New Features
Model File Versioning
The user is now able to upload multiple file versions to a single AI Model. Each file can be hot-swapped at any time and can contain a version number and a note about the file.
For more info, view the managing model files article.
Model History Backfill
Previously, when modifying a model or binding, the user could have to wait for historical data to collect. This is due to model historical data being kept in a high availability database as opposed to the longterm history. In this version, when the user does one of the following actions, the longterm database is queried and the appropriate amount of historical data is inserted into the high availability database:
Swaps to a model file that contains a greater required depth.
Changes an input binding to a different tag.
Increases the sample rate of a model.
Changed Device Reconnect Behavior
The keep alive attempts and max timeout scans parameters have been removed. The device will now attempt to reconnect at the set scan rate of the device. If the connection attempt takes longer than the scan rate, the device will flag it as an overscan and a connection attempt wont be made until the previous attempt has completely failed.
Historical Data Compression
Options were added to allow data to be stored more efficiently. The following two settings have been added:
Minimum Percent Change: When a tag value is added to long-term history, the value must move by this percentage before being stored again. This can greatly reduce storage demands on devices with fast scan rates. Set the value to zero to disable this feature.
Minimum Time: If a tag value is not moving much and is not being historized due to the minimum percent change setting, this setting will cause a "confidence sample" to force the value to be stored after an amount of time has passed. Set the value to zero to disable this feature.
Lookup Table Fallback
A lookup table can now have a fallback value that is used if none of the other cases are true. To use this, put an asterisk in the lookup field.
Quick Build Model Trend
From the model detail page dropdown, you can now easily generate a trend from the model’s bindings. If the bindings are changed, go to the trend and use it’s dropdown to refresh the trend’s tag traces.
Maximum Logger Debug Time
Leaving a logger in debug mode could quickly fill up your machine’s storage. Now there is a maximum time allowed for leaving a logger in debug before it will return back to the “Warning” level. This value can be modified from the logs dropdown menu.
New Services
Previously, the Data Collector handled many of the critical system services. These have now been broken out into there own services capable of being started and stopped independently, and with there own loggers:
Parameter Mapping
Performance Monitoring
Heartbeat
Model History Cleanup
Restructured Error Messages
Each object now has an error message and an error detail where the error message contains a brief summary of the issue and the error detail contains the raw error text. This allows for quicker debugging and a cleaner user experience.
Miscellaneous Improvements
Trend trace values and Local Values lists now update in realtime.
Can now set performance alarm limits from the performance page.
Added the ability to change the UI inactivity timeout from a new “System Settings” page.
An AI model now has the ability to be duplicated using the dropdown menu.
Added the ability to choose normalized or scaled data from the AI model detail quick trend.
Various QoL improvements.
Bug Fixes
Deleting an enabled device would cause the Data Collector to throw an error continuously. The Data Collector now properly cleans up deleted devices and there is a requirement to disable a device before deleting it.
ODBC drivers 17 & 18 were properly added to the container install. Versions 11 & 13 were removed and are no longer supported.
Known Issues
When backfilling history, there can be a race condition that causes the model to fail for a single scan before the new data has been backfilled.
Released 11/07/2024