Neural Network Mass Conservation Layer for Chemical Reaction Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data-based approaches for simulating chemical reactions in reactive systems, such as combustion engines and fuel cells, fail to accurately predict chemical species compositions due to neglecting mass conservation, leading to imprecise predictions and limited optimization.
Innovation Solution
A method involving an artificial neural network with an additional mass conservation layer is used to create a simulation model, where the neural network is trained with input and output data representing mass fractions of chemical species at the inlet and outlet of the reaction, and a linear correction is applied to ensure mass conservation of chemical elements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data-based approaches using machine learning algorithms are used to simulate chemical reactions, then prediction speed is improved, but mass conservation is not accounted for leading to imprecision
Solution Approach 1:
The neural network model is segmented into distinct functional layers: an intermediate model layer for prediction and a mass conservation layer for correction. This segmentation allows the model to maintain fast prediction capabilities while separately ensuring mass conservation constraints are met, resolving the contradiction between speed and accuracy.
Solution Approach 2:
A mass conservation layer is introduced as an intermediary component between the neural network prediction and the final output. This intermediary layer adjusts the predicted concentrations to satisfy mass conservation constraints without requiring complete redesign of the underlying neural network architecture, thus maintaining prediction speed while improving accuracy.
2Measurement precision
If an additional step of checking mass conservation is carried out outside the model, then mass conservation is taken into account, but optimization of the prediction is limited
Solution Approach 1:
The mass conservation checking step is merged with the neural network model by integrating it as an additional layer within the network architecture. This combination allows the model to internally enforce mass conservation constraints during the prediction process itself, rather than requiring separate post-processing steps, thereby maintaining both accuracy and optimization capability.
Solution Approach 2:
The mass conservation constraints are built into the model structure in advance, during the model design phase. This preliminary incorporation ensures that mass conservation is automatically satisfied for all predictions without requiring additional computational steps or external verification, simplifying the overall process while maintaining precision.
Data Source
AI summary
The present invention is a method of building a simulation model (MSI) of at least one chemical reaction, which builds an intermediate model using an artificial neural network (RNA) and a training base (BAP), and then builds the simulation model by adding an additional mass conservation layer (RNC) to artificial neural network (RNA). Furthermore, the invention relates to a method of simulating (SIM) a chemical reaction implementing the simulation model (MSI).