Machine Control Model Selection for Accurate Fuel-Mix Commands
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine control systems face challenges in creating practical engine control maps for combined fuel types, as manually creating control maps for each combination is complex and costly, and existing machine learning techniques may not achieve sufficient accuracy with a small number of learning models.
Innovation Solution
A machine control system that selects and combines multiple learning models based on acquired state quantities to calculate a composed command value, allowing for the output of a specific command value that satisfies a specific condition, even with a small number of learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple learning models are used to improve command value accuracy for specific conditions, then the accuracy improves, but the device complexity and computational load increase
Solution Approach 1:
The patent segments the learning model selection into discrete categories based on operating conditions (e.g., fuel type, load condition). Each segment has a dedicated learning model, allowing the system to choose the most appropriate model without evaluating all possibilities, thus maintaining accuracy while reducing complexity.
Solution Approach 2:
The system changes parameters such as fuel type, engine load, and operating temperature to select different learning models. By organizing models according to parameter ranges, the system achieves high accuracy for specific conditions without requiring a single complex model to handle all scenarios.
2Device complexity
If a small number of learning models are used to reduce system complexity, then the device complexity decreases, but the ability to obtain accurate command values for specific conditions deteriorates
Solution Approach 1:
Each learning model is designed to be universal within its designated operating range, handling multiple parameters (fuel type, load, temperature) simultaneously. This allows a small number of models to cover diverse conditions effectively, maintaining accuracy while limiting system complexity.
Solution Approach 2:
The patent introduces an intermediary selection mechanism that matches current operating conditions to the most appropriate learning model. This mediator layer enables accurate command value generation for specific conditions without requiring the system to maintain all possible specialized models.
3Measurement precision
If manual control maps are created for each fuel combination to ensure accuracy, then the command value accuracy improves, but the manufacturing complexity and cost increase significantly
Solution Approach 1:
The learning models automatically adapt and optimize control parameters based on real-time sensor data from the engine, eliminating the need for manual creation and calibration of control maps for each fuel combination. The system self-adjusts to maintain accuracy across different fuels and conditions.
Solution Approach 2:
The system continuously monitors engine performance and uses feedback from sensors to refine learning model predictions. This closed-loop approach ensures high accuracy for each fuel combination without requiring pre-manufactured control maps, as the system learns optimal parameters during operation.
Data Source
AI summary
According to one embodiment, a machine control system includes: a selecting unit which acquires a state quantity of a machine converted from data acquired by a sensor provided in the machine to select two or more learning models according to the acquired state quantity; a composing unit which inputs the state quantity acquired by the selecting unit to each of the two or more learning models selected by the selecting unit to calculate a composed value using a command value output from each of the learning models; and a learning unit which outputs a command value with respect to the machine in a range based on the composed value calculated by the composing unit, acquires a state quantity of the machine, searches for a command value of a specific condition from a combination of the output command value and the acquired state quantity, and outputs the searched command value to the machine, thereby creating a new learning model.