Incorporating machine learning (ML) techniques into real-time digital twins takes their power and simplicity to the next level. It can be challenging to write analytics code that sifts out subtle issues emerging over time in a stream of telemetry. In many cases, the underlying processes which lead to device failures are not well understood or are complex to describe. However, application developers can train ML algorithms on thousands of known patterns to recognize abnormal telemetry. Automatic retraining allows digital twins to continuously improve ML algorithms as they process real-world data.
Generative AI from OpenAI takes real-time monitoring to the next level. It can assist users in visualizing data and checking for anomalies, avoiding the need for managers to constantly watch dashboards for emerging issues.
When ML algorithms detect issues, digital twins can send alerts to many public alerting providers, including Splunk, Slack, and PagerDuty.
Developers can select from a wide variety of binary classification algorithms in ML.NET and TensorFlow and evaluate them before deployment.
Digital twins can optionally make use of straightforward business rules to extend their functionality when incorporating ML algorithms.
ScaleOut’s digital twins incorporate Microsoft’s popular machine learning library, ML.NET, or TensorFlow to run ML algorithms. Using the ScaleOut Machine Learning Training Tool™, developers can select, train, evaluate, deploy, and test ML algorithms for use in real-time digital twins. Once deployed, the ML algorithm runs independently for each data source, examining incoming telemetry within milliseconds after it arrives and logging abnormal events.
ScaleOut’s ML algorithms can perform spike and trend detection for a single parameter or anomaly detection that looks for more complex patterns across multiple parameters. For example, after training with data acquired from live readings, anomaly detection can look at several parameters from a motor or pump to detect abnormal behaviors that might otherwise go unnoticed.