where welding is performed with
automated machines that connect pieces of metal together by pressing them and passing
high current electricity through them. Development of ML approaches for welding qual-
ity monitoring used in Bosch follows an iterative workflow that includes data collection
(Step 1), task negotiation (Step 2), data preparation (Step 3), ML model development
(Step 4), result interpretation and model selection (Step 5), model deployment (Step 6).
Challenges. Development of such ML approaches is complex and costly and at Bosch
it is mostly affected by the following 3 challenges:
– Transparency: Steps 2 and 5 of welding quality monitoring require collaborative
work of experts from different areas. The asymmetric knowledge backgrounds, in-
cluding complexity of engineering practices in manufacturing and sophistication of
ML algorithms that constrain the transparency of ML results and models, make the
communication time consuming and error-prone.
– Data preparation: Step 3 requires to integrate data from dozens of sources and this
is a labor-intensive effort that requires necessary understanding of multi-faceted
domain knowledge and plentiful data complications.
– Generalisability of ML quality models: each ML model developed in Step 4 is
typically tailored to a specific dataset and one process and thus its reuse for other
data or processes, which is often needed, requires a significant effort.
Semantic Enhancement of ML Development. In order to address these challenges we
propose to rely on semantic technologies to enhance ML pipelines that are based on
feature engineering and developed a system SemML [3] that implements our ideas (see
Fig. 1). The core of our approach is to incorporate domain, e.g. welding, and machine
learning knowledge in the ML development in such a way that it allows us to automate
data integration, ML modelling, and improve model explainability and generalisability.
In particular, we capture the domain knowledge as ontologies (Fig. 1) and rely
on two high level ontologies: Core that captures high level manufacturing knowledge,
e.g., of discrete manufacturing processes, and ML that captures ML aspects like feature
groups, feature processing and ML algorithms. We also rely on a set of specific ontolo-
gies of two categories: Domain and ML Pipeline that focus on particularities of specific
manufacturing processes, e.g., welding, and specific ML-pipelines for such processes.
Moreover, we developed Manufacturing and ML templates that allow to encode domain
knowledge (see the Knowledge encoding box in Fig. 1) by constructing and extending
Domain and ML-pipeline ontologies in accordance to the Core and ML ontologies.
”Copyright c 2020 for this paper by its authors. Use permitted under Creative Commons License
Attribution 4.0 International (CC BY 4.0).”
Fig. 1: Overview architecture of our approach and the SemML system.
By reasoning over ontologies we automate ML modelling. In particular, when weld-
ing experts annotate raw welding data with features from the Domain (see Data annota-
tion box in Fig. 1), the reasoner derives from these features the relevant feature groups,
and then with the help of the ML Pipeline it derives relevant feature processing algo-
rithms, and relevant ML algorithms. Moreover, we use domain ontologies to annotate
resulting ML models and their selected features, thus enhancing their explainablility.
Evaluation at Bosch. We implemented and evaluated our approach at Bosch for weld-
ing quality monitoring and prediction. The purpose was to enable predictive measures
such as equipment (re-)configuration or maintenance and thus to prevent welding qual-
ity failures. The evaluation was done in the offline mode on two car manufacturing lines.
These lines generated large volumes of heterogeneous data such as historic data with
sensor measurements, welding machine configurations, manufacturing specifications,
and the quality estimates of finished welding operations. The raw data came in SQL,
Excel, plain text files, RUI-files and resulted in 263 fields with 53.2 million records, and
1.4 billion items after data integration. The accuracy of welding quality prediction by
ML models constructed with the help of SemML is good: the mean absolute percent-
age error is within 1.61%–2.50% comparing to 3.19%–7.74% for simple baselines [2].
Moreover, we evaluated the usability of our system [1] with very promising results:
Bosch engineers were able to construct domain and ML-pipeline ontologies relatively
fast with minor training and gave high overall usability scores to our system.
Outlook. Our current results on semanticfication and thus simplification and automa-
tion of ML model development for monitoring of manufacturing processes are promis-
ing. At the same time there is still a number of important steps to be done to bring them
the necessary degree of maturity and deploy them in production at Bosch by integrating
them into a new welding control system and automation platform. We plan to further
develop our ML pipelines with more methods, e.g. Feature Learning, ARIMA, and con-
duct more extensive study specifically focusing on analysis of input feature importance.
We also plan to evaluate all our ML pipelines on scenarios different than welding.
References
1. Svetashova, Y., Zhou, B., Pychynski, T., Schmidt, S., Sure-Vetter, Y., Mikut, R., Kharlamov,
E.: Ontology-Enhanced Machine Learning Pipeline: A Bosch Use Case of Welding Quality
Monitoring. In: ISWC (2020)
2. Zhou, B., Svetashova, Y., Byeon, S., Pychynski, T., Mikut, R., Kharlamov, E.: Predicting
Quality of Automated Welding with Machine Learning and Semantics: a Bosch Case Study.
In: CIKM (2020)
3. Zhou, B., Svetashova, Y., Pychynski, T., Kharlamov, E.: SemFE: Facilitating ML Pipeline
Development with Semantics. In: CIKM (2020)