31
Uncertainty Quantification (quoFEM) / Re: Choosing the best parameters using quoFEM
« on: September 20, 2021, 04:24:56 PM »
The DesignSafe Data Depot is back in service. You can now download the app.
Show Posts
This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.
32
Uncertainty Quantification (quoFEM) / Re: Choosing the best parameters using quoFEM
« on: September 15, 2021, 03:50:57 PM »
The option to pass in calibration data is available in the latest released version of the quoFEM app. I recommend installing this new version. You will find the link to download the app on this page - https://simcenter.designsafe-ci.org/research-tools/quofem-application/
Note that the DesignSafe-CI page is currently undergoing maintenance and will be available in a few hours.
Note that the DesignSafe-CI page is currently undergoing maintenance and will be available in a few hours.
33
Uncertainty Quantification (quoFEM) / Re: Choosing the best parameters using quoFEM
« on: September 14, 2021, 09:12:02 PM »
Hello Rashad,
To find the best parameter values, you will need to make some changes to the way you defined the analysis in quoFEM:
1) In order to find the best value of the parameters of your model, you should use the 'Parameters Estimation' method category in the ‘UQ’ panel of quoFEM, under the ‘Dakota’ UQ Engine. You also need to provide data, which are target values of the output of the model. These target values could be, for example, measurements of the output quantity of interest from experiments. Then, quoFEM will find the best values of the parameters that makes the model output match the target values of the output as well as possible.
The data for parameter estimation must be provided in a text file which should satisfy the following requirements:
•Each row contains data from one experiment
•The number of data in each row must equal the sum of the length of all response quantities defined in the QoI panel
•The order of the response quantities must match in the QoI panel, the calibration data file, and results file from analyses
So, you should provide the target value of the output Node_16_Disp_1 in a text file and provide the path to this file as input to the parameter estimation method in the UQ panel. (Refer attached figure of the UQ panel)
2) The parameter estimation method uses optimization algorithms to find the best value of the parameters, given the target value of the outputs of the model. Here, ‘the best value’ means ‘that value of the input parameters which minimizes the sum of the squared error between the outputs of the model and the data’. You will need to define the lower and upper bounds on the parameter values and the initial point for the optimization algorithm in the RV panel. (Refer attached figure of the RV panel)
3) Upon completion of the analysis, the 'best parameter' values will be displayed in the RES panel. (Refer attached figure of the RES panel)
To find the best parameter values, you will need to make some changes to the way you defined the analysis in quoFEM:
1) In order to find the best value of the parameters of your model, you should use the 'Parameters Estimation' method category in the ‘UQ’ panel of quoFEM, under the ‘Dakota’ UQ Engine. You also need to provide data, which are target values of the output of the model. These target values could be, for example, measurements of the output quantity of interest from experiments. Then, quoFEM will find the best values of the parameters that makes the model output match the target values of the output as well as possible.
The data for parameter estimation must be provided in a text file which should satisfy the following requirements:
•Each row contains data from one experiment
•The number of data in each row must equal the sum of the length of all response quantities defined in the QoI panel
•The order of the response quantities must match in the QoI panel, the calibration data file, and results file from analyses
So, you should provide the target value of the output Node_16_Disp_1 in a text file and provide the path to this file as input to the parameter estimation method in the UQ panel. (Refer attached figure of the UQ panel)
2) The parameter estimation method uses optimization algorithms to find the best value of the parameters, given the target value of the outputs of the model. Here, ‘the best value’ means ‘that value of the input parameters which minimizes the sum of the squared error between the outputs of the model and the data’. You will need to define the lower and upper bounds on the parameter values and the initial point for the optimization algorithm in the RV panel. (Refer attached figure of the RV panel)
3) Upon completion of the analysis, the 'best parameter' values will be displayed in the RES panel. (Refer attached figure of the RES panel)
34
Wind Engineering (WE-UQ) / Re: correlation between RVs in LHS method
« on: April 19, 2021, 09:51:17 PM »
Yes, in the current version, the RVs are uncorrelated. And the samples of each RV are combined in the shuffling approach of the LHS sampling method using the "restricted pairing" mode to force near-zero correlations between the uncorrelated RVs.
In a future release of WE-UQ, a feature to specify a correlation matrix for the RVs will be made available.
In a future release of WE-UQ, a feature to specify a correlation matrix for the RVs will be made available.
35
Uncertainty Quantification (quoFEM) / Re: FEATURE Request: Option to read in data from a separate file
« on: March 15, 2021, 03:59:41 PM »
This request has been entered into our list of tasks and a feature will be available in the next release that will provide the option of importing calibration data from text files.
36
Uncertainty Quantification (quoFEM) / Re: FEATURE Request: Ability to handle non-scalar response quantities
« on: March 15, 2021, 03:58:19 PM »
Thanks - this might be useful for many users. Yes, it is possible to support non-scalar responses. A feature will be added to specify if the response quantities are scalar or non-scalar, and will be available in the next release of the tool.
37
Uncertainty Quantification (quoFEM) / Re: FEATURE Request: Option to calibrate multipliers on the error covariance
« on: March 15, 2021, 03:57:08 PM »
This has been added to our list of tasks. A feature will be developed that provides the option to pass in the covariance structure for error, and specify that multipliers on the error covariance must be calibrated.
38
Uncertainty Quantification (quoFEM) / Re: FEATURE Request: Option to have a default log likelihood formulation
« on: March 15, 2021, 03:56:26 PM »
Currently, the TMCMC algorithm in the UCSD_UQ engine requires the users to specify the log-likelihood function in a script. This will be made optional and a Gaussian log-likelihood will be made the default. This has been added to our list of tasks.
Bayesian inference using the Dakota engine assumes that the log-likelihood is Gaussian, and does not require the user to write a separate script defining the log-likelihood function.
Bayesian inference using the Dakota engine assumes that the log-likelihood is Gaussian, and does not require the user to write a separate script defining the log-likelihood function.