How we introduced predictive maintenance in the mining industry

Abstract: For a client in the mining industry, we developed machine learning models that continuously predict the wear of a complex machine over the next days and weeks. The AI can learn which factors and machine conditions drive the highest wear, all while explaining how it came to certain predictions in a clear and concise way. Through this approach, we’ve not only managed to predict future wear (therefore enabling an optimized maintenance schedule), but we’ve also created optimal machine setting guidelines to reduce future wear.


Our customer is a large industrial engineering company whose services include building mining equipment. They decided to add data-driven, analytics-based software products to their traditional hardware-based business in order to generate new revenue channels. One of the key problems they set out to tackle was usage optimization, with the goal of reducing costly downtime.


At SPRYFOX, we began by testing the suitability of the customer’s existing historical data—specifically for use in predictive algorithms. Initial testing showed that while data cleaning and consolidation were necessary, there was also an opportunity to demonstrate business value through continuous wear prediction.

With this in mind, we implemented a full analytics pipeline composed of preprocessing steps that cleaned, aggregated, and consolidated the data before evaluating a set of suitable machine learning algorithms. These included both classical and neural network-based algorithms.

Common challenges included sourcing missing data, navigating complex heterogeneous data inputs, and gaining familiarity with several important features that could only be measured indirectly.


Our final solution was a full prediction engine trained over one year of data. The AI is now able to predict future machine wear over the next 3-4 weeks with a small margin of error. Additionally, it can identify which factors and machine conditions are driving the highest machine wear.

Explainable AI—or the trained model’s ability to explain how it came to certain predictions in a clear and concise way (understandable to humans)—was key to the success of the project.

As a result of this approach, we not only managed to predict future wear (and implement an optimized maintenance schedule), but also employ more optimal machine settings to reduce future wear.