Instrumented, connected assets generate volumes of operational data - structured and unstructured - that can be used to identify risks if the organizations have analytic tools to convey this insight to personnel responsible for asset operations.
In this journey, we'll show how to build and apply custom machine learning models to identify risks and suggest proactive maintenance to avoid service disruption. These models will also estimate how long a mechanical asset can be used before needing maintenance/replacement.
We'll also show how to import the custom models into a Maximo instance. Maximo is a system used for managing assets and workflow processes. This can be used to increase efficiency by automating processes such as work orders, notifications, anomaly detection, etc.
- Build custom machine learning model in Watson Studio, and export custom model as python package
- Publish sensor data from field assets to Maximo
- Periodically pull asset data from Maximo to Python server, and generate predictions based on packaged custom model
- Publish predicted "Remaining Useful Life" value to Maximo
- (Within Maximo) Create "automation scripts" to execute when "Remaining Useful Life" is updated. If RUL below X number of days, create a work order to have asset replaced / updated.
# OS X
brew install python python-pip
# Linux
apt-get install python python-pip
# Install python packages
pip install pickle scikit-learn flask
Create the following services
Navigate to your provisioned Watson Studio instance
Click "Get Started"
Create a project. Select "Standard" project, and give it a name.
Select your created project and click "New Notebook"
Now, you can create your own notebook for a custom use case, or use one of our predefined notebooks. Here, we'll upload our preconfigured "Advanced Regression" notebook, which can be seen in further detail in the notebooks directory of this project.
We'll do so by selecting "From File" in the menu, then entering a name, and the path to the local notebook file within this repo.
In this notebook, we'll be exploring a public dataset supplied by NASA. This dataset was generated by collecting sensor data from Turbofan engines. Each Turbofan has 20 sensors, as well as a "Cycles" field that signals how long a Turbofan has been running.
After going through each step in the notebook, we can then export our model for consumption by Maximo PMI. This can be done by following the steps in the notebook at notebooks/PMI-Custom_Model_Development.ipynb. You'll need the credentials that were included in your Welcome letter for the Maximo PMI instance.
Once this has been completed, we'll see the custom model available in the Maximo APM View.
As an alternative, we can also export the model by leveraging the pickle python package. This can be done by adding the following command to the notebook
pickle.dump(model, open("model.joblib", 'wb'))
Now that we have our custom model built, the next step is to register assets in Maximo. This can be done by navigating to your Maximo instance.
In the "Find Navigation Item" field, enter "Asset Templates".
Click "New Asset Template" in the left hand menu.
Give the Asset Template an ID (We used "Turbofan" here). We can also provide asset type, description, and other optional fields. Be sure to also enter "Turbofan" or some identifier in the description so we can use that as a search query.
Directly below, select "Add Row" in the "Meters" section.
In this case each meter will correspond to a sensor value. For simplicity, we'll add a single meter first, which measures the engine pressure.
Save the Asset Template. Next, click "Generate New Assets", and create one or more assets.
Next, we can search for our Asset(s) by entering "Turbofan" in the description.
Select an Asset, scroll down to the "More Actions" field and select "Enter Meter Readings", and then a value for each Meter.
These meter readings can also be updated via an HTTP POST request, we'll demonstrate this functionality in the next step.
Now, we'll leverage a set of python scripts to periodically query our registered assets, and collect the latest meter data. As each asset's associated data is collected, we'll load our custom model generated from Step 2 to estimate the remaining useful life of the asset.
In this particular implementation, we'll collect the asset data via the Maximo API.
To do so, we'll need to authenticate with a base64 encoded set of our Maximo credentials. These credentials are then provided in the following HTTP request as the "maxauth" header. We can query the API to get all meter readings associated with an Asset by sending a GET request to the following endpoint like so.
curl -H "maxauth: <base64 user:pw>" -H "Content-Type: application/json" ${MAXIMO_URL}/oslc/os/mxasset?oslc.select=assetmeter,expectedlife&oslc.where=assetnum=${ASSET_ID}
The database query is embedded at the end of the Maximo endpoint:
The oslc.select value determines which asset attributes will be returned (meter readings in this case)
The oslc.where value allows for us to only select data that matches a specific condition (matching our asset id)
After receiving the Asset data, we can estimate the Remaining Useful Life for our asset by loading our previous model using the pickle library
classifier = pickle.loads(load('model.joblib'))
predictions = classifier.predict(meter_data)
Finally, we can update the value for remaining useful life (expectedlife) in the Maximo database with the following POST request
requests.post("/oslc/os/mxasset/" + restid,
data={"expectedlife": predictions[0]})
The full Maximo asset query/update workflow can be executed by running the included server.py script like so
export TOKEN=$(echo ${MAXIMO_USER}:${MAXIMO_PASS} | base64)
python server.py ${MAXIMO_URL} ${TOKEN} ${ASSET_ID}
In conclusion, here we have demonstrated how to create assets in Maximo, query their associated meters, and pass the metered data through a custom machine learning model. This allows for organizations to predict of when their various assets may fail, and proactively take action beforehand.
This code pattern is licensed under the Apache Software License, Version 2. Separate third-party code objects invoked within this code pattern are licensed by their respective providers pursuant to their own separate licenses. Contributions are subject to the Developer Certificate of Origin, Version 1.1 (DCO) and the Apache Software License, Version 2.