Similarity Search of Time Series Data (SIFT)

The goal is to test and evaluate a technical solution for data mining and similarity search in large multivariate industrial time series in a real-world environment.

The project "Similarity search For Time series - SIFT" will test and evaluate a solution for making large amounts of multivariate industrial time series more searchable, a solution developed in the predecessor project DFusion. 

The solution has been inspired by embedding models used in natural language processing, but instead of embedding text, time series are embedded. The embedded time series vectors can then be stored in a vector database; a database endowed with similarity search functionality. This will, in turn, enable anomalies to be detected, events to be compared, and simplify searching in data that is usually difficult to search. T

he solution will be tested in real manufacturing environments, both at cable manufacturer Nexans in Grimsås and at Nord-Lock in Mattmar, which manufactures locking washers. Both companies have multiple projects and initiatives that involve collecting large amounts of data. This will require new methods for analysis and visualization, and the solution in the SIFT project is an important component. 

Nexans and Nord-Lock are driven by continuous improvement, and the SIFT-project will help  them improve productivity, reduce waste, and understand root causes of unplanned stops. Sensor manufacturer IFM will also participate in the project, where their monitoring software Moneo can serve as a platform for further development. The research will be conducted by RISE and Chalmers and builds upon results developed in the DFusion project.

The research will be conducted by RISE and Chalmers, building on results developed in the DFusion project.