Queries on Synthetic Images for Large Multivariate Engineering Data Base Searches

Abstract

Engineers are often interested in searching through historical engine testing data sets to explore systems behaviours across experiments or simply to validate the integrity of data by reference to past experience. Engine tests usually consist of design of experiments (DoE) tests carried out at different operating conditions on an engine. While DoE tests record a time or measurement field so could fall under the multivariate time series domain, however, the engineering experiments have no important connection with order. Using this knowledge we have developed a method of visualising the engineering data as a set of normalised histograms to manufacture synthetic images for every test. From these images and their pixel rows we propose a list base indexing method to efficiently index the engine test images and a fast algorithm to search both the univariate and multivariate cases. Querying the data for similar behaviours thus becomes much simpler for the engineer, compared to manually searching files. Our data is of a heterogeneous form which can often mean the absence of columns where they are present in other data sets. It can even sometimes mean the comparison between different data types. This data inconsistency is something that from our research is rarely reported or attempted to traverse, but is consistently occurring in real multivariate data case studies. In order to overcome this realistic problem our method will allow for comparison of univariate data with differing quality scores.