Phase 2 of the Detecta project addresses the development of an Event Prediction Prototype, an innovative tool that has made it possible to anticipate the average utilization of the milling machine, overcoming the unique challenges presented by highly variable production due to different part characteristics.
The unique nature of the production process, where every part worked on is different, has presented a significant challenge due to the absence of stable patterns for predictive analysis. However, advanced time series and deep learning techniques, such as the N-HiTS model, optimized with Optuna, have been applied to effectively address this variability. The potential of this approach lies in anticipating the need for maintenance and possible failures, while at the same time improving workload management and optimizing production planning.
The choice of the size of the prediction window and the meticulous selection of variables have been essential to fine-tune the AI model. The Detecta F2 team has been able to identify the variables with the highest correlation with the average machine utilization, which has improved the accuracy of the predictions. Through hyperparameter optimization with Optuna, the model has been tuned to minimize the mean absolute error, reflecting the high accuracy of the predictions in a highly variable environment.
Test results on different signal spans have demonstrated the model’s ability to track the overall trend of machine utilization with a remarkable degree of accuracy. Despite the inherent variability in production, consistent trend tracking and adequate handling of peaks and valleys has been observed, although there are still opportunities to improve the prediction of sudden fluctuations.
The Detecta F2 project is funded by the Ministry of Industry and Tourism through the line of support grants for Innovative Business Clusters, in its 2023 call under the Recovery, Transformation and Resilience Plan.
Results published in February 2024.
