Virtual Sensor Data Generation via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional vehicle-in-the-loop systems require accurate physical sensor models for each vehicle sensor, making it costly and time-consuming to simulate multiple vehicle sensors effectively.
Innovation Solution
A method for generating sensor data that reduces the effort required to create sensor models by using virtual sensors and machine learning algorithms to simulate sensor data in real-time, allowing for more efficient and cost-effective vehicle simulation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If accurate physical sensor models are used for each vehicle sensor, then measurement precision is improved, but device complexity and manufacturing effort increase
Solution Approach 1:
The patent creates virtual copies of physical sensors (virtual first vehicle sensor, virtual second vehicle sensor) that replicate sensor functions in the simulation environment. These virtual sensors generate sensor data through algorithms rather than physical measurement, eliminating the need for complex physical sensor models while maintaining measurement precision in the simulation.
Solution Approach 2:
The patent replaces physical sensor models with algorithm-based sensor data generation. Instead of using complex physical models to simulate sensor behavior, the system uses trained algorithms (neural networks, machine learning models) to generate sensor data directly from environmental data, substituting mechanical/physical modeling with computational approaches.
2Measurement precision
If multiple accurate physical sensor models are implemented, then measurement precision is improved, but productivity and development time decrease
Solution Approach 1:
The patent performs preliminary training of sensor data algorithms using training datasets before actual simulation. The algorithms are pre-trained to map environmental features to sensor data outputs, so during runtime, sensor data can be generated quickly without needing to evaluate complex physical sensor models, thus improving productivity while maintaining precision.
Solution Approach 2:
By creating virtual sensors that copy the functional behavior of physical sensors through algorithms, the system avoids the time-consuming process of developing and validating physical sensor models for each sensor type, accelerating simulation development while maintaining measurement accuracy.
3Measurement precision
If physical sensor models are used for real-time simulation, then measurement precision is improved, but computational effort increases
Solution Approach 1:
The patent substitutes computationally intensive physical sensor model evaluations with pre-trained algorithms that generate sensor data more efficiently. The algorithms use learned patterns from training data to directly produce sensor outputs, reducing computational effort while maintaining measurement precision in real-time simulation.
Solution Approach 2:
By pre-training the sensor data algorithms offline using comprehensive datasets, the system transfers computational effort from runtime simulation to offline training. During real-time simulation, the trained algorithms generate sensor data with minimal computational overhead, reducing energy consumption while preserving accuracy.
Data Source
AI summary
A method for generating sensor data. The method includes providing first sensor measurement data of at least a first vehicle sensor, calculating virtual object data of a virtual vehicle surroundings model that includes a virtual second vehicle sensor, wherein a sensor acquisition range of the first vehicle sensor and a sensor acquisition range of the second vehicle sensor overlap spatially and/or temporally in the vehicle surroundings model, calculating at least second modeled sensor data of the virtual second vehicle sensor using a trained first sensor data algorithm depending on the virtual object data, wherein the first sensor data algorithm is based on a training process using training data that include selective first sensor measurement data of the first vehicle sensor, outputting a sensor data set that at least includes the first sensor measurement data and the second modeled sensor data.