Machine Learning for Solvent Route Lubricant Base Oil Mixture Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The production of group I lubricant base oils through the solvent route is inefficient due to the lengthy process of determining optimal mixtures of distillate cuts, which requires extensive testing and often results in suboptimal quality oils that fail to meet market and regulatory standards.
Innovation Solution
A computer-implemented method using machine learning tools, specifically Random Forests and Gradient Boosting, to model and predict the properties of raffinate and dewaxed products, allowing for the optimization of operating conditions and the estimation of mixture percentages for achieving desired lubricant base oil specifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If extensive testing work is used to determine optimal mixtures of distillate cuts, then the quality of lubricant base oil is improved, but the time and cost required for formulation increases significantly
Solution Approach 1:
The patent creates a virtual model (copy) of the formulation process using machine learning algorithms trained on historical data from the pilot plant. This digital twin allows prediction of optimal mixture compositions without physically testing every combination, thereby maintaining quality while dramatically reducing time and cost.
Solution Approach 2:
The system performs preliminary analysis by training machine learning models on historical formulation data before actual production. This pre-computed knowledge base enables rapid prediction of optimal mixtures for new formulations, eliminating the need for extensive real-time testing.
2Ease of operation
If mass and solvent balances are used to determine mixing limits, then the formulation process is simplified, but the accuracy and reliability of the mixture optimization decreases
Solution Approach 1:
The patent introduces machine learning models as an intermediary between the simple mass balance calculations and the complex reality of mixture behavior. These models are trained on extensive experimental data and act as a bridge, providing accurate predictions while maintaining computational efficiency.
Solution Approach 2:
The system transforms the formulation approach from using only basic mass balance parameters to incorporating multiple parameters including pilot plant experimental data, chemical composition analysis, and process conditions. This multi-parameter approach significantly improves accuracy while the automated ML framework maintains operational simplicity.
3Reliability
If traditional pilot plant testing is used to evaluate mixture compositions, then reliable data is obtained, but the productivity and speed of formulation development is reduced
Solution Approach 1:
The system implements a feedback mechanism where pilot plant test results are continuously fed back into the machine learning models to improve their predictive accuracy. This creates a self-learning system that becomes progressively more reliable while maintaining high speed, as the models learn from actual experimental outcomes without requiring repeated physical testing.
Solution Approach 2:
The patent creates a virtual replica of the pilot plant testing process through machine learning models. This digital copy can simulate formulation outcomes instantly without the time and resource constraints of physical pilot plant operations, while the models are periodically validated and refined against actual pilot plant data to maintain reliability.
Data Source
AI summary
The present invention relates to data-driven methods for simulating the behavior of the solvent route. Specifically, a method is used in a simulator that estimates the behavior of the process. The methods comprise a prediction and an optimization method. The prediction method infers the properties of both the raffinate and the dewaxed from the properties of the feedstock and the manipulated variables. The optimization method defines which values of the manipulated variables generate the raffinate and the dewaxed with the desired properties.