Hybrid Process Models Using Acausal Modules and Auto-Differentiation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current industrial process control systems face inefficiencies due to the lack of modularity and interpretability in causally defined black box models, particularly when integrated with other optimization systems, and struggle with the computational demands of complex data sets from digitalized operations.
Innovation Solution
A system comprising acausal modular parameterized models that include both physical and neural network sub-models, utilizing reverse-mode automatic differentiation to generate gradients and update model parameters efficiently, allowing for improved parameterized control and optimization of industrial processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If causally defined black box models (neural networks) are used to model industrial processes, then the ability to handle complex data and reduce manual modelling hours is improved, but modularity, interpretability, and reuse possibility deteriorate
Solution Approach 1:
The patent segments the monolithic black box neural network into modular components that can be independently trained and reused. Each module represents a specific function or subsystem, allowing independent development, validation, and deployment while maintaining overall system performance.
Solution Approach 2:
The patent introduces intermediate representation layers and interface standards that enable different neural network modules to communicate and integrate. These intermediaries facilitate modularity by providing standardized connection points while preserving the black box functionality of individual components.
2Ease of operation
If standalone gradient descent approach is used for training universal function approximators, then training simplicity is improved, but efficiency deteriorates when integrated with other optimization systems
Solution Approach 1:
The patent merges the gradient descent training approach with other optimization systems by implementing a unified optimization framework. This allows multiple optimization algorithms to work together synergistically, improving overall efficiency while maintaining the simplicity of individual training processes through modular integration.
3Adaptability or versatility
If acausal declarative modelling languages are used to increase knowledge reuse, then modelling scalability is improved, but computational complexity increases
Solution Approach 1:
The patent segments the acausal declarative model into hierarchical levels of abstraction, where high-level knowledge representations can be reused across different applications while low-level computational details are handled separately. This reduces the apparent complexity while maintaining reusability.
Solution Approach 2:
The patent implements efficient copying and instantiation mechanisms for reusable knowledge modules. Once a model component is defined, it can be efficiently instantiated multiple times with different parameters, reducing computational overhead while maximizing knowledge reuse.
4Measurement precision
If digitalization of operations is pursued to create complex models, then data availability and model accuracy are improved, but computational resource requirements increase
Solution Approach 1:
The patent implements partial training and incremental learning approaches where models are trained on subsets of data or updated incrementally rather than retraining from scratch. This reduces computational resource consumption while maintaining model accuracy through selective processing of digitalized operational data.
Data Source
AI summary
There is provided a method and corresponding systems and computer-programs for evaluating and/or adapting one or more technical models related to an industrial and/or technical process. The method comprises obtaining a fully or partially acausal modular parameterized model of an industrial and/or technical process comprising at least one physical sub-model and at least one neural network sub-model, including one or more parameters of the parameterized model. The method further comprises generating a system of differential equations based on the parameterized model, and simulating the dynamics of one or more states of the industrial and/or technical process over time based on the system of differential equations. The method also comprises applying reverse-mode automatic differentiation with respect to the system of differential equations when simulating the industrial and/or technical process in order to generate an estimate representing an evaluation of the model of the industrial and/or technical process.