Difference between revisions of "AIEC Deployment"

From epsciwiki
Jump to navigation Jump to search
 
(29 intermediate revisions by 3 users not shown)
Line 1: Line 1:
  
 
+
{| class="wikitable" style="width: 100%;"
{| class="wikitable"
 
 
|-
 
|-
!Experiment
+
| style="width: 10%" | Experiment
!Date
+
| style="width: 10%" | Date
!Wiki & reports
+
| style="width: 10%" | Wiki & reports
!Model
+
| style="width: 30%" | Model
!Other details
+
| style="width: 40%" | Other details
 
|-
 
|-
 
|PrimEx
 
|PrimEx
Line 19: Line 18:
 
*Shifters ran the scripts to read EPICS and suggest HV, before setting HV and starting the run
 
*Shifters ran the scripts to read EPICS and suggest HV, before setting HV and starting the run
 
*Much of the run was with empty target.  Shifters were not supposed to use the AI w empty tgt but they did.  
 
*Much of the run was with empty target.  Shifters were not supposed to use the AI w empty tgt but they did.  
*The shifters were not asked to set the HV to 2125V before running the AI. This was not ideal, but saved time.  
+
*The shifters were not asked to set the HV to 2125V before running the AI. This was not ideal, but saved time.
 
|-
 
|-
 
|Cosmics
 
|Cosmics
Line 37: Line 36:
 
|CPP
 
|CPP
 
|June 2022
 
|June 2022
 +
|[https://halldweb.jlab.org/wiki/index.php/CPP_DAQ/Monitoring_Test_Spring_2022#AIEC_planning CPP plans]
 
|
 
|
|
+
*GPR with MLP (multilayer perceptron) learned prior, trained on GlueX-2020 + PrimEx2021  
*Linear regression with Bayesian learned prior???, trained on GlueX-2020 + PrimEx2021  
+
*Uncertainty quantification used to determine when to return to the comfort zone (3%: is threshold for standard_deviation_of_prediction/ideal_GCF)  
*Uncertainty quantification used to determine when to return to the comfort zone (3%?)  
 
 
*Default HV  2125V
 
*Default HV  2125V
 
*HV predictions rounded to nearest 1V
 
*HV predictions rounded to nearest 1V
*/group/halld/AIEC/utilities/CDC_control_ai/cpp/d_cdc_modelneeds/gpflow_params.pkl
+
*/group/halld/AIEC/utilities/CDC_control_ai/ cpp/d_cdc_modelneeds/gpflow_params.pkl
 
|
 
|
 
*RoboCDC summoned by DAQ 'GO' to read EPICS and set HV.
 
*RoboCDC summoned by DAQ 'GO' to read EPICS and set HV.
 
*RoboCDC scaled the HVB current reading to what it would have been at the default HV, before predicting GCF.  
 
*RoboCDC scaled the HVB current reading to what it would have been at the default HV, before predicting GCF.  
 
*Overall on/off switch added to CDC HV control gui.
 
*Overall on/off switch added to CDC HV control gui.
 
+
*[https://halldweb.jlab.org/wiki/index.php/CPP_DAQ/CPP_AIEC_config CPP AIEC CONFIG]
*running exactly once at run start with a control window of [2110.0, 2150.0] Volts.
+
|-
*trailing epics window of 15 (seconds)
+
|PrimEx 2022
 
+
|Aug-Dec 2022
 +
|[[PrimEx_2022]]
 +
|
 +
*Same model as for CPP, fed fixed HVBi
 +
*GPR with MLP (multilayer perceptron) learned prior, trained on GlueX-2020 + PrimEx2021
 +
*Uncertainty quantification used to determine when to return to the comfort zone (3%: is threshold for standard_deviation_of_prediction/ideal_GCF)
 +
*Default HV  2125V
 +
*HV predictions rounded to nearest 1V
 +
*/group/halld/AIEC/utilities/CDC_control_ai/ cpp/d_cdc_modelneeds/gpflow_params.pkl
 +
|
 +
*HVB current is fixed at the value read during initialization
 +
* RoboCDC switches itself off when the target is changed to empty
 +
* GPFlow Model (trained on 2018+2020+2021PrimeX+2021PrimeX_Empty_Target) saved in a model efficient format (tf function). Load model code was updated to use the new format.
 +
* HVBi current input value is fixed for run period duration
 +
|-
 +
|GlueX 2023
 +
|Jan 12 - Mar 22 2023
 +
|[[GlueX_2023]]
 +
|
 +
*TBD
 +
*Default HV  2125V
 +
*HV predictions rounded to nearest 1V
 +
|
  
 
|}
 
|}

Latest revision as of 15:04, 15 December 2022

Experiment Date Wiki & reports Model Other details
PrimEx Oct-Nov 2021?? PrimEx_instructions
  • GPR, trained on GlueX-2020
  • Default HV 2125V
  • HV predictions were rounded to nearest 5V
  • Shifters ran the scripts to read EPICS and suggest HV, before setting HV and starting the run
  • Much of the run was with empty target. Shifters were not supposed to use the AI w empty tgt but they did.
  • The shifters were not asked to set the HV to 2125V before running the AI. This was not ideal, but saved time.
Cosmics March 2022 Cosmics_test_plans
  • GPR, trained on GlueX-2020 + PrimEx2021
  • Default HV 2130V
  • HVB current fixed at 9A
  • Temp fixed at 299K??? at start of run
  • HV predictions rounded to nearest 1V
  • RoboCDC ran autonomously, setting the HV every 5 minutes.
  • Half of the CDC was kept at fixed default HV.
  • Temp fixed at 299 or 300K until FDC LV brought temp into trained region
CPP June 2022 CPP plans
  • GPR with MLP (multilayer perceptron) learned prior, trained on GlueX-2020 + PrimEx2021
  • Uncertainty quantification used to determine when to return to the comfort zone (3%: is threshold for standard_deviation_of_prediction/ideal_GCF)
  • Default HV 2125V
  • HV predictions rounded to nearest 1V
  • /group/halld/AIEC/utilities/CDC_control_ai/ cpp/d_cdc_modelneeds/gpflow_params.pkl
  • RoboCDC summoned by DAQ 'GO' to read EPICS and set HV.
  • RoboCDC scaled the HVB current reading to what it would have been at the default HV, before predicting GCF.
  • Overall on/off switch added to CDC HV control gui.
  • CPP AIEC CONFIG
PrimEx 2022 Aug-Dec 2022 PrimEx_2022
  • Same model as for CPP, fed fixed HVBi
  • GPR with MLP (multilayer perceptron) learned prior, trained on GlueX-2020 + PrimEx2021
  • Uncertainty quantification used to determine when to return to the comfort zone (3%: is threshold for standard_deviation_of_prediction/ideal_GCF)
  • Default HV 2125V
  • HV predictions rounded to nearest 1V
  • /group/halld/AIEC/utilities/CDC_control_ai/ cpp/d_cdc_modelneeds/gpflow_params.pkl
  • HVB current is fixed at the value read during initialization
  • RoboCDC switches itself off when the target is changed to empty
  • GPFlow Model (trained on 2018+2020+2021PrimeX+2021PrimeX_Empty_Target) saved in a model efficient format (tf function). Load model code was updated to use the new format.
  • HVBi current input value is fixed for run period duration
GlueX 2023 Jan 12 - Mar 22 2023 GlueX_2023
  • TBD
  • Default HV 2125V
  • HV predictions rounded to nearest 1V