Predictive models

Use predictive models to generate the information you need to make informed operational, maintenance, repair, or component replacement decisions.

This section describes the steps that are needed to build predictive models in the predictive maintenance area by using IBM Predictive Maintenance and Quality (PMQ). It also covers some sample use cases in the manufacturing field. Later, it highlights the steps involved, starting from the business/data understanding to deploying the predictive models built for a given use case.

The following models form the basis of the predictive models in IBM Predictive Maintenance and Quality:

  • The Maintenance predictive model
  • The Sensor Health predictive model
  • The Top Failure Reason predictive model
  • The Integrated Health predictive model

Sample predictive models are provided. For more information, see IBM SPSS artifacts.

The training and scoring process

The steps for training and scoring the predictive models are as follows:

  1. The modeling node estimates the model by studying records for which the outcome is known and creates a model nugget. This is referred to as training the model.
  2. The model nugget can be added to any stream with the expected fields to score records. By scoring the records for which you already know the outcome (such as existing customers), you can evaluate how well it performs.
  3. After you are satisfied that the model performs acceptably well, you can score new data (such as health score of an asset or life time of an asset) to predict how they will perform.

Optimized recommended actions

When an asset or a process is scored and identified as having a high probability of failure, recommendations can be generated.

Define recommended actions by using rules in IBM Analytical Decision Management. Use IBM Analytical Decision Management to understand the drivers that are used to define the rules, and to determine what happens based on the scores received. For example, if a score breaches a threshold, what is the resulting action? You can automate alerts for recommended actions by integrating with other systems or by defining a routing rule to send emails. Depending on the manufacturing execution systems (MES) that you use, the recommendation may be acted on automatically. You can also predict the success rate of the corrective action that is based on previous actions.

When IBM Predictive Maintenance and Quality generates recommendations, for example, to inspect an asset, you can configure the system so that the recommendation results in a work order that is created by IBM Maximo. The work order is populated with the information needed to complete the task, for example, a device identifier and a location.

Prioritize application template

Use the prioritize application template when you have a good understanding of the predictive analytics scores and the interaction between the predictive scores. You can use the template OptimizedAssetMaintenance.xml to prioritize your business objective that is based on, for example, profit maximization, or downtime minimization.