Interpretable Predictive Maintenance for Turbofans
Using Optimal Survival Trees to predict a machine's probability of survival over time
Simplifying predictive maintenance
Predictive maintenance aims to find the right moment to perform maintenance so that an industrial system's components are not prematurely replaced while ensuring the reliability of the whole system. As complexity of industrial systems grows, understanding the ways in which they can fail becomes all the more challenging.
Optimal Survival Trees can predict a machine's probability of failure over time, and provides visual insights to easily understand the underlying failure mechanisms. Let us showcase its capabilities on the Turbofan dataset, released by NASA to study engine degradation.
Understanding a machine's behavior
Optimal Survival Trees is fed daily observations from 25 sensors for 100 recorded machine failure trajectories. For each machine, we need to compute the time remaining before the engine failure occurred, classically called remaining useful life.
The resulting tree splits machines into cohorts based on their operating conditions, displaying the distribution of time before failure as well as the expected survival time of each cohort. On the right of the tree, in leaf node 7, an extreme failure mode characterized by high pressure conditions is highlighted. Under such conditions, most engines will fail in the next 19 days!
Empowering workers to improve processes
In an operational setting, workers and engineers can analyze the output decision tree and adjust their maintenance activities based on the common failure modes that were discovered.
More than improving maintenance scheduling, the tree's insights can be used to modify and improve the machines themselves and thus eradicate certain failure modes in the future.
Unique Advantage
Why is the Interpretable AI solution unique?
-
Detecting interpretable paths to failure
Optimal Survival Trees can automatically display paths to failure, as well as healthy behaviors, featuring correlations between several machine metrics simultaneously
-
Understanding probability of survival
Each machine ends up in one of the tree's leaf nodes, and is assigned a curve representing its probability of survival over time
-
Adaptable to low data availability
If easily accessible data is scarce and comes from a short time frame, interesting findings can still be found using Interpretable AI’s software modules