Yanomaly

• Robust and scalable AI-based anomaly detection and predictive analytics
• Specific algorithms for detecting various issues with different equipments: sensor drift, sensor errors, (impending) failures of electric drives and motor, pumps, valves, degradation of control loop performance, …
• Context-dependence of algorithms – algorithms take into account machine operating conditions (product types, startup vs regime) to enhance accuracy

Input(s):

• Various data sources – flat files, InfluxDB, REST API, OSISoft PI, MQTT

Main feature(s):

• Analytics (in particular anomaly detection) on machine-generated timeseries data
• Various visualisations for rapid analysis by non-data scientists
• Various browser-based UI screens for training, management and deployment of data sources and models
• Scalable architecture for robust deployment of hundreds of AI models

Output(s):

• Predictions, anomaly scores, root causes and alarm notifications to various data sinks

Yanomaly is deployed on Kubernetes, a scalable container orchestration platform. It can be deployed on cloud native Kubernetes service providers (such as Google Kubernetes Engine, Amazon Elastic Kubernetes Service or Azure Kubernetes Service) as well as on single node installations (such as k3s). Although single node deployments of Yanomaly sacrifice redundancy and scalability, it is still a valid mode for proof-of-concept, demo or smaller production setups. The installation procedure is tested for deployment to a server running Ubuntu Focal (20.04) and CentOS. Other versions from Ubuntu, starting from Bionic will probably also work. YFM will run on another Linux OS as well, as long as Kubernetes environment is installed. Other installation requirements:

• SSH access using public key authentication.
• Passwordless sudo on the target host (with !requiretty setting) Internet access from the target host. The installation procedure supports installations from behind an HTTP proxy. A list of endpoints to be whitelisted can be provided on request.

Consumption of predictions / notifications: Process engineers, maintenance engineers, operators, production managers. Building, maintenance of models: data scientists, process engineers.

Software license based on functionality and number of monitored tags.