DSM at VSL: The Impact of Models, Algorithms and AI on Trust in Measurements

Blog

DSM at VSL: The Impact of Models, Algorithms and AI on Trust in Measurements

 

In almost every sector, decisions are made based on measurements – from emissions reporting and permits to energy bills and medical images. Behind those measurements lies a complex combination of models, software and – increasingly – AI to interpret measurement data. So the question is not only whether the measurement is correct, but also whether the surrounding modelling is reliable.

At VSL, the Data Science & Modelling (DSM) department focuses precisely on this field, where measurements, data and models together determine what is considered “true”.

Would you like to know more about our services?

Our experts are happy to help.

Adriaan van der Veen
Chief Scientist Chemistry and Data Science & Modelling

compass@vsl.nl

From Individual Measurements to a Coherent Overview

 

For DSM, measurements are only the starting point. The data‑science side of DSM focuses on what is visible in the data itself: statistical analysis of measurement series, quantification of measurement uncertainty, detection of trends and patterns, and identification of anomalies and drift over time. At VSL, data science is used exclusively in a metrological context that connects to the expertise of other departments. It always concerns data that originate from measurements – of cases like gas flows, electrical quantities, sensor networks and emissions – and not for cases like customer behaviour or phone usage. The questions are always the same: What do the measurements show? Within which uncertainty margins? And where might this point to systematic problems in practice?

The modelling side of DSM builds on this and describes the behaviour of measurement systems and methods in mathematical terms. This can be a single instrument, but also complete gas and electricity networks, or emission and dispersion processes around livestock farms. Models capture how signals, errors and uncertainties propagate through such a system, which assumptions are made, and how the whole behaves in different scenarios. Those models can in turn be tested against real measurements, for example using background measurements of ammonia in outdoor air or field data from networks.

Together, these two perspectives ensure that individual measurements become part of a consistent and traceable picture of reality. Data science helps derive reliable patterns and uncertainties from measurement data, and modelling links those patterns to the underlying situation. In this way, a shared basis emerges on which parties such as grid operators and permitting authorities can make well‑founded decisions.
This modelling goes further than simply drawing a smooth curve through data points. DSM explicitly examines the assumptions, the uncertainties and the connection to practice. It also looks at alternative models and algorithms so that an appropriate solution is chosen. Models must not only be mathematically correct, they must also stand up in the field.

A View of the Entire Chain

 

An important difference from other departments is that DSM looks at the entire system. A gas meter can be calibrated very precisely in the laboratory by VSL experts, but for the uncertainty of the total gas volume consumed over a year, other factors also play a role:

  • how often and in what way the meter is read
  • which assumptions are made in data processing
  • how variations in usage and gas quality are taken into account

This is when DSM becomes involved. Organisations do the straightforward calculations themselves; DSM is called in when everything has been done according to the guidelines and there is still friction between the model and reality, or when the calculations are simply very difficult and complex.

Always Practically Applicable

At VSL, we prefer to deploy DSM in practical applications that have a positive impact on society. For example:

  1. Emissions of Ammonia and Methane
    Together with the Chemistry department, DSM models emissions in livestock housing and the dispersion of, for example, ammonia around farms. Outdoor measurement facilities are used to determine background levels so that dispersion models can be properly tested. This is crucial in debates about the agricultural sector, where models and measurements sometimes appear to contradict one another.
  2. Electricity Networks and Grid Congestion
    In cooperation with the Electricity department, measurement data from networks are used to better understand and mitigate congestion. Models support decisions on capacity and flexibility, but only if their uncertainties and limitations are clearly defined.
  3. Gas Networks and Gas Quality
    In gas networks, variation in gas quality plays a major role for both measurement systems and gas consumers. DSM investigates how meters and other instruments respond to that variation, and what this means for the total uncertainty in volumes and for impacts such as annual billing. DSM also assesses whether consumers receive gas of the required quality. In both cases, the focus is explicitly on the step from accurate instrument calibration to behaviour in a complex, dynamic network.

In all these examples, DSM seeks a realistic balance between models that provide sufficient detail to support decisions, but which are also explicit about the limits of their validity.

Software as an Indispensable Link in the Metrological Chain

 

Where traditional metrology is often associated with hardware, such as meters or sensors, software is now at least as decisive for the final result of measurements. Especially in sensitive environments such as laboratories and hospitals, strict requirements apply to software validation. At VSL, DSM assesses, among other things, whether software and firmware:

  • Implement mathematical models correctly and in a stable manner;
  • Handle measurement signals in a reproducible way;
  • Does not contain any “hidden” assumptions or error‑handling routines that distort the outcome.

For manufacturers, firmware may be a side issue; for VSL, it is a fully fledged part of the metrological chain. Where necessary, DSM also develops its own solutions so that validated models can be implemented in practice and used correctly.

AI as a New Type of Measurement System

 

AI is increasingly playing a role in systems that we use to measure and assess. Machine‑learning models are used, for example, in gas and electricity networks to forecast consumption or detect faults. In the medical field, deep‑learning models support image analysis so that tumours and other abnormalities in tissue can be identified more quickly.

 

DSM does not build these AI models itself, but investigates whether their output is reliable, traceable and reproducible. Typical questions for DSM include: How sensitive is an AI model to variations in data? How do you ensure that a model does not gradually drift over time without being noticed? And how do you quantify uncertainty in AI outputs in a traceable way? These are the kinds of AI questions DSM addresses.

Trust as the End Goal

 

Most of DSM’s work is invisible to end users, and that is exactly the intention. Someone who receives a gas bill, reads an emissions report or relies on a medical diagnosis does not want to think through every step of measurement, modelling and software. Trust in the result should be self‑evident.

Data Science & Modelling at VSL contributes to the conditions for that trust. It does so by examining, modelling and validating models, software, AI applications and complete measurement chains, and by clarifying the associated uncertainties and assumptions. In a world in which data and (AI) models form the basis for ever more important decisions, this kind of reliable modelling is not optional – it is an absolute necessity.