Virtual Sensor Design Using Lipschitz Functions for Emission Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating emissions from heat engines, such as those used in the automotive field, face challenges in accuracy due to the complexity of combustion systems and production spread, leading to suboptimal performance and potential overfitting or local minima issues in parametric-statistical estimation methods.
Innovation Solution
A non-parametric estimation process using a Lipschitz function is employed, which selects a function that minimizes estimation errors by considering production spread and experimental data, resulting in a 'robustly optimal' virtual sensor that estimates emissions accurately across varying conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If parametric-statistical estimation methods are used to design virtual sensors, then the estimation can be performed using available data, but the method may converge to local minima or overfit, reducing estimation accuracy
Solution Approach 1:
The patent creates multiple copies of the training data by generating synthetic data through perturbation of original measurements. This augmented dataset is then used to train the neural network, allowing the model to learn from diverse variations and avoid overfitting to a single dataset, thereby improving robustness while maintaining estimation accuracy
Solution Approach 2:
The patent performs preliminary data augmentation and synthetic data generation before the actual training process. By pre-processing the data to include various perturbed versions and synthetic samples, the method prepares a more robust training foundation that prevents convergence to local minima during the optimization phase
2Measurement precision
If physical sensors are installed for direct measurement of emissions, then accurate measurement information is obtained, but the cost of sensor and installation, weight, and periodic maintenance increase
Solution Approach 1:
The patent introduces a virtual sensor based on neural network as an intermediary between the available engine operating parameters and the desired emission measurements. This virtual sensor processes readily available data (fuel injection quantity, intake air amount, EGR rate, etc.) to estimate emissions, eliminating the need for direct physical sensor installation while maintaining measurement capability
Solution Approach 2:
The patent creates a virtual copy of the physical sensor's measurement function through software-based neural network estimation. Instead of installing physical sensors to directly measure emissions, the system creates a computational model that replicates the measurement function using available operating parameters, thereby avoiding the complexity of physical sensor installation and maintenance
3Measurement precision
If more data is collected to improve estimation accuracy, then the model can better handle production spread, but the risk of overfitting increases
Solution Approach 1:
The patent generates synthetic copies of the training data by applying perturbations and transformations to the original dataset. This data augmentation technique effectively increases the dataset size and diversity without simply collecting more raw data, allowing the model to learn robust patterns while the regularization techniques prevent overfitting to the augmented data
Solution Approach 2:
The patent applies parameter perturbations to the training data, systematically varying parameters within realistic ranges to simulate production spread. This approach allows the model to learn invariance to parameter variations while maintaining accuracy, and the use of regularization ensures that the model generalizes well rather than overfitting to the perturbed parameters
Data Source
Figure 1~3
Figure 4
Figure 5~7
AI summary
Described herein is a process for designing a virtual sensor that is able to estimate a variable of interest v as a function of a set of available variables u i . The process comprises the steps of: - acquiring (1002) a design data-set D d comprising a number N of measured values v(ti) of the variable of interest v and corresponding measured values ũi(ti) of the available variables u i ; - determining a limit δ on the disturbances of the available variables u i and a limit η on the errors of the method of measurement of the variable of interest v; - selecting (1004) a Lipschitz function f* with a respective Lipschitz constant γ, which is able to estimate the variable of interest v(t) as a function of a number n of past values of each available variable u i , by executing the following steps one or more times for different numbers n: a) determining a value for the Lipschitz constant y; b) defining (1006) a maximum limit f(r(t)) and a minimum limit f (r(t)) for the estimate of the variable of interest v as a function of the design data-set D d , and moreover the number n, the value for the Lipschitz constant y, the limit δ on the disturbances of the available variables u i , and the limit η on the errors of the method of measurement of the variable of interest v, and choosing a Lipschitz function f* comprised between the maximum limit f(r(t)) and the minimum limit f (r(t)); c) determining (1008) an estimation error ε*(f*) for the Lipschitz function f* and selecting the Lipschitz function f*, associated to which is a respective Lipschitz constant y* and a respective number n*, that presents the minimum estimation error ε*( f*(y*, n*)); and - implementing (1012) the selected Lipschitz function f* in an electronic circuit.