Neural Network Lab Develop a Neural Network to simulate the temperature exiting a heat exchanger: We will use a simulated heat exchanger in DeltaV, EX2_SIM
Heat Exchanger We will configure the network to determine the tempered water exit temperature based in the inlet temperatures and flow rates
Inputs Output, what we want to see from The Neural Network
Heat Exchanger
Heat Exchanger Calculations The amount of heat transferred is based on the heat transfer coefficient, U and h
Heat Exchange Calculations Heat transfer coefficient non linear function depending on viscosity, thermo conductivity mass flow rate
Inputs Select TT2-1, TT2-2, FC2-1 and FC2-2 Use a 1 second Historical Sampling Rate for this example
Under Advanced Control Lab Entry Block NNet Block under Advanced Control Tab
Right click on NN block Select Extendable Parameters to change Number of inputs
Select 4 inputs
Select the sampling rate, 1 sec in our case
Browse for input parameter, Any floating point variable can be selected
Select the CV or current value
Enter the identifier to articulate the point In the Neural Network toolkit
Wire TT2-4 to the NN sample input
NN block OUT_SCALE should match the Range of the output
Wire to NN_TEMP So we can trend the point
Save the case and Download the network
NN block reference inputs are available through the Historian, but the Historian must be Enabled, Assigned to the Area and Node
NN block reference inputs are available through the Historian, but the Historian must be enabled, Assigned to the Area and Node, then Downloaded!
DeltaV Neural Now that we have built the DeltaV Neural Network, we need to launch ( start up ) the application, either from the Control Studio or from the Explorer
Right Click the block and select Advanced Control > Neural
Start > DeltaV >Advanced Control >Neural
DeltaV Neural Block Reference Inputs and Sample
Right click to set scale
Green shaded area contains the Training set, right click to select Red area for excluded area
DeltaV Neural We need to configure the sample multiplier. This entry should allow the Time to Steady State to be that which you estimate the process to get to steady state after a disturbance
(1/50) * Time to Steady State DeltaV Neural In DeltaV, each input is actually 50 sampled inputs delayed by (1/50) * Time to Steady State This way the model will sense not only the final value but the approach to steady state
In our case, we estimate 5 minutes to get to steady state, multiplier set to 6
Note: You cannot have Control Studio running when training! Press Autogenerate to start the training process …
Note the sensitivities and if the network uses the input! We have to download the network to get it to work!
You can delete other training results, but you cannot delete the current running result.
DeltaV Neural – Display Editing We need to add a faceplate to the operators display. Open the display and do a quick edit. Locate the NN pre configured block and drag it on the display Browse to select the Neural Network
Faceplate
Training and test errors are shown
Expert mode allows User to select data screening range
Show inputs And delays, Click here Sensitivity and input selection Shown. Summ of all inputs = 1.0
Delays are shown
Detail, Can shift in expert mode
Verification Selectors
Bad points here
Excluded Area