Gear Reducer Parameter Optimization With Reinforcement Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The design process for gear reducers is time-consuming and prone to human error due to the manual evaluation of conflicting engineering requirements, often resulting in rushed and sub-optimal designs, especially when no existing product meets the application's needs.
Innovation Solution
An automated gearbox design method using reinforcement machine learning, which instantiates a gearbox model, analyzes its performance, and iteratively updates parameters based on rewards to satisfy performance targets, reducing human intervention and improving design efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual design process is used to evaluate different kinematic concepts, then design flexibility and human expertise can be applied, but the design process becomes very time-consuming and prone to human error
Solution Approach 1:
The patent replaces the manual mechanical design evaluation process with an automated computational system using machine learning agents. The system automatically evaluates different kinematic concepts, performs sizing and analysis of components, and optimizes designs without human intervention, thereby eliminating human error while significantly reducing design time through automated parallel processing.
Solution Approach 2:
The design system is self-optimizing through reinforcement learning agents that automatically evaluate designs, receive feedback on performance, and iteratively improve designs without human intervention. The system performs self-correction and self-optimization by automatically adjusting design parameters based on performance metrics and constraints.
2Productivity
If the design process is accelerated to meet short timelines, then productivity increases, but design quality deteriorates resulting in rushed and sub-optimal designs
Solution Approach 1:
The patent replaces manual design processes with automated machine learning-based design optimization. The system can evaluate thousands of design configurations in parallel, perform comprehensive analysis including finite element analysis and dynamic simulation, and optimize designs automatically, achieving both high speed and high quality without the trade-off inherent in manual processes.
Solution Approach 2:
The system performs preliminary automated evaluation of multiple design concepts simultaneously, conducting sizing, stress analysis, and performance prediction before final selection. This preliminary automated screening ensures that even under tight deadlines, comprehensive analysis is performed on all candidate designs, preventing rushed decisions.
3Reliability
If comprehensive component sizing and analysis is performed for each design candidate, then design reliability improves, but the time investment increases significantly
Solution Approach 1:
The system performs comprehensive analysis on a selected subset of promising design candidates identified through preliminary automated screening. The machine learning agents first perform rapid initial evaluation to identify top candidates, then apply full comprehensive analysis only to those candidates, achieving efficient resource allocation that maintains design robustness while improving overall throughput.
Solution Approach 2:
The design evaluation process is segmented into multiple stages: initial concept generation, preliminary screening with simplified analysis, detailed analysis of promising candidates, and final optimization. This segmentation allows comprehensive analysis to be applied selectively rather than uniformly to all candidates, improving productivity while maintaining reliability for critical design decisions.
Data Source
AI summary
A method for automated gearbox design includes: instantiating the gearbox model having an initial parameter state in a modeling environment; analyzing and/or characterizing the gearbox model in the modeling environment to determine gearbox model performance; and determining whether the gearbox model performance satisfies a performance target. Upon a determination that the gearbox model performance does not satisfy the performance target: a reward is calculated based on the gearbox model performance; a reinforcement machine learning agent determines a parameter change action based on the reward and a current parameter state of the gearbox model; and an updated parameter state of the gearbox model is determined based on the parameter change action.