Smart EngineeringAutomated anomalies detection for the IoT Ecosystem
Using Machine learning and closed loop Streaming Analytics to detect Anomalies
Using machine learning to detect early indicators of failureUsing advanced machine learning algorithms and a closed-loop streaming analytics with model update framework we are able to train several models to detect anomalies in the underlying time-stamped data streaming into the system.
Automatic upgrade of modelsModels are continually monitored and automatically updated, without operator intervention.
End to End SystemThis closed-loop system (device -> gateways -> edge analytics appliance -> server/cluster/cloud) is critical to the success of large scale deployment of IoT and Smart Cities/Nation where there would be thousands and millions of streaming data and it would be impossible to find and hire a large enough pool of data scientist to build models by-hand.
Scalable IoT DeploymentAn automated self-learning, model-building and updating system is the best way to scale an IoT deployment.
Current methods for detection of failures or anomalies are based on legacy Statistic Processing Control (SPC) charts or arbitrary thresholds set by the operator. This method fails to detect anomalies happening within the threshold or +-10% or 25% boundaries typical in SPC methods. The method is not suitable for detecting failures about to happen.
Instead of comparing with pre-determined fixed limits, a machine learning algorithm can detect patterns in degradation of performance, and learn the optimal time for maintenance. This ensures the optimal balance between maintenance cost, production quality and uptime of the production line.