Affectionately known as our “time machine”, the VOS is a historian of the future. It stores the outputs of each state (e.g. predicted, optimised) of the digital process model. This is a powerful what if tool. The VOS can be used to run non-intrusive simulations.
In any simulation there are multiple axes to consider:
Multiple states (e.g. predicted, simulated, observed and optimised, etc..)
Multiple timestamps (e.g. at 5, 10, 15, 20 min, etc..)
Multiple models (e.g. Physical, Neural Network, etc..)
Each Digital Process Model stores the required/desired simulated future states of a model output. The examples below show the simulated performance variable in the following cases:
Actual (the historic data of what happened in reality) (blue line)
Predicted (no change in the control or feed variables) (Red Line)
Simulated (back filling the control or feed variables) (Yellow Line)
Optimised (recommended changes in control variables) (Green Line)
Note: For the VOS to produce outputs a digital process model (i.e. a prediction simulator) needs to be present. And to produce an optimised state a Value Driver and Decision Optimizer need to be configured.
Multiple Time Stamps
Each Digital Process Model has multiple simulation steps into the future. The maximum time stamp predicted is process specific:
In fast moving (i.e. less variable) processes (e.g. Grinding) the prediction window can be only minutes
In slower moving processes (e.g. Thickeners) the prediction window can be in hours
For longer term mine planning models, the prediction window can be in days/weeks
Note: the prediction window can be made as long as possible the above defines the most "useful" prediction window - As a rule of thumb: The longer the prediction the higher the uncertainty and lower the accuracy
Below is an example of the metric ProcessPlant_DIT5005_SCLD at different time stamps N+X, X being the time stamp in minutes.
These Metrics can be trended over time, the user can see how the delta between predictions and the actual variable increase as the length of the time window increases.
Multiple Digital Process Models can be run in parallel (often different types) the example below compares the predicted "Clear Water Height" generated from 3 different models:
AC = Actual Value
NN = Neural Network (i.e. Machine learning)
PDE= Partial Differential Equation (i.e. physical modelling)
BM = Basic Model (which was generated by the user themselves)
Note: Brains.app can run a dual modelling approach in which multiple models can run in parallel to simulate the same variables therefore sense checking the outputs and increasing the users trust and confidence.
The VOS stores each of these potential situations. These situations as "what ifs" are stored on a virtual version of the asset, to keep the integrity of the original asset.
The VOS generates all of these future outcomes every time new data is ingested
For most digital plant based applications this is every minute
For longer term digital mine applications like stockpile, this can be every shift
The uncertainty i.e. how sure the model is of the future varies from in each situation. Where applicable, brains.app models come with uncertainty metrics.