We present a data validation tool called UDAVA (Unsupervised Data Validation), that automatically discovers process behavior patterns in sensor data for AI-enabled Industrial Internet of Things applications.
Watch the video above to learn how the tool works and how to use it.
UDAVA is a tool using unsupervised machine learning to process time series data. The tool extracts feature vectors from subsequences of the time series, and performs clustering on these vectors. The time series is then labeled based on the clusters discovered by the machine learning model. UDAVA also computes a deviation metric that indicates parts of the time series that has features which deviates from the norm. This can be used to discover anomalies and errors in production.
UDAVA is developed in Python and includes an unsupervised machine learning pipeline that is orchestrated using DVC (Data Version Control). The source code of UDAVA can be found on GitHub.
Download the tool by cloning the GitHub-repository:
git clone https://github.com/SINTEF-9012/Udava/You can install the required modules by creating a virtual environment and install the requirements.txt-file (run these commands from the main folder):
mkdir venv
python3 -m venv venv
source venv/bin/activate
pip3 install -r requirements.txtStart the server by running:
python3 src/api.pyThe web interface will run on localhost; use the URL displayed in the terminal when you start the server. In this interface, you can both create new models and use existing ones for inference. For more information, watch the video above.