Research Data Management for Bespoke Experiments in Mechanical Engineering - Semantic digital data sheets, automated configuration and documentation, and reproducible data storage

When conducting one-of-a-kind experiments with bespoke test rigs using custom software, there are innumerable degrees of freedom for building a Research Data Management (RDM) framework. Thus, the challenge lies in developing generalisable approaches that use commonplace tools and interoperable technologies, and which give guidance among the available options. This contribution presents a set of approaches and technologies for addressing the challenge of RDM for bespoke experiments in mechanical engineering. This set of approaches addresses the issues of: (i) documenting the properties of sensors and actuators that are assembled in the test rig; (ii) using a machine- and human-readable technology for configuring an experimental run with automatic self-documentation; (iii) storing the resulting data in an interoperable format together with all the documentation for reproducing that run as well as further context information. Regarding (i), the presented work uses semantic digital data sheets as proposed in [1] and operationalised by Rexer et al. These data sheets store component information in the form of semantic graphs using standardised vocabularies from common ontologies. The work of Rexer et al. was extended beyond sensors, test rigs and substances to actuators and sensor-actuator combinations. The data sheets are stored on a public repository on the GitLab instance of RWTH Aachen, accessible through the GitLab API. For (ii), files in the JavaScript Object Notation (JSON) format are used. These are passed to the test rig operating software, which subsequently retrieves the information from the file for automatically configuring and running the experiment. The data in the recipe file is validated, avoiding faulty user input or infeasible configurations. The file remains available for documenting an experimental run and should be associated with the resulting data. Concerning (iii), the resulting data is stored in a Hierarchical Data Format (HDF5) file together with the content of the recipe file, information on the system on which the experiment was run and serialisation of the software code used for running the experiment. This approach was presented by Inturri et al. with the library Co-Simulation Of Functional Mock-up Units (FMU) with Integrated Research Data Management (SOFIRpy). The library allows loading runs from the file, manipulating the settings and rerunning them, leading to fully reproducible data. The documentation is extended by joining the HDF5 file with a metadata file, following the Research Object Crate (RO-Crate) standard. The presented solutions are implemented and validated on a test rig for studying the resilience of urban water distribution systems. The test rig uses the National Instruments (NI) Compact Data Acquisition system for measurement and is operated using a custom-written Python library, which relies on the Python API of NI and uses the framework of SOFIRpy. The recipe file is used for orchestrating the entire measurement setup: retrieving the semantic digital data sheets, instantiating classes for sensors and actuators, instantiating classes for the experimental setup and configuration. Before the orchestration, the recipe file gets validated using Pydantic. The framework for running the experiment is an adaptation of the SOFIRpy library. This allows storing the resulting experimental data according to the schema presented in [3]. To summarise, this contribution proposes standardised semantic digital data sheets for components, human- and machine-readable configuration files for configuring and documenting experimental runs, and a comprehensive data storage concept for reproducible experimental data. The proposed approaches have been validated using a physical test rig.