Vehicle Transmission State Estimation With Physics-Constrained AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting the state of a vehicle transmission, such as DE 10 2006 007717 A1 and US 5 557 521 A, face challenges in accuracy and reliability, particularly in determining physically viable gear states, especially when dealing with unobservable behavior and varying driving conditions.
Innovation Solution
A method combining a generative model with a physical model, where the generative model processes route information and vehicle speed to produce an intermediate state, and the physical model applies limits to ensure the output is physically plausible, enhancing accuracy and reliability by incorporating physical constraints and vehicle parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a probabilistic model is used to model unobservable behavior, then the model can handle unobservable states, but it becomes difficult to prove that the predicted states are physically viable
Solution Approach 1:
The patent combines a generative model (for predicting gear states) with a physical model (for validating physical viability). The hybrid architecture merges the strengths of both approaches: the generative model handles unobservable behavior and provides predictions, while the physical model constrains these predictions to ensure they satisfy physical laws and vehicle dynamics, thereby resolving the contradiction between adaptability and reliability
2Measurement precision
If a purely data-based model is used, then the model can learn from training data, but it cannot reliably extrapolate to unseen scenarios such as different numbers of gears
Solution Approach 1:
The physical model acts as an intermediary that bridges the gap between training data and unseen scenarios. It provides domain knowledge and physical constraints that guide the generative model's predictions, enabling reliable extrapolation to scenarios like different gear configurations without requiring extensive training data for each specific case
3Productivity
If only generative models are used, then the system can predict gear states, but the predictions may not satisfy physical constraints and vehicle dynamics
Solution Approach 1:
The physical model provides feedback to the generative model by validating predictions against physical constraints. If predicted gear states violate physical laws or vehicle dynamics, the physical model identifies these violations and the system can adjust predictions accordingly, ensuring that final outputs are both productive and physically reliable
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A device, a machine learning system (400) and a method for determining a state (116) of a transmission for a vehicle, the method comprising providing an input for a first in particular generative model (102) depending on a route information (106a), a vehicle speed (106b), a probabilistic variable (108), in particular noise, and an output of a second in particular physical model (104), determining an output of the first model (102) in response to the input for the first model (102), wherein the output of the first model (102) characterizes the state (116), wherein the first model (102) comprises a first layer (102a) that is trained to map input for the first model (102) determined depending on the route information (106a), vehicle speed (106b) and the probabilistic variable (108) to an intermediate state (110), wherein the first model (102) comprises a second layer (102b) that is trained to map the intermediate state (110) to the state (116) depending on the output of the second model (104), providing an input for the second physical model (104) depending on at least one vehicle state (114) and/or the route information (106a), determining an output of the second model (104) in response to the input for the second model (104), wherein the output of the second model (104) characterizes at least one limit (112) for the intermediate state (110).