Create Utilities Anomaly Detection Model

Maximo Predict comes with notebook templates to assist in streamlining data uploads to Maximo. This notebook will create the following resources using provided csv files:

Anomaly Score
Anomaly Score History

These instructions use the notebook named 3_PMI - Anomaly Detection -UnSupervised-HPU.ipynb file with the Substation Transformer for Health and Predict for Utilities Demo Assets. Note that this uses simulated Pump Data for the sensor readings.

In this exercise you will use Watson Studio and Health and Predict - Utilities to:

Upload the and Run the Anomaly Detection Notebook using a template to Train and Display an Anomaly Detection Model
Confirm the Data Has been uploaded for your assets

Note

You must complete the previous exercise for Setup Watson Studio before you start this exercise.

Pre-requisites

Review Predict documentation for the list of available models.
Ensure your MAS Predict environment is running and you have access. Try your server URL that might look something like: https://main.predict.ivt11rel87.ivt.suite.myhost.com/ibm/pmi/service/rest/system/info
Ensure you have Access to Asset data files for the Health and Predict Utilities Demo Data
Complete the Load Data into Manage lab for the Utilities data.
Complete the Load Utilities Health Scores via Notebook lab
Complete the Create Utilities Predict Group and Upload Sensor Data lab
Have the following information from the previous lab: Predict_Envs.JSON and Fast_Execution.JSON

Note

It is best to perform this lab in your own Watson Studio Project created using Setup Watson Studio instructions. If you are using a shared project, ensure you append each file uploaded with your initials and update the file paths in the notebooks to include that change.

Upload and Start the Anomaly Detection Unsupervised Notebook

Upload or open the Anomaly Detection template notebook to your Project. Use the steps from the previous exercise Add Notebook From File to a Watson Studio Project If you are using a shared project, rename the notebook template by prepending your initials to the template. If this is done, ensure any paths or file names within the notebook are updated as well. If you already have uploaded the notebook, open it with Watson Studio. Select the 3_PMI - Anomaly Detection -UnSupervised-HPU.ipynb notebook template.
Open the notebook. Click on the pencil icon next to your notebook
If the notebook fails to start, restart it. Click on the i icon , Environment tab, Running status drop down select box and choose Restart

Run the Notebook

Install the Maximo Predict SDK

Run the cell to uninstall the libraries. This is done to ensure the right version is installed later in the process.
Run the next cell to define the requirements and some environment variables to run this notebook. This cell also checks that the precusor notebook has been run
Run the next cell to define the API keys used to call Maximo Predict from the Predict_Envs.JSON file
Run the next two cells to import the os, trim the provided base url to be used when downloading pmlib in the second cell
Run the final 4 cells in this section to import the pmlib, srom, and logging libraries

Set up the Model Training Pipeline

Run the first cell to obtain the Asset Group ID from the JSON file produced in the Create Utilities Predict Group and Upload Sensor Data lab. Note that it prints out the Predict Group ID after being extracted.
Run the next cell to extract the device type and the desired features to be considered in the anomaly model.
Run the next cell to define the scoring strategy. There are three different scoring strategies shown to choose from.
Run the following cell to import all the anomaly detection models to be run through the pipeline and other functions necessary for this process
Run the following two cells to define the model pipeline for the determined features. This cell defines the settings to be used within the model algorithm during the training process.
Run the next cells to define the model group containting the pipeline, and the asset group id that the model will be trained on

Train, Register and Enable the Model

Run the first cell to train the model. Some models take time to train.
Once that process is complete, register the model to your asset group by running the next cell
Finally, Run the next cell to enable it and determine how often it will be run in monitor. This is the last cell to be run in this notebook.

Confirm Model Registration

Navigate to Maximo Health and Predict for Utilities within your environment
Use the left-hand menu to go into Predict Grouping
Select your asset group
Click into your asset group and ensure you have the Anomaly Detection Model listed under Trained instances registered for this group and select an asset to go to the Health Dashboard
Scroll down and expand the Predict section to ensure the anomaly detection history and the anomaly detection score is visible.

Predict Grouping

Note

Recall in the Create Utilities Predict Group and Upload Sensor Data lab only some assets have sensor data. If an asset does not have sensor data, it will not have Predict data

Congratulations you have created an Anomaly Detection model and associated it to your assets!