MATHMET

The European Centre for
Mathematics and Statistics in Metrology

Metrology for the Factory of the Future

An EMPIR project proposal for TP Industry 2017
Share
Stacks Image 15

To remain competitive, European industrial production and assembly lines have to be transformed following the concept of the “Factory of the Future” (FoF) – a manufacturing environment of inter-connected devices and with an autonomous flow of information leading to automated decisions. This requires reliable and formally described information about the manufacturing processes, derived from trustworthy measurement data at every stage of data analysis. Currently there is no corresponding metrological infrastructure to support these developments. An international metrology, measurement and data analysis infrastructure is needed to support the digitisation of European industry.

The Proposal

The proposed project consists of four research areas, which together aim at establishing a metrological treatment of measurement data in modern connected and smart factories. This contains the calibration of sensors with digital, pre-processed output and the incorporation of calibration information as well as other data quality issues in the treatment of industrial sensor networks. As the data from such sensor networks is often analyzed using machine learning algorithms, the incorporation of data quality for such algorithms has to be investigated, too. Finally, in order to demonstrate usability of the developed procedures and frameworks, actual implementation in testbeds will be realized.
Calibration framework for sensors with digital pre-processed outputs
Many measurement systems in Factory of the Future (FoF) environments provide only digital output of pre-processed data and calibration information is typically very sparse or available only on a sensor-type basis. However, reliable information about the sensors’ measurement capabilities is necessary to evaluate data quality. This can be addressed by developing a calibration framework for distributed networks of sensors with digital outputs and internal signal processing. The ability to extrapolate measurement uncertainty derived from individually calibrated sensors to other individual sensors of the same type is required. As measurement environments are complex and usually dynamic, corresponding dynamic calibration methods and identification of dynamic parameters of the production process are needed that allow the metrologically sound quantification of uncertainty contributions due to dynamic effects.
Metrological treatment of industrial sensor networks
A basic feature of FoF environments is the aggregation of data from various sources of information that vary over time and the adoption and extension of industrial internet technologies. The poor synchronization of measurements and the lack of precision of the underlying time axis in such scenarios can be addressed by the development of data aggregation and harmonization methods with quantitative assessment and propagation of data quality in sensor networks that take into account these limitations. Partial redundancy of sensors can be exploited to reduce effects of ambient conditions, especially temperature, but also to assess the performance of individual sensors. This allows identification of defective sensors and, ideally, replacing faulty data by data estimated from other sensors in the system to increase the overall data quality and, thus, system robustness and performance. Several low-cost sensors might provide better performance and reliability than one expensive sensor, but this concept also needs to be integrated into the required metrology framework. The increasing availability of versatile sensors and, thus, potential points of measurement lead to challenging questions regarding the definition of required measurement coverage and accuracy to meet process output quality targets. Software and metrological reference models are needed to allow for optimization of the measurement infrastructure and the identification of critical input sources.
Machine learning techniques in industrial sensor networks
Owing to the volatility of data and information in FoF scenarios, data analysis has to be carried out in real-time or close to real-time. Typically, machine learning methods are employed to this end, which extract information from data online, e.g., using statistical methods or artificial neural networks. Despite their wide use and successful application in various examples, there is a lack of methodologies for the assessment of the reliability and quality of such methods in a metrological acceptable way. The required data analysis methods should take into account individual sensor limitations, quantitative evaluation of data and information quality and consider self-calibration, which can reduce uncertainty and increase trust in data through corroboration of other measurands in the network. As the amount of data in industrial sensor networks can easily be considered as big data, analysis methods have to scale well with the number of sensors and provide performance and reliability statements for single sensors as well as for large sensor networks.
Concepts and testbed implementations to accelerate actual up-take by industry
Testing the practicality of the approaches identified in the previous three objectives requires the provision of test-beds and factory-like environments in which the new techniques can be demonstrated in action and to allow trouble-shooting and validation prior to commissioning new networks and robot systems. This includes the development of methodologies of gradually improving existing infrastructures. Synchronization with international consortia such as the “Industrial Internet Consortium” (IIC) and corresponding national initiatives and the adoption and extension of common technologies such as OPC UA, oneM2M or Semantic Web are needed to ensure the up-take of the developed metrological framework by a wide range of industries. The integration of the metrological framework in industrial standards requires a high-level, abstract formulation of data protocols and sensor network characterisations including data quality.

Background information

Smart sensors, connected and distributed manufacturing, automated process control and intelligent data analysis are becoming the de-facto standard in modern industrial environments. Concepts titled as, for instance, "Industrie 4.0", "Factory du Future" or "Factory of the Future" envision a manufacturing landscape in which the whole process - from first designs to the actual use and maintenance - is connected. The general aims are to improve resource efficiency, decrease development and production time, allow for small lot sizes in mass production and to develop new innovative business models based on intelligent data analysis and reliable communication of information.

Metrology is an essential part of the backbone of today's quality infrastructure with the provision and further development of high-precision measurements, calibrations and conformity assessment as well as standardization and harmonization. These key areas are facing enormous challenges resulting from the digital transform in industry:
  • New algorithm-based manufacturing methods, such as additive manufacturing, and the fast development of 5G communication technologies require new measurement capabilities in order to enable traceability.
  • Digitally transformed industries require machine-readable, automated certification processes.
  • Intelligent sensors with pre-processed data output require new frameworks and standardization to enable traceability and allow quality assessments.
  • Automated data communication and interpretation requires reliable knowledge about dimension, unit, type and quality of the data being transferred.
  • Intelligent data analysis methods, such as deep neural networks, need to be assessed regarding their reliability in order to gain trust in their results.

The proposed project in this regard aims at bridging "traditional" metrology and the concepts of a "factory of the future", in which data from large networks intelligent sensors is analyzed and used automatically. For the individual sensors methodologies are required to calibrate their digital pre-processed outputs with the aim to utilize the built-in computing power of today's sensors to allow for an "advanced traceability", for which calibration information is used reliably on the sensor itself and the corresponding (dynamic) uncertainties are being communicated together with the processed values. Based on such a framework a corresponding data communication protocol is to be developed which takes into account the communicated sensor uncertainties, network communication issues and allows for reference sensors in the network for "self-calibration". With industrial sensor networks reporting data quality (e.g. uncertainties), data analysis methods could take advantage of this information in order to produce more reliable results. Therefore, an assessment of the uncertainty propagation abilities of the machine learning tools in use is required.

The Consortium

EMPIR project proposals are carried out in two main stages: First, a proposed research topic (PRT) is submitted and evaluated by the EURAMET committee. In the second stage, for the selected research topics (SRTs) detailed proposals for joint research projects are written. The consortium for the second phase will be established at the partnering meeting 29-30 June 2017 in Berlin. Information on how to register can be found here.
The consortium of the first stage, namely the co-authors of the PRT (download here), were:
  • from Germany: PTB, HBM, Endress+Hauser, Imagineon, Univ. Saarland, TU Ilmenau, Univ. Kassel, Fraunhofer FOKUS, Fraunhofer IPK
  • from France: LNE, CEA-LIST-LADIS
  • from the UK: NPL, Strathclyde Univ., Cambridge Univ.
  • from Finland: VTT, Mapvision, Aalto Univ.
  • from Italy: INRIM, SPEA, CSP
  • from the Netherlands: VSL, NEN, TNO, Sioux LIME
  • from Turkey: TUBITAK UME
  • from Romania: Romanian Measurement Society
  • from the USA: National Instruments
  • from Belgium: Vrije Univ. Brussel

The EMPIR Programme

The European Metrology Programme for Innovation and Research (EMPIR) is a co-funded initiative by the European Union's Horizon 2020 research and innovation programme and the EMPIR participating states. For 2017 the initiative has open calls for Industry, Fundamental, Normative, Research Potential and Support for Impact. Just recently, EURAMET published the list of selected research topics (SRTs) based on which until 2nd October applications for joint research projects (JRPs) can be submitted. Interested parties can register themselves at the EURAMET website: msu.euramet.org/calls.html

Joint research projects in EMPIR are coordinated by a staff member of a partner NMI (national metrology institute) elected during the partnering meetings. The project duration is 36 months and average funding for the Industry Call 2017 is approx. 2.8 Mio Euros, from which 30% is for external (non NMI/DIs) partners. Interested parties can join the project either as funded partner (if eligible), collaborator or unfunded partner. Details about funding and types of collaboration can be found here.
 

Interested? Get in contact:

The next step is the preparation of a Joint Research Proposal (JRP) until the end of September 2017. If you are interested in this process or want to provide us your feedback on your specific requirements, ideas and suggestions - please let us know!
x
This website uses cookies occasionally to provide you with the best web browsing experience. However, no web-analytics tracking based on cookies is employed here.