What are some failure prediction models in predictive maintenance?

As we all probably know, there are a lot of ways for equipment to fail. With predictive maintenance (PdM), the name of the game is less about knowing every single way of failing - rather, it's understanding an asset's most probable failure modes and monitoring those conditions. To do this, we can use failure prediction models to understand equipment failures and fix them before they arise.

Time-based models
One type of model is a time-based model called a regression model. This kind of model allows a PdM system to predict the number of days (or other metrics, like cycles, months, or products made) that are left before the system fails.

This model collects some amount of historical data and uses that data to look over every machine failure, building a model based off of the timing of failures in regard to current maintenance practice. It will then make a prediction for a date or specified time interval for failure.

Irregular behavior models
Perhaps a more common model, some failure prediction models look at so-called "anomalous" or non-normal behavior in an asset and use that behavior to predict failures.

For example, let's say a facility has a robot gripper arm that usually moves to Point A, picks up a part, and moves the part to Point B. However, when the machine is close to failure, the machine arm starts to take longer when it moves to Point B, freezing in place with the part in its arm.

We can point to this irregular behavior and understand it as a failure marker, using it as a way of diagnosing how close the asset is to failure.

Survival model
Survival failure prediction models ask the question: "How does the failure risk of an asset change over a period of time if we look at X amount of characteristics?"

If a facility is tracking a bunch of different parameters about an asset (like heat, vibrations, and sound), a survival failure prediction model will use all of these things to estimate how the potential for failure changes. So while this model may not tell you exactly when an asset will fail, it will tell you how failure risk goes up or down based on those characteristics. From there, you can monitor those specified parameters and assess for equipment failure.