Data Sampling

Data Sampling

DOEs are typically used to collect new data from a system. In many cases, sufficient data has already been collected. Often in these scenarios, the data collected has been accumulated over long periods of time, and there is enough data that analysis is simply intractable. For example, health monitoring data from sensors on fielded components may capture live data continuously over the entire operating life of the component. SmartUQ’s data sampling tools can divide the data to mimic a space-filling DOE consisting of subsets of the full data set. Unlike DOEs which are developed before data collection, data sampling like subsampling and sliced sampling takes existing input-output data pairs and selects the points that will represent the design space well.

Workflow of the Data Sampling Tool. Data Sampling Workflow
The workflow of data sampling tool starts with a collected data set from simulation and/or physical data. The data set tends to be large, and using the whole data set to perform analysis may be computationally demanding. SmartUQ data sampling tools can divide the data set in a smaller subset that represents the design space well.

Subsampling Algorithms

Subsampling algorithms sample a user specified number of points from an existing large data set to adequately represent the original data set. Unlike arbitrarily dividing a data set in two, the subsampling tool considers points that will mimic a space-filling DOE, and thus reduces the potential bias in the subsampled data set. By only using a subset of the initial data, the computational burden is significantly reduced, while the ability to accurately perform advanced analytics is maintained through the intelligent selection process. The larger, remaining subset of the data can be used to perform model validation, thus saving simulation and testing resources while remaining confident in prediction accuracy.

Subsampling Process Subsampling Process
Subsampling an existing data set where the subsample is mimics a space-filling DOE.

Sliced Sampling Algorithms

Like the sliced design, the sliced sampling algorithms divide the data set into groups. However, the initial data set is no longer assumed to be a space-filling design. Each slice can be used to build an emulator or perform model validation.

A unique emulation process enabled by the slicing structure is the divide and combine method of emulation. Each slice that is used for training has its own emulator, and then all the emulators are combined into one emulator for the final results. This allows for parallel computing and reduces the memory requirements necessary to perform advanced analytics of large data sets.

Sliced Sampling Process Sliced Sampling Process
Slicing an existing data set into four slices where each slice mimics a space-filling DOE.