Limestone, better known in the industry as calcium carbonate, is a chemical compound used in many industries; usually as burnt lime: in building materials, steel production, as fertilisers, in the pharmaceutical industry as well as in glass, paper and plastic production. A product that we are all faced with every day without being aware of it. More than 100 million tonnes are needed worldwide annually. Heavy industrial machinery is used to produce the precursors of limestone calcium carbonate: so-called classifiers, which sort the rock according to its size, and rock grinders, which grind the stone. A world-leading producer of industrial minerals planned to work with Trivadis to develop a solution that could detect damage to its industrial machinery in advance.
Predictive maintenance applications should avoid sudden failures, predict downtime and speed up repairs, all in all exploiting the benefits of an IoT predictive maintenance model in the industry. In the highly competitive global market and in light of the high demand for calcium carbonate, broken machines cost the company a lot of money, especially with deferred sales and delivery dates, as well as resources that are required unexpectedly for troubleshooting.
In accordance with the Trivadis motto “Better Together”, the solution was to be developed in close cooperation with the employees of the industrial minerals producer. A solution was developed together and Trivadis conveyed the basic principles of predictive maintenance to the employees in the project team. Typically, a vast amount of historical data from sensors from many of the same machines is available for predictive maintenance applications. These are analysed to find clues related to actual errors that have occurred. The challenge in this case was that the supplier of industrial minerals only uses a few of these large and heavy machines, which they also develop and build themselves. Therefore, there was only a very limited data pool available for the evaluation. The demands placed on Trivadis’ data scientists were therefore high.
Despite the limited basic data, Trivadis experts were able to read and interpret machine data based on database extracts with sensor information to turn data into valuable information. Measurement data was analysed based on the query language R and models were created and verified. The customer’s employees learned the statistical methods and how to use R and at the same time contributed the necessary specialist knowledge about the use and function of the machines. Within a short time, it was possible to develop a robust predictive maintenance model for proactive maintenance that could predict machine damage up to 45 days in advance.
For this model, suggestions for operationalisation were then developed, such as what a cloud infrastructure based on Microsoft Azure would have to look like in order to collect and historicalise sensor data and introduce predictive maintenance as a process. The Head of Engineering, Machines & Maintenance Group Operations summarises: “Trivadis has developed an IoT predictive maintenance model for us that allows us to identify any machine damage up to 45 days in advance. This enables us not only to better plan the maintenance of the machines, but also to significantly reduce costs.”