The “Factory of the Future” (FoF) as an inter-connected production environment with an autonomous flow of information and decision-making constitutes the digital transformation of manufacturing to improve efficiency and competitiveness. Transparency, comparability and sustainable quality all require reliable measured data, processing methods and results. This project will establish a metrological framework for the complete lifecycle of measured data in industrial applications: from calibration capabilities for individual sensors with digital pre-processed output to uncertainty quantification associated with machine learning (ML) in industrial sensor networks. Implementation in realistic testbeds will demonstrate the practical applicability and provide templates for future up-take by industry.
Traceable calibrations, harmonised treatment of measurement uncertainties, and industrial standards and guidelines are the major components of a comprehensive metrological infrastructure that has enabled globalised manufacturing and international trade. Digitalisation and data science are rapidly changing almost all aspects of this landscape: e.g. sensors are becoming smart, large networks of sensors are being used together with ML algorithms to make automated decisions and manage production processes. The combination of these technological elements constitutes the FoF, a paradigm that is evolving rapidly worldwide.
According to the 2016 UK “Workshop on Data Metrology” and other recent surveys, one of the top priority industrial needs in the FoF is data quality. This project addresses the need for data quality interpreted as the need for a measurement uncertainty framework supporting a metrological infrastructure. In order to address the complete flow of information this infrastructure has to cover traceable calibration of smart sensors taking into account dynamic effects, metrological treatment of complex sensor networks and uncertainty evaluation for the data aggregation and decision-making methods. Previous projects developed the foundation of some of these aspects: EMRP IND09 established a metrological infrastructure for analogue dynamic measurement of mechanical quantities; EMPIR 14SIP08 implemented the mathematical methods from EMRP IND09 into software tools and guidelines for industrial end users; and EMRP ENG63 developed mathematical methods for sensor network metrology focusing on electrical power grids.
However the calibration facilities need to be extended to digital-only sensors, which requires new concepts to deal with the internal time keeping of sensors. Cost-efficient traceable calibration of Micro Electro Mechanical Systems (MEMS) sensors for ambient conditions is needed to associate their output with reliable uncertainties. Methodologies for sensor network metrology also need to be extended and real-time ML methods need to be developed to address uncertainty evaluation in industrial sensor networks.
The project contains three technical workpackages covering different aspects of the research and technical work to be carried out. The actvities in this work package are linked with each other via exchange of information, models, data, sensors, etc.
The Met4FoF project comprises the following main objectives:
- To develop calibration methods for industrial sensors of dynamic measurements such as acceleration, force and pressure with digital data output (data streams) and internal digital pre-processing, including the extrapolation of the measurement uncertainty from individually calibrated sensors to other individuals of the same type by means of co-calibration and statistical modelling.
- To develop and demonstrate methods enabling digital sensors to provide uncertainty and/or data quality information together with the measurement data.
- To develop a cost-efficient in-situ calibration framework for MEMS sensors measuring ambient temperature for their integration into an industrial sensor network with metrological quality infrastructure.
- To develop and assess data aggregation methods for industrial sensor networks based on machine learning and efficient software architectures, addressing synchronisation of measurements, making use of redundancies of measurements, taking into account uncertainty from calibration and network communication issues, including strategies for balancing cost versus uncertainty and explore methods to identify the measurement coverage and accuracy required for process output targets.
- To improve existing industry-like testbeds for sensor networks in manufacturing environments towards the implementation of a metrological quality infrastructure and to facilitate the take up of the project outputs by the stakeholders, especially the manufacturing industry.
The technical work also focuses on two very common challenges in manufacturing – process optimisation and predictive maintenance – represented by three specific testbeds using different types of sensor networks.
- The SPEA Automatic Test Equipment (ATE) for MEMS temperature sensor testing uses a network of reference temperature sensors, where the optimal implementation and usage of this sensor network determines the efficiency and reliability of the ATE results.
- The STRATH testbed considers radial forging using pre-heated metallic material and vibrating hammers. The testbed will be used to try and optimise the heating and forming process based on a range of different sensors in order to improve the production output quality.
- The ZEMA testbed uses a range of sensors measuring different quantities for end-of-line tests and condition monitoring methods for electromagnetic cylinders.
For all three testbeds, uncertainty in the whole flow of information, from the individual sensors to the data analysis output, will be considered consistently.