Baseline Correction #319
Baseline Correction #319
Conversation
|
I still need to create unit tests for the Also, since the syntax page is so long, perhaps we should discuss reorganizing it. Maybe I can move the part where I talk about the numerical implementation to the theory manual and move the part where I present demonstrations to examples. |
crswong888
force-pushed
the
baseline-correction-296
branch
from
b6fc0fb
to
043a132
Compare
| const Real & beta, | ||
| const Real & dt); | ||
|
|
||
| /// Solves linear normal equation for minimum acceleration square error |
There was a problem hiding this comment.
Is there a simpler explanation to the phrase linear normal equation?
There was a problem hiding this comment.
@somu15 This is what it is referred to as in terms of linear algebra jargon. For example, see section 8.2 (page 510) of this textbook, or see this Wikipedia page.
It's just the equation that solves the least squares polynomial. I suppose I could be straight-up and call it this instead.
| MooseUtils::DelimitedFileReader reader(getParam<FileName>("data_file"), &_communicator); | ||
| reader.read(); | ||
|
|
||
| // Check if specific column headers were input |
There was a problem hiding this comment.
Do you need to always have the column header names? Currently, as far as I know, when we input accelerations to MASTODON, we are not required to have column header names.
There was a problem hiding this comment.
No this is not required, but, here, I enforced the case where if one variable was specified a header (e.g., time), then the other variable (acceleration), must also be specified a column header. This is to avoid errors when the file is read by the DelimitedFileReader class.
If no header is specified for either, then the default is to read the first column as time and the second column as acceleration.
| @@ -0,0 +1,409 @@ | |||
| # Baseline Correction | |||
|
|
|||
| `BaselineCorrection` is a [`Function`](syntax/Functions/index.md) object that applies a [!ac](BLC) to an acceleration time history, $\ddot{u}(t)$, and outputs the adjusted values. The corrections are performed by first computing the nominal (*uncorrected*) velocity, $\dot{u}$, and displacement, $u$, by the [Newmark-beta method](manuals/theory/index.md#time-integration). An $n$-th order polynomial approximation of the acceleration, velocity, and/or displacement is then found by the method of least squares and each are successively subtracted from the nominal acceleration to obtain the corrected one, $\ddot{u}^{\ast}$. That is, the corrected acceleration time history is given by the following: | |||
There was a problem hiding this comment.
An $n$-th order polynomial approximation of the acceleration, velocity, and/or displacement is then found by the method of least squares and each are successively subtracted from the nominal acceleration to obtain the corrected one, $\ddot{u}^{\ast}$. looks a bit confusing and you may want to break it up. Also, consider explaining more each are subtracted from.
There was a problem hiding this comment.
@somu15 the next section about the numerical implementation explains this. Please read on.
|
@crswong888 Nice PR. I made a few comments for now. Please make changes if you agree with them. @cbolisetti may have additional comments. |
|
@somu15 Thank you very much for taking the time to go through my work. I replied to most of your comments, perhaps you will continue some discussions there. I am in agreement with many of your comments and will make those changes. |
|
Hi @crswong888 What's the status of this? |
|
@somu15 I still haven't gotten around to addressing your comments. I can do this by the end of next week, the latest, if this is okay. |
crswong888
force-pushed
the
baseline-correction-296
branch
from
043a132
to
6273f54
Compare
crswong888
force-pushed
the
baseline-correction-296
branch
from
6273f54
to
4845be3
Compare
…these as an example problem (refs idaholab#296 idaholab#344)
…re's no good reason why this shouldn't be allowed (refs idaholab#296)
…cement time history (refs idaholab#296)
…er and a placeholder page to move all of the demonstrations to (refs idaholab#296 idaholab#344)
crswong888
force-pushed
the
baseline-correction-296
branch
from
a5a9089
to
90c34a6
Compare
|
@somu15 I have addressed some of the things we have talked about. I have made some pretty significant changes, mainly, I've made it possible for the function to output the corrected values for any of the three types of time series: acceleration, velocity, or displacement, since there's no reason why this wasn't allowed. Please review my work when you get the chance. |
|
All jobs on 90c34a6 : invalidated by @cbolisetti re-running. |
|
Could you please exclude the .gif file? We thought that it's big, memory wise. |
|
How much time are the baseline correction tests taking now, Chris? |
There was a problem hiding this comment.
@cbolisetti I am done with my review. I have some additional comments for Chris.
| @@ -0,0 +1,6 @@ | |||
| <!--placeholder page for BaselineCorrection examples, do not link this to examples index--> | |||
There was a problem hiding this comment.
Can you please list this example in the index.md file as well?
|
|
||
| <!-- reference action system here when/if it becomes available--> | ||
|
|
||
| The corrected acceleration output from the `BaselineCorrection` object can be passed to [`PresetAcceleration`](source/bcs/PresetAcceleration.md) to be used as a dynamic excitation on a mesh boundary. However, the initial corrected acceleration may not be zero-valued, and by default, MOOSE assumes that $\ddot{u}(0) = 0$ for a given instance of `NewmarkAccelAux`. Thus, to set the correct initial condition, i.e., $\ddot{u}(0) = \ddot{u}^{\ast}(0)$, so that integration is properly handled by the boundary condition object, a user may invoke the `FunctionIC` object, as demonstrated in [example-3]. |
There was a problem hiding this comment.
I thought that this documentation is really good. Can the user know which combination (i.e., AVD or VD, etc.) will work the best apriori? If not, is there any guidance for this? Also, the optimal polynomial order can only be determined by trial and error, right?
|
|
||
| The corrected acceleration output from the `BaselineCorrection` object can be passed to [`PresetAcceleration`](source/bcs/PresetAcceleration.md) to be used as a dynamic excitation on a mesh boundary. However, the initial corrected acceleration may not be zero-valued, and by default, MOOSE assumes that $\ddot{u}(0) = 0$ for a given instance of `NewmarkAccelAux`. Thus, to set the correct initial condition, i.e., $\ddot{u}(0) = \ddot{u}^{\ast}(0)$, so that integration is properly handled by the boundary condition object, a user may invoke the `FunctionIC` object, as demonstrated in [example-3]. | ||
|
|
||
| !listing baselinecorrection/twenty_sines_with_bc.i |
There was a problem hiding this comment.
If I want to use the VD combination, should I set acceleration_fit_order to zero?
| design = functions/BaselineCorrection.md | ||
|
|
||
| [test_correction] | ||
| requirement = "The BaselineCorrection class shall accurately adjust a raw acceleration time history according to the method of using least squares polynomial fits." |
There was a problem hiding this comment.
The BaselineCorrection class shall accurately adjust a raw acceleration time history according to the method of using least squares polynomial fits for
| input = half_sine_scaled.i | ||
| csvdiff = half_sine_scaled_out.csv | ||
|
|
||
| detail = "The output should be a displacement time history scaled by a factor of 4.886 that does not exhibit drifting behavior. The input acceleration time history is a half cycle of `9.87 \sin(\pi t)`." |
There was a problem hiding this comment.
a half-sine acceleration input.
| input = missing_value_param_error.i | ||
| expect_err = "In BaselineCorrection corrected_accel_func: Either `data_file` or `time_values` and `acceleration_values` must be specified exclusively." | ||
|
|
||
| detail = "This error shall occur since the `time_values` and `acceleration_values` parameters were not provided together." |
There was a problem hiding this comment.
the time_values and acceleration_values parameters are not provided together.
| input = missing_file_param_error.i | ||
| expect_err = "In BaselineCorrection corrected_accel_func: A column header name was specified for the for the time data. Please specify a header for the acceleration data using the 'acceleration_name' parameter." | ||
|
|
||
| detail = "This error shall occur since the `time_name` and `acceleration_name` parameters were not provided together." |
There was a problem hiding this comment.
the time_name and acceleration_name parameters are not provided together.
| input = size_mismatch_error.i | ||
| expect_err = "In BaselineCorrection corrected_accel_func: The length of time and acceleration data must be equal." | ||
|
|
||
| detail = "This error shall occur since the lengths of the time data and acceleration data are not equal" |
There was a problem hiding this comment.
the lengths of the time data and acceleration data are not equal.
| input = no_data_error.i | ||
| expect_err = "In BaselineCorrection corrected_accel_func: The length of time and acceleration data must be > 0." | ||
|
|
||
| detail = "This error shall occur since no data was provided at all." |
There was a problem hiding this comment.
no data is provided at all.
| input = no_order_error.i | ||
| expect_err = "In BaselineCorrection corrected_accel_func: No values were input for parameters 'accel_fit_order', 'vel_fit_order', or 'disp_fit_order'. Please specify an integer value from 0 to 9 for at least one of these parameters." | ||
|
|
||
| detail = "This error shall occur since no least squares polynomials would be computed and, therefore, no adjustments would be made." |
There was a problem hiding this comment.
no least squares polynomials can be computed and, therefore, no adjustments can be made.
| void buildFromFile(); | ||
|
|
||
| /// Builds data from pairs of `time_values` and `acceleration_values' | ||
| void buildFromXandY(); |
There was a problem hiding this comment.
@crswong888 you might already have looked into this, but why not inherit from PiecewiseTabularBase? I think that class is more general than what you are trying to do here. If not, you can just override the buildFromFile() method. You can also set/suppress some of the params in there that are not needed for baseline correction.
| params.addRangeCheckedParam<unsigned int>( | ||
| "accel_fit_order", | ||
| "(0 <= accel_fit_order) & (accel_fit_order < 10)", | ||
| "If this is provided, the acceleration time history will be adjusted using an nth-order " | ||
| "polynomial fit of the nominal acceleration data, where n = accel_fit_order (only integer " | ||
| "values from 0 to 9 are supported)."); | ||
| params.addRangeCheckedParam<unsigned int>( | ||
| "vel_fit_order", | ||
| "(0 <= vel_fit_order) & (vel_fit_order < 10)", | ||
| "If this is provided, the acceleration time history will be adjusted using an nth-order " | ||
| "polynomial fit of the nominal velocity data, where n = vel_fit_order (only integer values " | ||
| "from 0 to 9 are supported)."); | ||
| params.addRangeCheckedParam<unsigned int>( | ||
| "disp_fit_order", | ||
| "(0 <= disp_fit_order) & (disp_fit_order < 10)", | ||
| "If this is provided, the acceleration time history will be adjusted using an nth-order " | ||
| "polynomial fit of the nominal displacement data, where n = disp_fit_order (only integer " | ||
| "values from 0 to 9 are supported)."); |
There was a problem hiding this comment.
Can we add defaults to addRangeCheckedParams? If so, can you please specify defaults?
| mooseError( | ||
| "In BaselineCorrection ", _name, ": The length of time and acceleration data must be > 0."); | ||
|
|
||
| // check that at lease one least squares fit will be applied |
|
@cbolisetti @somu15 i will be addressing your comments very soon |
|
@cbolisetti @somu15 I am going to close this PR and start from scratch. My algorithm has changed significantly. |
@cbolisetti and @somu15,
I created a
Functionobject,BaselineCorrection, to perform baseline correction on acceleration time histories. The object behaves likePiecewiseLinear, except the input is first adjusted according to the least squares method for baseline correction, and then linearly interpolated over time before outputting. ABaselineCorrectionobject can also be passed to aPresetAccelerationobject to be used as a BC, so long as the initial acceleration is properly set by aFunctionICobject. If this PR is merged, I could come back and create an action system that handles all three of these objects.I also created several TestHarness objects for
BaselineCorrection, many of which are described in detail in the MooseDocs syntax page I wrote. The syntax page may seem usually long, mostly, because I want to be sure that MASTODON developers understand how I came up with this code, and what its use is. My approach is also somewhat of a novel one that builds upon some references which I mentioned in the docs. I will likely try to publish a paper that describes my method, at which point, I can come back and add a reference to it. I'll have to describe some verifications I've made by comparing against SeismoSignal and Abaqus, and so I will create an SQA page forBaselineCorrectionat that point as well.I started a separate
dict()for the MooseDocs Acronyms Extension atdoc/acronyms.ymland updated theconfig.ymlfile to include both the one from MOOSE and this new one. This way, anyone can start adding acronyms that are unique to MASTODON documentation if they need to. This change was @aeslaughter approved ✔️Interesting Demos
I describe several demonstrations in the syntax page at
doc/content/source/functions/BaselineCorrection.md, which I encourage you guys' to run a live serve of on your own machines, but a couple of them are particularly interesting:test/tests/functions/baselinecorrection/twenty_sines_with_bc.iapplies a baseline corrected BC. I also ran this input file without baseline correction, and I rendered the two solutions, simultaneously, in the following GIF:BaselineCorrectionto it in the test attest/tests/functions/baselinecorrection/chichi_A01263000_N.i. I used Newmark integration to get the displacements from both the raw record and the corrected one and superimposed their time history plots:TODOs
The following items should be addressed in a future PR:
(closes #296)