ANN EONOx Prediction System for Diesel Engine Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing EONOx sensors in diesel engines are unreliable due to high temperature and particulate matter in the exhaust environment, leading to unstable and untrustable data, especially during dynamic exhaust conditions.
Innovation Solution
An intelligent exhaust gas prediction system using an Artificial Neural Network (ANN) is deployed in the engine controller to predict EONOx levels instead of or in conjunction with traditional sensors, eliminating the need for complex software strategies and lengthy calibration efforts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If EONOx sensors are used to measure exhaust gas emissions, then EONOx information can be provided to control systems, but the sensors become unreliable due to high temperature and particulate matter in the exhaust environment
Solution Approach 1:
The patent introduces an intermediary system (neural network-based estimation system) that mediates between the unreliable sensor and the control system. The neural network processes multiple input parameters (engine load, temperature, pressure, etc.) to generate reliable EONOx estimates, effectively replacing the unreliable direct sensor measurement while maintaining the required measurement accuracy for control purposes.
Solution Approach 2:
Instead of relying on the physical sensor that fails in harsh environments, the system creates a virtual copy of the EONOx measurement through neural network estimation. This virtual sensor replicates the measurement function by learning from training data and processing readily available engine parameters, providing reliable EONOx information without exposing measurement components to the harsh exhaust environment.
2Speed
If EONOx sensors are used under dynamic exhaust conditions, then real-time EONOx data is available, but the data becomes unstable and untrustable during rapid changes
Solution Approach 1:
The neural network system implements feedback by continuously monitoring engine operating parameters and adjusting EONOx estimates in real-time. The system processes multiple input signals (engine load, temperature, pressure, airflow) and dynamically updates the estimation based on current conditions, providing stable and reliable data even during rapid transients where traditional sensors fail.
Solution Approach 2:
The system performs preliminary action by pre-training the neural network with extensive training data that covers various dynamic operating conditions. This pre-learning enables the system to quickly and accurately estimate EONOx during dynamic events without requiring physical sensors to respond in real-time, thereby providing stable data during rapid changes.
3Measurement precision
If traditional EONOx estimation methods are used, then the system can provide EONOx information, but complex software strategies and lengthy calibration efforts are required
Solution Approach 1:
The patent replaces complex mechanical/software-based estimation systems with a neural network-based intelligent system. The neural network, once trained, provides accurate EONOx estimates through straightforward computation of weighted inputs, eliminating the need for complex software strategies and lengthy calibration procedures while maintaining or improving estimation accuracy.
Data Source
AI summary
The Method and Arrangement for Predicting Engine Out Nitrogen Oxides (EONOx) includes training multiple candidate Artificial Neural Network (ANN) architectures using training data, and then selecting an ANN architecture from the candidates using an automated ANN architecture selection algorithm and testing data. An intelligent EONOx prediction or estimation system using the selected ANN architecture then provides an EONOx output variable, which is used along with the output of an EONOx sensor. The system is deployed into the engine controller. The training and testing sets of data include input variables from engine sensors and/or actuators that relate to EONOx, and may be acquired by testing a target engine. Selecting the optimal ANN architecture may be based on Root Mean Squared Error (RMSE) analysis using the automated ANN architecture algorithm and the training set of data.