Test Stand Active Learning for Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current test stand operations for motor vehicle systems, such as exhaust aftertreatment systems, face inefficiencies in training system models due to the lack of targeted input data selection, leading to suboptimal uncertainty reduction and predictive accuracy.
Innovation Solution
An active learning method is employed, where an optimization problem is defined based on input variable measurements, and a gradient is determined to select informative input data points for training, iteratively updating the system model with pairs of input and output data, focusing on reducing uncertainty and improving model quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If statistical test planning with predefined input variables is used, then measurements can be carried out systematically, but the training efficiency and uncertainty reduction of the system model is suboptimal
Solution Approach 1:
The test plan is made dynamic by iteratively adapting the selection of input data points based on the current state of the system model. Instead of using fixed predefined input variables, the method dynamically determines which input points to measure next based on where the model has highest uncertainty, making the training process adaptive and efficient
Solution Approach 2:
The method implements feedback by using the trained system model to inform subsequent measurement selections. The model's predictive accuracy and uncertainty estimates feed back into the test planning process, guiding which input points should be measured next to maximize training efficiency and reduce uncertainty in the most impactful areas
2Adaptability or versatility
If comprehensive input data is collected for training, then model coverage is improved, but the complexity and cost of test stand operations increases
Solution Approach 1:
The method extracts only the most informative input data points for training based on where the model has highest uncertainty. Instead of collecting comprehensive data across all possible input variables, it selectively extracts and measures only those specific input points that will provide the greatest benefit for improving model coverage and reducing uncertainty
Solution Approach 2:
The approach changes the parameter of data selection from comprehensive coverage to targeted selection based on uncertainty metrics. By changing how input data is selected (from fixed predefined sets to dynamically chosen points based on model state), the method achieves good model coverage with fewer, more strategically selected measurements
Data Source
AI summary
A device and a method for operating a test stand. A set of measurements of input variables of a system model of a component of a machine is provided. An optimization problem is defined using a set of measurements of input variables. A gradient for solving the optimization problem is determined as a function of the set of measurements. A solution to the optimization problem, which defines a design for input data for the test stand for a measurement on the component, is determined as a function of the gradient. A measurement of output data is acquired on the component on the test stand as a function of the input data. Pairs of training input data and training output data are determined as a function of the input data and the measurement of output data. The system model for the component is trained as a function of the pairs.