Engine Control Device Using AI Model and Learning Table
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning-based engine combustion models require significant arithmetic processing time and are nonlinear, making it difficult to derive mathematical models for inverse functions, which complicates real-time on-board control and increases calculation load.
Innovation Solution
An engine control device that uses a multi-layer neural network AI model to reproduce combustion state indexes like thermal efficiency and pressure rise rate, optimizing operation amounts through a learning control table that associates engine operating conditions with control target values, allowing for fast and efficient control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a deep learning-based combustion model (multi-layer neural network) is used to reproduce engine combustion indexes, then the accuracy of combustion state reproduction is improved, but the arithmetic processing time increases significantly
Solution Approach 1:
The patent pre-calculates and stores the relationship between operation amounts and combustion indexes in a lookup table during an offline training phase. During actual engine control, the system only performs table lookup and simple interpolation operations, avoiding real-time execution of the complex neural network forward propagation. This preliminary preparation transfers the heavy computational burden from online control to offline training, resolving the contradiction between accuracy and processing speed.
2Adaptability or versatility
If a deep learning-based combustion model is used, then the ability to capture nonlinear combustion characteristics is improved, but the difficulty of deriving inverse functions for control increases
Solution Approach 1:
The patent creates a forward model (combustion reproduction model) and uses it to generate training data for an inverse model (control amount determination model). Instead of directly deriving inverse functions from the complex neural network, the system copies the forward model's input-output relationships and trains a separate inverse model using supervised learning. This approach simplifies the control implementation while maintaining the ability to handle nonlinear characteristics.
3Extent of automation
If optimization control is performed using a multi-layer neural network to simultaneously control multiple operation amounts, then the comprehensiveness of control is improved, but the calculation load increases
Solution Approach 1:
The patent segments the control task into two distinct phases: offline training phase where the neural network performs comprehensive optimization control to learn the relationships between multiple operation amounts and combustion indexes, and online control phase where the trained model is used for fast inference. This segmentation allows the system to achieve comprehensive control capability during training while maintaining low calculation load during actual engine operation through efficient table lookup and interpolation.
Data Source
AI summary
An engine control device includes a model that, based on engine operating condition and first-type operation amount, reproduces at least one index from among various indexes of combustion state of engine, and a processor that executes a process including deciding on second-type operation amount, by optimization using the model so as to treat at least one of the indexes, which are reproduced by the model, as estimated value of control amount, and ensure that the estimated value of the control amount follows control target value, associating the second-type operation amount with the control target value and the engine operating condition, rewriting a learning control table in which operation amount corresponding to the control target value and the engine operating condition is registered, and calculating operation amount according to the learning control table based on the control target value and the engine operating condition.