Why should you use MLnext Execution instead of your own machine learning solution?
The use of MLnext Execution offers the following benefits compared to custom programmed algorithms for anomaly detection:
- Parameterization instead of programming - Through configuration files, the data flow can be flexibly set in the given parameters.
- Cloud / Edge / Premise execution - The solution can be executed in different cloud applications, on edge or on premise without customization of the algorithms.
- Isolated execution - Each algorithm gets its own runtime environment, so the processes do not interfere with each other, which ensures stable execution.
The architecture features:
- a periodic execution service of pre-defined workflows
- a web interface to monitor and control workflows
- a prediction endpoint provided as a REST API
A workflow is defined in a three-tier hierarchy. A task defines global namespace for jobs. A job contains a list of steps which are executed in interval. A step is a basic operation such as:
- data collection from databases (e.g. MySQL, MSSQL, InfluxDB) and MQTT-based services (e.g. Kafka)
- data preprocessing with scikit-learn pipelines with modules from MLnext, DaskML and scikit-learn
- processing with ML models written in Tensorflow
- result evaluation with custom evaluation strategies
- result saving to databases (e.g. MySQL, MSSQL, InfluxDB) and MQTT-based services (e.g. Kafka)
The MLnext Framework is part of the "Digital Factory now" campaign by Phoenix Contact Electronics to support the solution portfolio "Anomaly Detection".