What is Knowledge extraction?

Knowledge extraction is the overall process of deriving valuable information from data. It involves several steps, including data cleaning, data selection, data mining and knowledge visualization.

knowledge extraction in the context of OPTIMIZATION

In the context of Flow Simulation and Optimization, Knowledge extraction seeks to extract valuable information from an experiment or optimization result. An experiment or an optimization often seeks to find a spefic result, for example, “Which is the solution with the highest throughput/h?” (given specific experiment parameters). Knowledge extraction on the other hand often aim to find more general findings such as “What do the top five procent of the solutions have in common?”.

applied on the result

Knowledge extraction do not normally interact with the Flow Simulation model or the Optimization. Instead Knowledge extraction is applied to the result of a Flow Simulation or an optimization.

large data sets

Knowledge extraction works best with large data sets. Therefor, there is no value in using knowledge extraction to a single Flow Simulation result. Knowledge extraction needs several different simulation runs to be able to extract any knowledge. Therefor it is used at optimization results or results from large experiments.

methods

There are several different methods within Knowledge extraction which can be used. One of the more simpler methods is to visualize all results in a parallel diagram where settings for each parameter can be set to filter out interesting results. Many software has the possibility to apply more advanced techniques such as different rule extraction or pattern identifying techniques.

Extract “rules”

A common goal of Knowledge extraction is to extract rules from optimization results. This typically involves selecting an area of interest from the result space, such as the top 10% in terms of throughput/h and the bottom 20% in terms of Work in progress (WIP).

In the choosen result space, a scan is made to identy similarities in the experiment parameters between the different results. These potential similarities becomes the extracted rules. The extracted rules could for example state that, the first machine requires an availability over 95%, or that the buffer between the third and fourth machines should contain between 12 and 16 slots.

Patterns

Another method used in Knowledge extraction is to automatically identify different patterns in the result space. These patterns are often used to group results into different categories that can be analyzed further, both manually and in combination with other techniques.