iImprove is a model-predictive optimization design tool for Intellect. Model-predictive optimizers are useful when you have good historical data about your performance and what drives it and you have control over at least one of those key drivers. The models can "inverted" to determine what value of the controllable key driver to apply to your process in order to achieve a desired performance. iImprove uses one or more models created with the Expert tool to find values for one or more setpoints, given current conditions on one or more uncontrollable inputs in order to maximize, minimize or target a set of objectives. You run off-line what-if analysis to validate your solution and perhaps just use iImprove for off-line optmization, but you can save the optimization strategy to disk and put it on-line in the Intellect Server. The iImprove optimizer then can run live, optimally controlling complex non-linear processes in real-time.

Build Optimization Schemes

iImprove "inverts" models to perform an optimization. Inverting a model means searching controllable input values while using current fixed values of uncontrollable inputs to achieve desired objectives. A controllable input is one that you can change in your operations, such as a setpoint on a process, or the amount of ingredients to put in a batch. An input that is not controllable is one that you don't, or cannot, directly change.

The models that are inverted are modeling systems built with Intellect Expert, including all the data pre-processing that may be used inside it and across the ensemble of models in the modeling system. Usually, the raw data going into a modeling system is the actual conditions coming from your sensors. That data might be then "transformed" inside the modeling system and used by the models. The predictions or estimates made by the models are merged to create a "System Model". So, the controllable raw input data variables are searched, internal pre-processing occurs and the output of the System Models is what is optimized.

"Optimized" means controllable input values are manipulated within hard constraints and at settable increments to seek the maximal value, minimal value or a target for the output(s). This finds setpoint values that achieve, or come as close as possible to, a desired result.

iImprove can use more than one modeling system at a time, thereby enabling you to seek multiple objectives at the same time. For example to target a quality metric while maximizing another result while minimizing waste. Or, maximize production while minimizing undesired by-products while achieving some target as well.

iImprove also supports costs and desirabilities. Costing is a linear function, with a fixed base cost and a cost in proportion to the value of the controllable input. Desirabilities mean when given alternative choices, you desire a value that is low, or high or near a value.

You can save your solutions for future use on the desktop or for use on-line within an Intellect 3.0 Server via an "iImprove Task".

When running on-line, current data is passed to the "iImprove Task", which "plugs in" the current values into the uncontrollable inputs, then it runs an optimization and broadcasts the best solution found. A downstream task may interpret that solution and send it to another task that writes the setpoints (controllable variables' values) to some destination, such as a control system via OPC.