FactoryTalk Analytics LogixAI software module adds machine-learning intelligence and lets users create better models for control variables and predictive analytics without requiring data science skills.
It's always a big moment when the spark catches, when students go from passively taking in facts and figures to when their curiosity ignites, and they start asking questions and seeking answers on their own. Software apparently isn't immune to this spark, and many computer programs are incorporating machine learning (ML) and other forms of artificial intelligence (AI) that let them take in new information, and learn, grow and change in response to it.
In the process control and automation realm, one of the latest instances of ML and AI is the FactoryTalk Analytics LogixAI add-on module from Rockwell Automation, which adds further intelligence to its ControlLogix platform. It employs machine learning to monitor processes and triggers a bit in the controller when a problem is predicted to occur. Without requiring a data scientist, LogixAI can build predictions in five steps, including: configuration and identification of data, automatic model generation, training, monitoring and integrating predictions.
“LogixAI makes predictive analytics more accessible to help users make better production decisions,” says Jonathan Wise, product manager, Rockwell Automation. “The module learns from their ControlLogix applications, tells them when things are changing in unexpected ways, and helps them get ahead of quality issues and protect process integrity.”
For instance, the module can help operators spot performance deviations in equipment like mixers that could affect product quality or lead to downtime. It can also be used as a virtual sensor, so instead of workers taking a manual reading, like the humidity of a packaged food product, the module can analyze variables from line assets like sprayers, dryers and burners to predict a measurement, virtually. Workers can then be notified of problems by integrating LogixAI’s user-defined tags (UDT) to further configure alarms on an HMI or dashboard.
"LogixAI takes the ControlLogix platform to the next level beyond control by becoming a way for controls engineering to deploy machine learning in chassis,” says Jennifer Mansfield, product marketing manager for Rockwell Automation's embedded analytics and mobile applications development team. "The module lets users create models for control variables without having data science skills."
Users employ a web-based interface in LogixAI that lets them describe the process and prediction they want to make by adding key control, state and manipulated variables. This information is then used to generate a UDT they can use to interact with the module in a Studio 5000 software environment from Rockwell Automation. The module has two primary operational modes: Operational Monitor performs anomaly detection by creating a model of normal operations and detecting variances; and Value Estimation works as a soft sensor to create a model using existing data to estimate a value that can’t easily be collected or measured.
"We simplified the data science process to allow users to monitor controller data and make predictions about key variables that they can act on," says Mansfield. "In its first release, LogixAI monitors, detects and triggers if there's high confidence that an anomaly will occur. The algorithm works with streaming data and, based on first principles of unit operations, identifies a physical representation of the operation, which is used as the predictive model.
Mansfield adds that LogixAI creates a platform for cyber-physical systems and lets users apply machine learning to control systems without investing in costly data science. "Why not solve problems on machines if you already have the primary source of data in the controller?" she asks. "The other advantage of LogixAI is it doesn’t require a connection to the cloud or an outside data store, so it's easy to install and easy to configure, working with the control-level technology you already have in place. Users can build and train their model, monitor the process, and learn if there's any anomalous behavior."
Click here for more information.
