Data Fit for Plant or Lab Data • Data Fit Fit is a tool for for fitting fitting simulation simulation models models to plant plant or lab data. Data Fit allows you to: – Fit plant/lab data to the same models that have been used for
design – Estimate any feed stream or block input variable from plant/lab
data – Reconcile any accessible variable with any amount of data
General Procedure for Data Fitting Process Data Review - Verify reproducibility - Verify steady state - Verify data feasibility
Preliminary Model Development - Develop a basic model - Ignore details (e.g. non-ideal mixing). - Specify temperature instead of duty - Specify volume instead of residence time
Literature Search - In-house lab data - Journals and Handbooks - Electronic databases
Trend Analysis - Use Sensitivity to evaluate trends - Compare predicted trends with data
Preliminary Model Fitting - Physical property data regression - Property constant estimation - Verify properties and phase equilibrium
Model Refinement - Use Data-Fit - Relax model assumptions as needed
Influence of Physical Properties • Physical property parameters influence reaction kinetics • Density (DNLRKT) - Concentration is proportional to density. Reaction kinetics depend on component concentrations • Vapor pressure (PLXANT, HENRY) –
The vapor pressure controls phase equilibrium of volatile components in vapor-liquid systems. The phase equilibrium strongly influences concentrations, which controls kinetics
• Enthalpy (DHFORM) –
The component enthalpies influence the predicted heat duties and temperatures in the model
• Heat Capacity (CPIG, CPL) –
The heat capacity controls the influence of temperature on enthalpies
• Phase equilibrium –
In multiphase reactors, the phase equilibrium determines the component concentrations in each phase which influences the reaction rates
Use of Data Fit • Any amount of data can be used • Any accessible variable including Property Set properties can be reconciled • Uses Maximum Likelihood Principle • Any accessible feed stream or non-integer block input variable can be estimated, including – User subroutines constants – Reaction kinetic parameters – Physical property model parameters
Type of Data • There are two types of Data – Point-data for continuous unit operations • • •
Operating conditions for steady-state unit operation models Feed streams for continuous process or batch charge streams Analytical data, measured flow rates, or composition data for product streams
– Profile-data for time or length profile data • •
Operating profiles for batch reactors or plug flow reactors Time series measured data for a batch reactor of data along the axial profile of a plug flow reactor
Fitting Kinetic Parameters • If no reference temperature is specified: knet = k0exp(-Eact /RT) – The activation energy controls the magnitude of the reaction rate as well as the temperature sensitivity of the reaction rate
• If a reference temperature is specified: knet = k0exp(-Eact /R(1/T - 1/T ref ) – With this approach, the pre-exponential factor controls the magnitude of the reaction rate at the reference temperature – The activation energy controls the temperature sensitivity of the rate constant
Scaling the Fitted Parameters • When several types of parameters are adjusted in the same run, the magnitude of the manipulated parameters may influence the convergence • Ideally, the magnitude of the manipulated parameters should be within several orders of magnitude of each other • To scale the manipulated parameters: – Define a Parameter variable and initialize it to a value of 1 in a
Calculator block – In another Calculator block, multiply the Parameter variable by the base case value – Manipulate the Parameter variable in the Data Fit instead of the actual variable
