SUCCESS STORIES – INDUSTRIAL ANALYTICS & BIG DATA

THE PROBLEM

Organizations face new challenges every day due to the need to be competitive in their markets and, at the same time, do so within the framework of higher quality requirements for their products, in order to meet the expectations of their most demanding customers. Our client, an international steel manufacturing company, had various issues with the quality of its products: In some cases, these problems manifested as deviations from the mechanical properties values of the product compared to its specification. This fact was only visualized at the end of the production process, where samples of the material are taken and tested in the laboratory to obtain the values of the mechanical properties. If the results showed deviations from the product specification, the material had to be downgraded, with the consequent business impact that entails.

THE SOLUTION

The mechanical properties of the material are intrinsic variables of the material itself and cannot be directly measured but rather through laboratory tests. Given the correlation between the mechanical properties of the material and the process variables that produced it, and the impossibility of measuring them directly, we developed a predictive model based on machine learning algorithms and trained it with historical data. This involved a thorough selection of the determinant process variables that feed into the model and a scoring of models until arriving at the final one, according to its performance. Predicting mechanical properties halfway through the process allows for the early detection of deviations, enabling corrective action downstream in the remaining processes to correct the deviations generated in the process.

THE RESULTS

The predictive model responded appropriately, with prediction errors of less than 5% for 90% of the population considered.