📄 chemistry

Integrated Evaluation Framework for Standard Reference Thermophysical Property Data in the Korean Thermophysical Properties Databank (KDB)

This paper presents and validates a structured, reproducible framework for generating uncertainty-aware standard reference thermophysical data within the Korean Thermophysical Properties Databank (KDB), which was successfully applied to process 82,488 literature data points into 2,363 vetted entries across six key properties.

Original authors: Beom Chan Ryu, Sun Yoo Hwang, Chang Geun Kim, Jeong Won Kang

Published 2026-09-21

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Beom Chan Ryu, Sun Yoo Hwang, Chang Geun Kim, Jeong Won Kang

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the world of engineering and safety, numbers are not just abstract figures; they are the blueprints for how we build, how we transport energy, and how we protect the environment. When scientists design a chemical plant or assess the risk of a fire, they rely on thermophysical properties—specific, measurable characteristics of materials like how easily they catch fire, how they dissolve in water, or how they behave under heat. These values are the foundation of trust in science. However, a single number without context is dangerous. If a researcher reports a temperature but does not explain how much that number might vary, or where the measurement came from, engineers cannot know if it is safe to use. The reliability of these data points depends not only on the value itself but on a clear record of its uncertainty and its origin. For decades, the scientific community has struggled to balance the need for open, free access to this information with the rigorous, often expensive process of verifying that every number is accurate and trustworthy.

A team of researchers from Korea University and the National Center for Standard Reference Data has addressed this challenge by creating a new, systematic way to organize and verify these critical numbers. They focused on six specific properties that are vital for process design and safety: the temperature at which a liquid gives off enough vapor to ignite, how a substance splits between water and oil, the temperature at which it spontaneously catches fire, the force holding its surface together, a factor describing how its vapor pressure behaves, and a measure of how well it dissolves in other substances. The team took a massive collection of 82,488 raw data points gathered from decades of scientific literature and subjected them to a strict, step-by-step evaluation process. Their goal was to transform this chaotic mix of reports into a clean, reliable set of standard reference data that anyone could use with confidence.

The process began by gathering every available measurement for these six properties, ensuring that each number could be traced back to its original source. The researchers then acted as a filter, removing duplicate entries and discarding data points that were too few in number to be meaningful. They applied a statistical method to identify and remove outliers—numbers that were so far off from the average that they likely represented errors or unique experimental conditions rather than the true nature of the material. For the remaining data, the team calculated a weighted average, giving more importance to measurements that came with smaller margins of error. Crucially, they did not just accept these numbers; they tested them against independent predictions. For some properties, they compared the results to computer models that predict behavior based on molecular structure. For others, they calculated what the value should be by combining the uncertainties of the fundamental properties that define it.

Only when a value passed every single test—showing it was traceable, statistically sound, and consistent with independent theories—did the team register it as a "standard reference data" entry. This rigorous filtering reduced the original 82,488 raw points down to 2,363 high-quality, verified entries. The remaining 1,198 values that did not meet every single criterion were still kept in the database but labeled as "reference data," indicating they are useful but carry a different level of certainty. This distinction allows users to choose the right level of confidence for their specific needs. The researchers also discovered that this cross-checking process acted as a diagnostic tool, revealing hidden errors in existing databases. For instance, they found that some previously stored values for the solubility of water and ethyl iodide were significantly incorrect, a mistake that was only caught because the new evaluation process compared the data against independent calculations.

The final result is a publicly accessible database where every number comes with a clear statement of its uncertainty and a quality classification. This means that an engineer designing a safety valve or a scientist modeling environmental risks can now look up a property and know exactly how much they can trust it. The study demonstrates that by combining careful data collection, statistical screening, and independent verification, it is possible to turn a scattered collection of scientific reports into a unified, reliable resource. The researchers have made this entire framework available to the public, ensuring that the path from a raw laboratory measurement to a trusted engineering standard is now transparent, reproducible, and open to everyone. This work does not just add more numbers to a list; it provides a method for ensuring that the numbers we rely on for safety and design are as solid as the structures they help build.

Technical Summary: Integrated Evaluation Framework for Standard Reference Thermophysical Property Data in the KDB

Problem Statement
Reliable thermophysical property data are critical for engineering applications, including process design, safety analysis, and environmental risk assessment. However, the reliability of such data depends not only on the reported values but also on transparent uncertainty quantification and source traceability. While open databases (e.g., CoolProp, OpenCalphad) offer broad accessibility, they often lack systematic uncertainty evaluation. Conversely, rigorously evaluated resources (e.g., DIPPR, NIST-TDE) typically require subscriptions. Furthermore, although the Korean Thermophysical Properties Databank (KDB) previously provided critically evaluated data, its evaluation workflow lacked a documented, coherent, and reproducible procedure for constructing standard reference data. Specifically, the platform-level description did not formally integrate source traceability, outlier rejection, uncertainty propagation, and expert review into a unified framework.

Methodology
The study formalizes a structured, three-phase framework for constructing standard reference data within the KDB, applied to six thermophysical properties: flash point (FP), octanol–water partition coefficient (log Kow), autoignition temperature (AIT), surface tension ( $\gamma$ ), acentric factor ( $\omega$ ), and Hildebrand solubility parameter ( $\delta$ ).

Data Collection and Refinement:
- Raw data were compiled from peer-reviewed literature, standard compilations (DIPPR, NIST-TDE), and existing KDB datasets, totaling 82,488 data points.
- Sources were classified as traceable (linked to specific experiments/publications) or non-traceable (pre-aggregated values). Non-traceable sources were merged and deduplicated, while traceable sources were retained in original form.
- Data were deduplicated based on source traceability.
Data Evaluation and Uncertainty Estimation:
- Outlier Detection: For compounds with $N \ge 6$ data points, an iterative $\pm 2\sigma$ (z-score) criterion was applied to remove outliers. This threshold was chosen over $\pm 3\sigma$ to remain effective for smaller sample sizes common in thermophysical databases. Compounds with $N < 3$ were excluded; those with $3 \le N \le 5$ proceeded to averaging without outlier screening.
- Weighted Mean and Uncertainty: A weighted mean ( $x_w$ ) was calculated using inverse-variance weights ( $w_i = 1/u_i^2$ ). If reported uncertainties were missing, estimated uncertainties were assigned based on average experimental uncertainties, with a penalty (doubling) applied to data from non-traceable sources. A cap of 0.5 was placed on the relative weight of any single data point to prevent dominance.
- Uncertainty Propagation: The combined standard uncertainty was scaled by the Birge ratio ( $R_B$ ) if $R_B > 1$ . Expanded uncertainty ( $k=2$ ) was reported.
- Derived Properties: For $\gamma$ , $\omega$ , and $\delta$ , uncertainties were propagated from basis properties (e.g., critical temperature, enthalpy of vaporization) using defining equations, treating correlation coefficients as fixed constants due to lack of reported uncertainties.
Cross-Checking and Expert Review:
- Cross-Checking: Evaluated values were compared against independent benchmarks.
  - Fixed Properties (FP, log Kow, AIT): Compared against predictive models (GC-GCN, XLogP3, ANN-GC).
  - Derived Properties ( $\gamma$ , $\omega$ , $\delta$ ): Compared against values calculated by propagating uncertainties of basis properties.
  - A value satisfied the criterion if its expanded uncertainty interval overlapped with the comparison value.
- Expert Committee Review: An expert committee evaluated candidates against eight criteria (e.g., traceability, measurement method adequacy, consistency with theoretical values). Since exactly seven of the eight criteria applied to any given property type (based on whether it was fixed or derived), a value was registered as Standard Reference Data only if all seven applicable criteria were met.
- Classification: Data meeting all criteria were registered as Standard Reference Data; those failing the cross-check but retaining reliable experimental evidence were classified as Reference Data.

Key Contributions

Formalized Workflow: The paper establishes a reproducible, end-to-end procedure for converting heterogeneous literature data into uncertainty-aware standard reference data, explicitly documenting traceability, outlier handling, and expert decision-making.
Integrated Uncertainty and Traceability: The framework ensures that every registered entry includes expanded uncertainty estimates and a clear provenance chain, addressing the gap between open-access databases and rigorous, subscription-based evaluations.
Dual-Role Cross-Checking: The study demonstrates that cross-checking serves both as a registration criterion (validating consistency with independent models or propagation) and a diagnostic tool for identifying errors in supporting database entries (e.g., correcting erroneous values for water and ethyl iodide in the original KDB).
Dataset Expansion: The methodology was applied to 82,488 raw data points, resulting in 2,363 registered Standard Reference Data entries and 1,198 Reference Data entries across six properties.

Results

Data Consolidation: The consolidation rate (Standard Reference Data / Raw Data Points) varied by property, ranging from 0.80% for log Kow (due to extensive duplicate removal from QSAR/QSPR compilations) to 8.89% for the acentric factor.
Cross-Checking Performance:
- For fixed properties, Standard Reference Data clustered closely around predictive model parity lines, while Reference Data showed greater scatter.
- For derived properties, cross-checking identified specific inconsistencies. For example, the propagated uncertainty for the acentric factor of ethylcyclohexane was artificially inflated due to a large uncertainty in the critical temperature from the source dataset, masking a discrepancy in central estimates.
- Errors in the original KDB entries for the Hildebrand solubility parameter of water and ethyl iodide were identified and flagged for correction.
Statistical Agreement: Standard Reference Data generally exhibited lower Mean Absolute Error (MAE) and higher coefficients of determination ( $R^2$ ) compared to Reference Data when compared against independent benchmarks, validating the registration criteria.

Significance
The paper claims that this framework provides a systematic methodology for consolidating heterogeneous literature data into reliable standard reference data with quantified confidence levels. By integrating traceability, rigorous uncertainty evaluation, and independent cross-checking, the study establishes a foundation for trustworthy process design and safety assessment. The resulting dataset, now publicly accessible via the KDB platform with full metadata (uncertainty, quality classification, and supporting data counts), offers a resource that bridges the gap between broad accessibility and high-quality, uncertainty-aware data. The authors note that the framework is designed as an operational protocol for the KDB, enabling the continuous registration of new literature data.

Technical Summary: Integrated Evaluation Framework for Standard Reference Thermophysical Property Data in the KDB

More like this