TrendMiner 2021.R2

12th of May 2021

The 2021.R2 release contains the following new functionality and improvements:

Presenting Machine Learning Model tags 

In the 2021.R1 release we introduced Embedded notebooks within TrendMiner, which allowed the Analytics Expert user to have more flexibility beyond the robustly packaged functionalities that TrendMiner is known for. You can now, based on data prepared in a TrendHub view, conduct advanced analytics via python scripting and create yet more advanced visualizations, which can be operationalized by embedding them within a DashHub dashboard and make them available to an entire organization.  

With this release, we are extending the capabilities to operationalize the notebook work by deploying custom created models into an embedded scoring/inference engine and making the model outputs available as Machine Learning model tags. As such, the model outputs are available for all TrendMiner users, as if they were tags originating from an enterprise historian or any other timeseries data source, unlocking ALL existing TrendMiner capabilities. For example: visualizing recent & historical data, searching for patterns or threshold values as well as monitoring. 

 

blobid0.png

To illustrate the potential of this solution, this release will also include a TrendMiner-created anomaly detection model. This model can be trained to recognize the normal behavior of a continuous process based on a provided for TrendMiner view with historical data. The outcome can be applied to recent / live data to detect anomalous process behavior, indicated by a high anomaly score (i.e. the model output). 

The minimum system requirements in order to use the new Machine Learning Model tags and/or run the Embedded Notebooks functionality released in our 2021.R1 are 48GB of RAM, i.e. an additional 16GB RAM is required on top of the minimal system requirements of 32GB of RAM. 

New Functionalities 

  • Asset specific context types: Additional configuration for the context types in the platform, letting administrators decide which context types are available for a certain part of the asset tree. 
  • Component filter in ContextHub adds support to easily include all context items related to a unit or equipment in your views. 
  • Pinning rows on the Gantt chart allows you to specify which components, types or fields should always be displayed, regardless of there being content available in the timeframe you are viewing 

Application Enhancements 

  • Improved search result and chart data export. 
  • Improved disablement strategy for monitors that better indicates system disabled monitors and allows for health-checks to re-enable these monitors. 
  • Improved DashHub user experience including a more compact tile view, presentation mode and scrollable dashboards. 
  • DashHub’s ‘Trend tile’ now lets you enjoy all the features like: stacked plot mode, grouping, axis visibility, scatter plot mode with histograms, etc. Well known tools in TrendMiner. 
  • Gantt chart ordering by drag and drop to move types, field or complete component block on the Gantt overview. 
  • Parallel connections: In ConfigHub it is now possible to configure the number of parallel connections per data source, giving the option to limit the load on low capacity data sources and increasing throughput to TrendMiner for high capacity data sources. 
  • Various scatter plot improvements. 
  • Support for TrendHub NextGen views in Notebooks. 
  • Miscellaneous security improvements. 
  • Full support for (latest version of) the Microsoft Edge browser  (as a replacement for support of Internet Explorer 11, which will be dropped in the 2021.R3 release) 
  • TrendMiner symbol for PiVision was updated to support PiVision 2020 installations 

Bug fixes 

  • Solved an issue where the recommender engine returns inaccessible tags as a result.  

Known issues  

  • Cross asset value based search is currently not supported (and blocked) for string tags (i.e. digital tags for which no numerical value mapping exists on the historian). Digital tags for which a numerical value mapping exists on the historian are supported.  
  • After restoring a backup the tm-zeppelin and the tm-zementis service need to be restarted manually. 
  • Scooter values may change when zooming in after adding the scooter. This is due to the fact that scooters currently interpolate the data points on the chart. 
  • The back button of the browser does not correctly restore the panel state.  
  • When calculated tags are deleted, they cannot be created with the same name again. 
  • When the trigger of a fingerprint monitor is deleted, the monitor will go to system-disabled. Running the health check will re-enable the monitor, but it will switch to Detect Matches as opposed to Detect Deviations 
  • Cloning a notebook will result in a copy of the notebook. The paragraphs of the copied notebook are still interlinked with the paragraphs of the original notebook. The "linked" paragraphs of both notebooks may be refreshed/updated based on the other notebook resulting in undesired behavior. It is advised to copy/paste your code manually for now. 

Important:   

  • The TrendMiner symbol for OSIsoft PiVision was updated to support installations of PiVision 2020. Starting in PiVision 2020 the Pi Web API is not installed by default anymore.  
    For existing customers, to continue to use the plugin, install the Pi Web API component to on your environment and upgrade your TrendMiner symbol to the latest version. 

Important:   

As from our 2021.R3 release, the minimum system requirements will change:  

  • vCPU 8 (Recommended: vCPU 16) 
  • RAM 64 GB (up from 32 GB) 
  • Disk space 100 GB (Recommended: 500 GB) 

Note: In case you are foreseeing issues in (timely) upgrading your system resources, please contact your CSM for more information, as we will make sure that the TrendMiner core functionalities will still fit 32GB RAM installations, albeit some functionalities might become limited or suffer from degraded performance. 

As from the 2021.R3 onwards, we will drop active support* for Internet Explorer 11 browsers, following Microsoft's recommendation to only use Internet Explorer for compatibility with older sites and applications. Ending support for IE 11 will allow us to use modern web practices to improve performance and user experience as we continue adding new features. Support for Microsoft Edge browser has been added instead.  

* users will still be able to use IE11 with TrendMiner, but new features might no longer work + IE11 specific bugs/issues will no longer be solved. In a later stage, existing functionalities will also be adjusted to no longer have to cope with the restrictions of IE11.

Synopsis functionality

Machine Learning Model Tags 

With the Machine Learning Model tags, TrendMiner now offers the capability to operationalise machine learning, AI and predictive models by deploying the models and scoring incoming data in real-time and storing the model output as a new type of tag in TrendMinerThese Machine Learning Model tags can be displayed and analyzed in TrendHub as any other tag available in TrendMiner. 

You'll use the Predictive Model Markup Language (PMML) standard to import and deploy predictive models via an embedded predictive analytics scoring/inference engine that is included within the TrendMiner installation, referred to as Zementis. 

blobid1.png

The deployed model will become available inside the Machine Learning Model Tag Builder menu which enables you to share the output of your models as a tag inside TrendMiner. 

blobid2.png

blobid3.png

Note: Creation of Machine Learning Model tags is only available with the appropriate license and by users assigned the necessary permissions.  

TrendMiner Anomaly Detection Model 

TrendMiner offers a custom model for multi-variate anomaly detection via its notebook and machine learning model tags functionality. The TrendMiner Anomaly Detection Model can be trained on a TrendHub View containing normal operating conditions of your process. After learning the desired process conditions, the model will then be able to detect anomalies on new incoming data. 

The model itself can be trained inside the embedded notebook functionality after loading in a TrendHub View as a DataFrame. 

blobid4.png

blobid5.png

Once the model is trained, new data can be evaluated. The model will return whether a new datapoint is an outlier or not based on a given threshold (anomaly class) or return an anomaly score. The higher the anomaly score, the more likely that the datapoint is an outlier. 

blobid6.png

Asset specific context types 

To make the creation of context items easier, you can now assign certain types to a specific set of assets. This results in a filtered list of types when creating new context items for an asset. In the context configuration section, you can find the asset specific types section. The settings can be changed by selecting an asset from the asset browser. By default all types are allowed but you can set a custom configuration. This configuration will be applied on this asset and all underlying assets.   

blobid7.png

Component filter in ContextHub 

When selecting components in the component filter of a context view, you can now choose to select only the selected component (component only, this is also the default) or all underlying components (include children) or ascendants (include parents).  

blobid8.png 

Pinning rows on the Gantt chart and more 

In the Gantt chart, you can now pin rows so that they are always visible in your Context view, even if there is no data available at that moment in time. You click the pin icon to pin a row, or via the component/type add additional rows that are not visible.  

We also made it easier to change the order of the different rows, simply drag and drop the rows in the position that you want. 

blobid9.png

Application enhancements 

Improved disablement strategy for monitors 

A new state has been introduced for monitors to indicate issues with the underlying tags. From now on, monitors can be in one of the following states: Enabled, Disabled, or System-disabled (see screenshot below). 

blobid10.png

Whenever issues are encountered during the automatic monitoring process (e.g. tags not accessible, incorrect minimum duration, trigger of fingerprint monitor no longer existing), the respective monitor will be set to System-disabled. This makes sure users always have a correct overview of which monitors have been enabled by themselves and which of them are in a healthy state. For each monitor that goes to System-disabled, a notification is sent out describing the root cause. TrendMiner will no longer check for results for System-disabled monitors.

Whenever the underlying issues regarding tag permission and tag accessibility are restored, the user can re-enable all System-disabled monitors at once by clicking the health check button blobid11.png in the header. Whenever the health check is finished, a notification is sent out indicating the number of monitors that could not be enabled.   

blobid12.png

Note: It is highly advised to run a health check whenever the underlying issue is resolved. This will make sure all events, since the monitor went to system-disabled, are still captured and context items (if set up), will be backfilled. In case a user manual disables and re-enables the monitor again, all missed events will be lost. 

Improved DashHub user experience  

blobid13.png

We added some features to DashHub to improve your experience. We changed the basic design to enable the addition of more tiles on 1 screen by making everything more compact. New dashboards are also scrollable now, so you can add up to 100 tiles to 1 dashboard.   

Important note: your current views won't be scrollable after the upgrade. You can, and we recommend, to migrate these views manually to a scrollable dashboard. This is a manual step because this change can result in some small rearrangements in your dashboard.  

We also added a presentation mode to our dashboards, removing all menus and toolbars, to facilitate use in control rooms on large screens or embedding in 3rd party applications (Tip: by simply adding ?mode=presentation and the end of the url, you can trigger the presentation mode without any manual interaction). 

DashHub's trend tile 

Exclusively for TrendHub beta views, we released a new and improved trend tile. It has now the same look and feel as the same view in TrendHub beta. The tile is a read only visualization of your trend view, including a stacked or scatter plot mode. Also features like hidden axes and grouping are now available on your dashboard.  

blobid14.png

 

Parallel connections 

To improve the internal performance of TM-datasources, it can open multiple connections to the connector per data source.  

Configure global default 

The global default parallel connections is configured in ConfigHub. Changing the value will require a restart of the tm-datasource service (a popup is shown forcing the user to restart or cancel the change). 

Changing the global default will not update the existing overrides. They will keep the current value set as override. 

blobid15.png 

Override per datasource 

Overriding the value per data source can be done on create or edit. While setting the value on create tm-datasource does not require a restart, changing the value on edit requires a restart of the tm-datasource service. The value can be switched back to the global default at any moment. 

blobid16.png 

 

Setting the correct value 

We do not recommend changing the settings when the system performs as expected. 

Setting the sweet spot of the parallel connections is dependent on the hardware of the TrendMiner system and remote connector. Setting the number of connections too high might cause the system to slow down. Some trial and error over a longer period of time may be necessary. 

Scatter plot improvements 

Improvements have been made with the scatter plot.   

We updated the position of the tags / attribute label to be visualized next to their corresponding axis when the histogram option is disabled. You can now also update one of the tags / attributes to another active tag / attribute using the labels which are positioned next to the axes or inside the histogram tile (if the histogram option is enabled) without returning to the scatter plot overview. 

Next to this, we also moved the correlation values when the histogram option is disabled. The correlation values are, in this case, displayed below the focus plot to maximize the single scatter plot on the screen. You will also notice that the right hand side label is moved under the single scatter plot. 

blobid17.png

blobid18.png

Although the search functionality is not yet present, we have already added the option to start drawing on a single scatter plot. The draw functionality can be enabled when clicking the 'Draw' button next to the 'Return to overview' button. The draw area itself has some improvements under the hood. Once an area is drawn, it can easily be adjusted by drag and dropping the corners to another desired point on the chart.  

Removing a drawn area is now done via an option which pops up when hovering over the area, and is no longer removed from your plot when accidentally clicking on the scatter plot again. Next to these single scatter plot improvements, the multi scatter plot already has some additional adjustments. The axis text/value will be less likely to overlap with other text/value.  

Finally, the title of each individual correlation tile "Correlation and equation" has been removed to provide a cleaner look on the multi scatter plot overview.