Use DeltaV Neural Networks to create virtual sensors that continuously predict a process output based on measured process inputs. In addition, neural networks calculate and provide as a block output the future value of the output based on the current input values. Use neural networks where the process output is only available through analysis of lab samples or online analysis is not reliable.
The NN function block is the basis of implementing neural networks in a DeltaV system. You define all parameters in the NN block using Control Studio. From Control Studio, you can launch the DeltaV Neural application to train the neural network.
If you are using a neural network to augment or replace the output of values determined by lab analysis, use the Lab Entry (LE) function block to provide the lab sample data to the NN block. When the neural network is providing backup for an online analyzer, use an Analog Input block to provide the measurement to the NN block. The neural network uses historical sample input data to train the neural network. After training, the neural network uses the sample input data to automatically correct the calculated output for changes in the process and unmeasured inputs. The ability of the NN block to adapt to changes reduces maintenance by minimizing the need to retrain the neural network to compensate for process changes.
Use the DeltaV Neural application to train and test neural networks. When a module containing an NN block is downloaded, all inputs and outputs are assigned to the Continuous Historian. The NN function block can predict the future output of a process given the current inputs, allowing for process dynamics. Refer to the DeltaV Neural topic for more information on neural network and training.
The operator runs the control through a DeltaV Operate picture constructed with the NN and LE (Lab Entry) dynamos.
The NN function block supports modes and status handling.
The NN function block is not supported in composites.
The following diagram shows the internal components of the Neural Network function block:
