Skill Level: Any Skill Level

This is the first recipe in a series of 6, showing you how you can build predictive maintenance models with Maximo APM - Predictive Maintenance Insights, using the data science process, CRISP-DM.


Welcome to the first in a series of recipes about predictive maintenance that I will publish over the coming weeks. This series will help you understand the various steps involved in developing custom predictive models using Maximo APM Predict.


The Maximo APM Predict offering has various out of the box models that you can quickly use to predict asset anomalies, failure dates, failure probabilities, and to help you generate probability of failure curves for your assets, as well as highlight the asset features that contribute most to failures. Also included is access to the IBM Smarter Resource and Operations Management machine learning library, which contains state-of-the-art algorithms developed by IBM Research for predictive maintenance.


However, if you’re curious about how to begin using machine learning for predictive maintenance, maybe even using your own custom models, this series of recipes will guide you through the steps, explaining the data science process, and showing you code snippets to help you on your way. We will follow the CRISP-DM process for data science, and each of the articles will focus on a step in that process.


1.     Business Understanding

2.     Data Understanding

3.     Data Preparation

4.     Modelling

5.     Evaluation

6.     Deployment





Business Understanding

The first step in this process is completed before you ever begin working with our software, but in many ways, it is the most important. Before beginning the development of any predictive maintenance solution, you must fully understand the business objectives. This step in the process should set the context of the problem you are trying to solve, assess the current situation, and fully understand all the requirements from the business. This step should set the expectations and any criteria for success. This will help align both the business and the team responsible for the development of the predictive maintenance solution. Typically, this will take the form of a project plan, detailing timelines, milestones, assumptions, constraints, unknowns, potential issues, and the criteria for success.



  1. Know the background

    Begin with researching the necessary background information about the current situation. This might include understanding the various business units that will be stakeholders in predictive maintenance, for example, the asset management, engineering, operations, and maintenance teams. Identifying the the key individuals in these teams is worthwhile. An internal sponsor that can provide financial backing and the necessary domain expertise is always useful.


    It is then necessary to understand the problem area and be able to describe it for a general audience. Set out the reasons for why you are under taking predictive maintenance. Understand whether your organization has used AI or predictive analytics in the past and were there any lessons learned. Preparing informational materials on predictive maintenance to share within the organization will help people better understand what you are trying to achieve.


    Take a look at what the current solutions are to the problem and be able to articulate the pros and cons of those solutions.

  2. Determine the business objectives

    After the research phase, you will need to work with the stakeholders in this project to clearly define the predictive maintenance objectives that you want to achieve, as specifically as possible. So, it might be something like: Reduce the failure rates of Acme Industry large centrifugal pumps by 10%


    In order to get to defined business objectives, you will also want to clearly define the problem you are seeking to solve, clearly address any business questions , and outlining any requirements from the business (for example, not increasing associated maintenance costs for the specified assets).

  3. Defining the success criteria

    Once the objectives are defined, you’ll need to think about the measures by which the success can be measured. For each objective, the success criteria should be documented as precisely as possible and have agreement from the necessary stakeholders, especially if the measurements are subjective and not necessarily quantifiable.

  4. Next steps

    The next step in the process is data understanding, where we will be getting our hands dirty with some code for data analysis. We’ll take a closer look at that in our next recipe so stay tuned!

Join The Discussion