Novel inverse predictive system integrated with industrial lubricant information

The development of lubricants with the required specifications is a time-consuming and costly process, given the need to explore numerous formulations that represent material types and their combinations for the lubricant. Although machine learning approaches have made significant progress in addressing these challenges, their practical application in industrial plants requires considerable expert knowledge to screen out unfeasible lubricant formulations. Hence, this paper proposes a novel inverse predictive system integrated with industrial information to recommend new and feasible formulations for the lubricant industry. The proposed system integrates a machine-learning-based model for property prediction with an optimization process that incorporates expert knowledge. The developed prediction models demonstrated high performance in predicting the lubricant properties (viscosity at 40 degrees C, viscosity at 100 degrees C, and density), with R 2 scores of 0.9839, 0.9779, and 0.9816, respectively. In addition, an optimization process using particle swarm optimization was employed to suggest formulations tailored to the specific requirements of various industries. The recommended formulations were validated using laboratory-scale specimens with errors of 5%-20%. This framework provides promising opportunities for recommending material types and their ratios in various industries.