Control Parameter Optimization Using Latent System Behavior Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for parameterizing control functions in technical systems, such as brake systems and steering assists, often require manual expertise and explicit quality functions, limiting their scalability and efficiency, especially when using Bayesian optimization or simulation-based inference.
Innovation Solution
An automated method using pre-trained variational autoencoders and neural posterior estimation models to iteratively optimize control parameters by simulating system behaviors and learning probability distributions, allowing for unsupervised training and reduced simulation needs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual parameterization by experienced specialists is used, then the control function can be properly configured, but the process is time-consuming and not scalable
Solution Approach 1:
The patent replaces manual mechanical parameterization processes with an automated neural network-based system. The neural network model learns from simulation data and automatically determines optimal parameter values, substituting the manual expert process with an automated computational system that maintains accuracy while dramatically reducing time requirements
Solution Approach 2:
The system enables self-service parameterization where the neural network automatically configures control parameters without requiring human expert intervention. The model serves itself by learning from simulation data and independently determining optimal parameter values, making the process autonomous and scalable
2Manufacturing precision
If Bayesian optimization with explicit quality functions is used, then parameter optimization can be performed, but the method requires explicit quality function specification and consistent data dimensionality
Solution Approach 1:
The patent substitutes the complex Bayesian optimization framework with a neural network-based approach. Instead of requiring explicit quality functions and Bayesian inference mechanisms, the system uses neural networks to directly learn the mapping from operational profiles to optimal parameters, simplifying the method while maintaining optimization precision
Solution Approach 2:
The system changes the fundamental parameters of the optimization approach by transitioning from Bayesian optimization with explicit quality functions to neural network-based learning. This parameter change allows the system to handle variable data dimensionalities and eliminates the need for explicit quality function specification while maintaining optimization effectiveness
3Extent of automation
If simulation-based inference with large amounts of simulated data is used, then parameterization can be performed without unsupervised pre-training, but the required simulation time becomes unviable
Solution Approach 1:
The patent applies partial action by using a manageable amount of simulation data that is sufficient for neural network training without requiring the excessive amounts of data needed for simulation-based inference methods. The neural network efficiently learns from this partial dataset, achieving automation without the prohibitive simulation time requirements
Solution Approach 2:
The system replaces simulation-based inference with neural network-based learning, substituting a data-intensive approach with a more efficient learning mechanism. The neural network achieves comparable or superior parameterization automation with significantly reduced simulation time requirements
Data Source
AI summary
A computer-implemented method for determining parameter values of parameters of an optimized parameter set for operating a particular technical system is disclosed. The behavior of the parameterized technical system can be simulated by way of operational variable profiles indicating time profiles of at least one input variable, at least one output variable and at least one status variable. The method includes the steps of providing a data-based representation model trained to associate operational variable time profiles of one or more technical systems with a latent representation vector in each case that characterizes the behavior of the technical system, and providing a data-based distribution model trained to associate latent representation vectors resulting from simulated operational variable profiles of the particular technical system with a probability distribution of parameter values of the parameters of the parameter sets. And the following steps are carried out iteratively (i) providing parameter values of an initial parameter set or selecting parameter values of a parameter set from a probability distribution of parameters by way of random selection, (ii) simulating or measuring the technical system parameterized with the parameter values of the parameter set in order to obtain operational variable time profiles, (iii) analyzing the obtained operational variable profile with the data-based representation model to obtain a latent representation vector, and (iv) further developing or retraining the data-based distribution model with a training data set from the obtained latent representation vector and the parameter set so that an updated probability distribution of parameters results from the data-based distribution model.