WP1: Calibration framework for distributed sensors with digital preprocessed output
Traceability for digital sensors in dynamic measurements and practical MEMS calibration services is currently completely lacking NMI-level or accredited calibration services. This is a consequence of the lack of adequate technical set-ups and procedures as well as normative standards.
The challenge in dynamic calibration is that traditionally, dealing with analogue signals, the time keeping was managed by the calibration system. Such centralised time management was the enabling technology for synchronous sampling and hence, calibration of phase-response. The autonomous timing of digital sensors is currently impeding the phase response characterisation and requires new concepts for the full characterisation of the dynamic response of sensors with digitally pre-processed output are needed.
The current challenge for in-situ MEMS calibration is the establishment of traceability to SI units whilst maintaining the overall efficiency of the automated testing. This work package will address this by developing an optimised fixture (i.e. framework) of reference sensors.
- In Task 1.1 sensors and interfaces compatible for the testbed measurement tasks (as part of WP3) and the existing calibration facilities will be determined and selected.
- The aim of Task 1.2 is then to digitally enhance the existing dynamic calibration set-ups of project partners involved whilst maintaining uncertainties in the range 0.1 % and 1 % and measurement ranges from DC up to 10 kHz as is currently achieved for analogue dynamic calibration measurements.
- Based on the outputs of Task 1.1 and the mathematical and statistical frameworks developed in EMRP IND09, the aim of Task 1.3 is to develop a prototype of a digitally enhanced sensor for dynamic measurements, realising “Smart Traceability”.
- The objective of Task 1.4 is to develop a calibration framework to provide measurement traceability in industry in ATE systems for MEMS temperature sensor testing. Target temperature ranges are from -60 °C to 200 °C.
WP2: Metrological infrastructure for the industrial sensor networks aggregation of data from industrial sensor networks
The overall aim of this work package is to take the outputs of WP1 and show how the digital sensors that were the subject of WP1 can be combined into networks to produce either aggregated measurements (a single measurement value, its associated uncertainty and confidence interval, derived from a number of measurements) or a distributed measurement (how quantities of interest, e.g. environmental quantities such as temperature are distributed in space and time). Different measurement models are required for aggregated and distributed measurements, as the measurement aims, and therefore the measurands are different.
The key challenge is to address synchronisation of measurements, make use of redundancies of measurements at ambient conditions, balance the potential of multiple low-cost sensors versus the performance and reliability of a few expensive sensors, and explore methods to identify the measurement coverage and accuracy required for process output quality targets, taking into account the complexity of the system, and to develop methods for assessing the quality, reliability and accuracy of ML methods in metrological sensor networks. Therefore, in a first step the process output quality targets for selected measurement scenarios with the SPEA, ZEMA and STRATH testbeds will be identified.
Agent based software frameworks are ideal to drive autonomous data capture and analysis in sensor-based networks because they provide a distributed, scalable, persistent environment that is easily reconfigurable based on the trade-offs obtained.
Using the agent-based approach it will be shown how the sensor network design and uncertainty reduction approaches can be used for condition monitoring in the FoF. The essential element of ML (ML) methods is the design of feature vectors by data pre-processing, feature extraction and feature selection. In addition to the data-driven ML approaches, a model-based method provided by ITRI will also be considered in this work package to extract training data.
The project will demonstrate these methods in action on the testbeds that are established as part of WP3. In particular, in collaboration with the testbed owners, the project will define target process outputs from the testbed that demonstrate an improvement on existing testbed performance, from which all future users of the testbed will benefit. As far as possible all methods we develop will align with the requirements of the GUM.
- Task 2.1 sets out the mathematical modelling required to approach sensor network metrology for the testbeds considered in Tasks 3.1-3.3 and lays the foundation for the rest of WP2.
- Task 2.2 extends the mathematical models from Task 2.1 to take into account potential timing issues by introducing additional input quantities to the mathematical model.
- Task 2.3 extends the mathematical models from Task 2.1 and Task 2.2 to situations where redundant information is available for aggregated or distributed measurement data.
- Task 2.4 extends the mathematical models from previous tasks with statistical features to develop a framework for the treatment of sensor networks with mixed quality.
- Task 2.5 joins together the mathematical models from Task 2.1 – 2.4 in an agent-based software framework for an efficient implementation of ML methods. The task will then assess and develop ML methods that take into account uncertainty associated with the sensor output data.
WP3: Industrial testbed implementations and facilitation of end-user uptake
The aim of this work package is to demonstrate the applicability of the methods and concepts developed in WP1 and WP2 both in industry-like environments and in an industrial premise. Moreover, the goal is to demonstrate the benefits provided by improved dynamic calibration and testing methods to relevant manufacturing processes.
Therefore, this work package considers three industrial application examples: process optimisation, condition monitoring and sensor network measurements. Each example is addressed in an individual task using an individual existing facility.
- Task 3.1 applies the sensor network setup of references temperature sensors, developed in Task 1.4 in an actual MEMS temperature calibration testbed at SPEA.
- Task 3.2 applies the ML methods developed in Task 2.5 and the calibrated digital sensors from Task 1.2 in the ZEMA testbed for condition monitoring of electromechanical cylinders.
- Task 3.3 applies the calibrated digital sensors from Task 1.2 and the sensor network and data aggregation methods from Task 2.5 to perform a process optimisation for the inductive heating in the STRATH testbed for radial forging.
All three testbeds have been developed prior to this project with the aim to support industry up-take of FoF methods. This project will extend these testbeds to a metrological infrastructure. For all testbeds, selected measurement data will be made available to the scientific community to foster the development of ML methods in metrology.