Improved throughput and uptime


Batch and process manufacturing use case

Without Falkonry

A mineral production company faces a common situation: instrumentation and data collection are producing large volumes of operational data, but efforts to turn this data into meaningful improvements in operational efficiency are falling short.

The production line experiences frequent, unexpected downtime events due to variations in raw material that adversely impact a critical process-line machine. These downtime events last anywhere from 2 to 24 hours per occurrence, costing $30,000 per hour and $720,000 a day.

Data, in the form of motor currents, temperatures, valve settings, and stoichiometric measurements are collected from the process line, stored in a process historian (OSIsoft PI), and are made available to the operations team through dashboards and other means. The thresholds, rules, statistical and engineering models that are being used are, however, unable to reliably identify conditions leading to downtime events.

Applying Falkonry ready-to-use machine learning

A Falkonry server is installed on-premise, and a Falkonry-supplied integration agent is used to connect it to the customer’s OSIsoft PI System.

Members of the process operations team are given a < 3 hour training on the Falkonry products and then complete the following tasks:

  • Configure a Datastream for pattern recognition (using the Falkonry/PI Agent UI)
    • Select the set of entities (elements in PI) and signals (attributes/tags in PI) for which pattern recognition is desired. Here there is a single entity (PI Element) corresponding to the process line, and 7 signals (PI Attributes) corresponding to selected motor currents, temperatures, and valve settings along the line.
    • Identify a window of historic data to learn from. This window is chosen to include example periods of known conditions such as the downtime event.
    • Create a PI Attribute/Tag to hold the Assessment results produced by Falkonry.
  • Find patterns in the signal data
    • Train an unsupervised model using the Falkonry UI
  • Add Facts corresponding to a few known examples in the history window of known downtime events and periods of normality.
  • Identify possible patterns that could indicate a bad raw material condition leading to the downtime event.
    • Label the conditions and create an updated (supervised) model based on the supplied Facts.
  • Test the updated model on other parts of the history window data, and confirm that model detects bad raw material conditions in advance of downtime events.
  • Set up notifications in PI to alert operators of a bad raw material condition.
  • Turn on live monitoring of the Datastream.

The Assessment stream produced by Falkonry is able to provide early warning of previously-hidden bad raw material conditions. This condition awareness enables the operations team to take corrective actions and avoid many of the costly downtime events that have plagued them previously.

Ready to discuss your use case?