With the explosion of big data technologies (BD), the possibility to integrate those tools into daily company operations is more a ordable and straightforward. On the other hand, Knowledge-based approaches such as graphs and di erent semantic approaches, although those have not been so popular in the industry in the past years, nowadays, thanks to the high availability of heterogeneous data inside of the company context, those tools are being used more to enhance or enrich data and processes, and make more informed decisions about the business. The SEMT platform is presented; this system combines a Big Data recollection approach from a legacy/manual sensor environment to perform a knowledge enhancement process to support the semi-conductor development and production operations inside a cleanroom.
The SEMT (Sensor Environment Monitoring Tool) proposed here was deployed in the company Leonardo SpA, in Rome headquarters, for monitoring a manufacturing environment, i.e., cleanroom. Cleanrooms are controlled R+D closed manufacturing environments that deliver controlled levels of particles, temperature, and other climate-related characteristics. The goal was to collect and analyze sensor environmental data over a production line, to support the cleanroom quality control process and the production of semi-conductors on this facility.
For these heterogeneous domains, semantic tools, such as Knowledge Base
approaches [
D. Rincon-Yanez et al.
as status reports or supply chain data. But, later arisen pattern discovery,
interrogating the system to obtain contextual answers related to the status of a
product using, e.g., SPARQL queries [
(b) Knowledge-Base Schema The system was designed to extract environmental data from heterogeneous sensors scattered in di erent laboratory segments. The general approach in this architecture was to create a data microservices-based supply chain; those services are divided into a set of layers grouped by functionalities at a higher level see Fig. 1a. All the components use HTTP REST messages allowing transparent and standard message interchange between the services in general. Storage layer provides persistence to the data, descriptors for sensors, collected data, and semantic data, The Connectors layer interacts directly with the device and his data; a connector for each device type should exist. The Scheduled Extraction layer provides the ability to coordinate the extraction from the sensor network, Visualization layer acts as a webapp for displays the dashboards, data reports, and device management interfaces. MQTT IoT layer can forward the recollected data via MQTT to the Leonardo 4.0 platform; the Semantic Layer will be debriefed in section 3. 3
Some of the most compelling challenges at the design phase on this system
were: (a) the inherent need to maintain historical data in an arbitrated way;
for later trace, product quality control, and (b) usage of temporal data stored,
both immediately and distant. For attending to these requirements, a simple
adaptation of a contextual sensor network ontology [
A long-term purpose is to consider a later knowledge-discovery process to combine this data with product-related data, sta access and activity logs, and other context-aware relevant data. Initially, in this same way, the semantic layer is also used to query the temporal stored data to perform reports and predictive analysis on the di erent cleanroom internal locations, aiming to evaluate and assist a risk management procedure that could compromise the product viability. 4
This work's main intention was to develop the closest to an industry-grade solution that brings value into an existing production line implementing cutting-edge approaches and delivers a product complying with the standards of industry 4.0.
As controlled and restricted environments, cleanrooms require fast and e cient technologies to deliver real-time, contextual, and aware data of the room's reality, enabling further monitoring, tracing, controlling, and predicting product characteristics. With the development of this platform, it was possible to demonstrate that a fast semantic-data ecosystem can be set up in restrictive and productive spaces and contribute with an added value to the process. Acknowledgements This research was partially supported by the MIUR (Ministero dell'Istruzione dell'Universita e della Ricerca) under the national program PON 2014-2020, Leonardo 4.0 (ID ARS01 00945) research project in the area of Industry 4.0.