Examples of good practice

This website lists examples considered to represent good practice case studies that may help others in advancing use of AI in legal metrology. The DTG Subgroup on AI welcomes any suggestions for other examples; these can be submitted directly to the DTG Chairperson, Dr Sascha Eichstädt (PTB).

Utilizing large language models (LLM)

Optimized conformity assessment of measuring instrument software

A document analysis tool, which was developed by PTB and is regularly used to test complex software of measuring instruments that are legally regulated in Europe, has now been enhanced using artificial intelligence and prepared for the use of large language models.

Litzinger, J. Neumann, D. Peters, F. Thiel: Streamlining conformity assessment of software applying large language models. Measurement: Sensors, 101792 (2025), DOI: 10.1016/j.measen.2024.101792

AI-supported digital conformity assessment ("AI.Cert" project)

The aim of “AI.Cert” is to establish a conformity assessment supported by Artificial Intelligence (AI) at PTB, which will make internal processes more efficient. Areas of application for AI are, for example, the automatic checking of manufacturer documentation for completeness, the automatic creation of a checklist for product testing or the first draft of a certificate of conformity. This step is made possible by the development of Digital Certificates of Conformity (DCoC) by PTB. 

Contact: ai.cert@ptb.de 

Combining AI with legally controlled measurements

Auditable linking of AI image recognition and weighing (Japan)

Kamacho Scale Co., Ltd. has piloted a framework on a recycling sorting line in which AI image recognition identifies objects, while a weighing instrument verified on the basis of OIML R 76 provides the mass. For each event, the image, the mass indication, the time and the evidence of how image and mass were linked are stored as a hash-chained, tamper-evident record. The AI components have no feedback path into the weighing function, and AI-derived quantities are labelled and kept separate from the metrologically traceable mass value. The records serve both as an auditable log and as training data for site-specific AI models. The author explicitly proposes the record structure as a use case for the DTG guidance on AI.

T. Kamada: Preserving legal-metrology integrity in AI-assisted logging. 2026 OIML Bulletin LXVII(3) 20260305

Trust in digital retail weighing systems with AI-assisted product recognition (Czech Republic)

An analysis from the Czech Metrology Institute and WELMEC Working Group 2 shows that modern retail weighing systems have become distributed transaction systems, which may include product-recognition applications or AI-assisted product selection alongside the non-automatic weighing instrument, POS software and cloud services. A field case in which a replaced barcode scanner with new firmware led the POS system to process ten times the actual weight and price illustrates that components treated as legally non-relevant can affect the legally relevant result. The article proposes transaction-chain assurance based on WELMEC Guides 2.2, 2.10 and 7.2, OIML R 76 and OIML D 31 – an approach that equally applies to AI-based components in the chain.

D. Šťastný: Designing trust in digital retail weighing systems. 2026 OIML Bulletin LXVII(3) 20260306

Introductory scientific papers

• Schwabe et al. The METRIC-framework for assessing data quality for trustworthy AI in medicine: a systematic review. Npj Digital Medicine. DOI: 10.1038/s41746-024-01196-4

• Kuzin et al. Practical aspects of applying artificial intelligence in metrology. DOI: 1007/s11018-024-02285-2
• Kuzin et al. Artificial intelligence-based neuro-consultant in the field of legal metrology. DOI: 10.1007/s11018-025-02400-x
• Galyshev et al. Control of measurement reliability in technical monitoring instruments using a cascade of autoencoders. Vestn. YuUrGU. Ser. Vych. Matem. Inform., 14:4 (2025), 5–24.
• Shestakov et al. Hybrid CNN–MLP for Robust Fault Diagnosis in Induction Motors Using Physics-Guided Spectral Augmentation. https://doi.org/10.3390/a18110722