ThermoData Engine: New Generation Expert System for Thermodynamic Data Critical Evaluation

ThermoData Engine is the first product fully implementing all major principles of the concept of dynamic data evaluation formulated at the NIST Thermodynamics Research Center of the Thermophysical Properties Division. This concept requires the development of large electronic databases capable of storing essentially all 'raw' experimental data known to date with detailed descriptions of relevant metadata and uncertainties. The combination of these databases with expert software designed to generate recommended data based on available 'raw' experimental data and their uncertainties leads to the possibility of producing data compilations automatically 'to order' forming a dynamic data infrastructure.

Intended Impact

Traditionally, critical data evaluation is an extremely time- and resource-consuming process, which includes extensive use of labor in data collection, data mining, analysis, fitting, etc. Because of this, it must be performed far in advance of a need within an industrial or scientific application. In addition, it is quite common that by the time the critical data-evaluation process for a particular chemical system or property group is complete (sometimes after years of data evaluation projects involving highly skilled data experts), it must be re-initiated because significant new data have become available. This type of slow and inflexible critical data evaluation can be defined as 'static.' These shortcomings have become magnified dramatically within the last 5 to 10 years due to the significant increase in the rate of publication of experimental and predicted thermodynamic data which need to be analyzed during the critical data evaluation process.

To address the weaknesses of 'static' evaluations, the concept of a dynamic data evaluation system was developed at the NIST Thermodynamics Research Center of the Thermophysical Properties Division. This concept requires large electronic databases capable of storing essentially all experimental data known to date with detailed descriptions of relevant metadata and uncertainties. The combination of these electronic databases with expert-system software, designed to automatically generate recommended data based on available experimental data, leads to the ability to produce critically evaluated data dynamically or 'to order' (see figure). This concept contrasts sharply with static critical data evaluation, which must be initiated far in advance of a particular need. The dynamic data evaluation process dramatically reduces the effort and costs associated with anticipating future needs and keeping static evaluations current.

Objective

Develop algorithms fully implementing the concept of dynamic data evaluation for thermophysical and thermochemical properties of all major types of chemical systems, and incorporate them within NIST SRD products assuring internal consistency of the information and multiple capabilities of "bundling" of this product with a variety of chemical engineering software applications.

Goals

Gradually increase the scope of the implementation of the dynamic data evaluation concept for the following systems and applications:

• Pure chemical compounds
• Equations of state on-demandBinary mixtures
• Chemical reactions
• Software for strategic measurement planning
• Ternary mixtures
• Software capable of propagating uncertainties of the thermophysical and thermochemical property data into uncertainties of "reaction streams"

Research activities and technical approach

Critically evaluated data produced by the deployment of the dynamic data-evaluation concept can rigorously be characterized with their quality assessments. That, in turn, provides the ability to propagate reliable data-quality limits to all aspects of chemical process design. Implementation of the dynamic data evaluation concept consists of the completion of a number of major tasks: (1) design and development of a comprehensive database system structure based on the principles of physical chemistry and capable of supporting a large-scale data entry operation for the complete set of thermophysical, thermochemical, and transport properties for chemical systems including pure compounds, binary mixtures, ternary mixtures, and chemical reactions; (2) development of software tools for automation of the data-entry process with robust and internally-consistent mechanisms for automatic assessments of data uncertainty; (3) design and development of algorithms and software tools to assure quality control at all stages of data entry and analysis; (4) development of algorithms and computer codes to implement the stages of the dynamic data-evaluation concept; (5) development of algorithms to implement, target, and apply prediction methods depending on the nature of the chemical system and property, including automatic chemical structure recognition mechanisms; and (6) development of procedures allowing generation of output in a format suitable for application in major commercial simulation engines for chemical-process design.

The ThermoData Engine (TDE) – NIST Standard Reference Database 103 - incorporates all major stages of the concept implementation including data retrieval, grouping, normalization, sorting, consistency enforcement, fitting, and prediction. The SOURCE data system is used in conjunction with TDE as the comprehensive storage facility for experimental thermophysical and thermochemical property data. In addition, the NIST/TRC Ideal Gas Database is used as a source of thermodynamic property data in the ideal-gas state.

In FY05, version 1.0 (now designated as NIST SRD 103a) was released through the NIST Standard Reference Data Office. This version was limited to thermodynamic properties of pure compounds. The software architecture emphasizes enforcement of consistency between related properties (including those obtained from predictions), assumes an imperfect source of original data, provides for flexibility in selection of default data models depending on the particular data scenario, incorporates a large variety of models for secondary fitting, and allows saving of critically evaluated data in the ThermoML format. The latter assures compatibility of the TDE software with any engineering application equipped with a ThermoML software 'reader.'

In the second version of TDE, released in 2007, the implementation of dynamic data evaluation was further expanded for pure compounds to generation of equations of state on demand. The four principal equations of state (original and modified volume-translated Peng-Robinson, Sanchez-Lacombe, PC-SAFT, and Span-Wagner) were included to assure adequate experimental and predicted data fitting depending on various “data scenarios”. Periodical Web updates of the local TDE-SOURCE database maintain its up-to-date status by use of a Web-Oracle infrastructure, providing new data to users of TDE soon after original publication in the literature. In 2008, the capabilities of the first two versions of the TDE were combined within the NIST Standard Reference Database 103a.

The third version of TDE (now NIST SRD 103b), released in 2008, expanded the implementation of the concept to the thermophysical properties of binary mixtures. TDE provides access to single-phase thermodynamic and transport property data, VLE, LLE, and SLE data for more than 30,000 mixtures and performs automated evaluation of most of those properties. Certain properties such as densities, critical, and transport properties are described by special fitting equations; phase equilibria data are described by activity coefficient models selected by the user from the set of supported models: Margules, NRTL, Redlich-Kister, UNIQUAC, van Laar, and Wilson. UNIFAC predictions are generated for mixtures covered by the UNIFAC method, including those for which experimental data are currently not available. Phase diagrams, isotherms, and isobars based on those models can be calculated and drawn for the user's convenience. Proprietary data can be entered for inclusion in the evaluation, and the user can influence the evaluation process by changing relative data weights or by rejecting particular data sets.