Engine Control Using Reference State Interpolation at Operating Limits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for controlling internal combustion engines face accuracy issues due to insufficient training data near the engine's performance limit, leading to unstable combustion and potential deterioration in control accuracy.
Innovation Solution
A control device that acquires state amounts from sensors and uses a neural network to compute control amounts by determining whether the state amounts fall within a set region, using in-region data for direct computation and out-of-region data to select reference state amounts for interpolation, thereby ensuring accurate control even at extreme operating conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a fuzzy neural network uses multiple state amounts as input for learning, then the control accuracy should improve, but the training data becomes insufficient near the engine's performance limit
Solution Approach 1:
The patent divides the state amount space into multiple regions based on data density. The region determination unit segments the input space into high-density regions (where sufficient training data exists) and low-density regions (where training data is insufficient). This segmentation allows the system to apply different processing strategies to different regions, resolving the contradiction by identifying where the data insufficiency problem occurs.
Solution Approach 2:
The patent introduces an intermediary mechanism (the region determination unit and reference state amount selection) that mediates between the fuzzy neural network and the insufficient training data. When out-of-region state amounts are detected, the system selects reference state amounts from in-region data and uses interpolation to generate estimated control amounts, thereby bridging the gap caused by insufficient training data.
2Measurement precision
If training data is collected near the engine's performance limit, then the learning accuracy should improve, but the combustion becomes unstable and control amounts cannot be accurately measured
Solution Approach 1:
The patent performs preliminary action by pre-determining regions of high and low data density before the actual control process. The region determination unit establishes the spatial distribution of training data in advance, identifying which state amount combinations have sufficient training data and which do not. This preliminary analysis allows the system to prepare appropriate handling strategies for each region.
Solution Approach 2:
The patent uses copying by selecting reference state amounts from stable in-region conditions and copying their control characteristics to estimate control amounts for out-of-region state amounts. Instead of relying on unstable direct measurements from performance limit conditions, the system copies proven control patterns from stable operating regions and adapts them to extreme conditions through interpolation.
3Productivity
If the number of training data pieces is reduced, then the learning process becomes faster, but the control accuracy deteriorates
Solution Approach 1:
The patent applies parameter changes by dynamically changing the effective number of training data pieces based on the input state amount. For in-region state amounts, the system uses the full training dataset, maximizing learning accuracy. For out-of-region state amounts, the system effectively uses fewer training data pieces by relying on reference state amounts and interpolation, thereby maintaining acceptable accuracy while reducing computational burden.
Data Source
AI summary
A control device for an internal combustion engine includes an information acquirer, a first computing unit, and a second computing unit. The information acquirer acquires information on a state amount that changes depending on the operation state of the internal combustion engine. A region determiner determines whether the state amount falls within a set region. The first computing unit uses an in-region state amount within the set region, as an input value to compute a control amount of the internal combustion engine by a neural network. The second computing unit selects a reference state amount within the set region, based on the out-of-region state amount, uses the selected reference state amount as an input value to compute a reference control amount by the neural network, and computes the control amount corresponding to the out-of-region state amount based on the computed reference control amount.