Creating anomaly detection scores
Anomaly detection in Maximo Predict and Maximo Health helps identify unusual patterns in the behavior of the asset, which might indicate potential failures or pre-failure behaviors. To detect anomalies and take appropriate action, unusual patterns, trends, events, and outliers are scored against normalized scores.
Before you begin
- Ensure that your environment in Maximo Predict is running and you have access to Watson™ Studio.
- Ensure that you have access to asset data files.
- Load historical asset class sensor data into Watson Studio from local storage in IBM Cloud Pak® for Datafrom a cloud object storage on IBM Cloud® or from the data loader by using App Connect.
About this task
Prepare the CSV files to in the correct format for historical data. For more information, see
data by using notebooks.
- Provide two data sets for anomaly detection: normal data for training and validation data for anomaly scoring and evaluation.
- Train and score models.
- Collect data.
- Select input format for each type of data.
- Format data as CSV files.
- On the Assets details page, find the missing values.
- In Watson Studio, find the correct project.
- Duplicate the notebook.
- Click View all assets.
- Browse or search for the notebook name.
- Select a supervised, unsupervised, WS, or PMI notebook.
- Copy and rename the notebook in the new environment.
- Upload data from the notebook and check its quality.
- Insert the name of the data file.
- Run each cell.
- Change column or header names to match.
- Set the threshold by giving the model some criteria.
- Complete missing values.
- Run and train relevant algorithms.
- View the best pipeline.
- Use the recommended pipeline to create the PMI notebook in Maximo Monitor.