INITIAL SITUATION

Our customer collected a large amount of manufacturing and maintenance data and was struggling to make sense of it.

The vision was to develop a predictive maintenance platform to save costs by reducing expensive repairs.​

This can be achieved by detecting failing components before they break.​

SOLUTION

We have developed a prediction model based on a time-dependent training set.

The ML model is able to predict potential problems for a given machine based on many different characteristics (when it was manufactured, when it was sold, components, etc.)

For each prediction, a confidence interval for the prediction is also provided to increase the interpretability of the model.

BUSINESS VALUE

Our customer was able to better visualize and analyze maintenance issues.

Multiple customers could be contacted to replace specific components while still under warranty.

The operations teams were also able to identify patterns and significantly reduce high repair costs.

FRAMEWORK & TOOLS

The entire pipeline was developed using Big Data technologies (Spark, Hadoop, Python) and completed with a dashboard for visualization in Tableau.

Finally, the project was handed over to the operations team after the production rollout.