ML Surrogate Model for Design Feedback Loop
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In industrial manufacturing, finding an optimal design for complex technical systems like hybrid cars is challenging due to the vast number of possible component combinations, leading to inefficient design processes where engineers rely heavily on experience and simulation feedback, which is only available for fully specified designs.
Innovation Solution
An automated system using a machine learning model that computes probability distributions and predicted impact values for design key performance indicators, allowing engineers to evaluate component additions during the design process and guide critical decision-making.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If engineers rely on experience and simulation feedback to find optimal designs, then design quality can be improved, but the design process becomes extremely time-consuming due to the vast design space
Solution Approach 1:
The machine learning model is trained in advance on historical design data and simulation results to learn the relationship between component configurations and KPI performance. This preliminary training enables the model to provide rapid predictions during the actual design process without requiring time-consuming simulations for every evaluation
Solution Approach 2:
A machine learning model is introduced as an intermediary between the design space exploration and simulation evaluation. The model approximates simulation results for partial designs, providing fast feedback that guides engineers toward promising design regions before committing to full simulations
2Manufacturing precision
If the design space is fully explored to find optimal component combinations, then design optimality is improved, but the complexity of the design process increases dramatically
Solution Approach 1:
The design process is segmented into iterative steps where engineers gradually build partial designs by adding components. The machine learning model evaluates each partial design configuration, providing feedback that guides the next component selection without requiring evaluation of all possible complete designs
Solution Approach 2:
Instead of evaluating complete designs only, the system performs partial design evaluation using the machine learning model on incomplete configurations. This partial evaluation provides early guidance and reduces the need to explore the entire design space exhaustively
3Measurement precision
If simulation feedback is used to evaluate design performance, then measurement accuracy of KPIs is improved, but the feedback loop becomes too long for efficient iterative design
Solution Approach 1:
The machine learning model creates a computational copy or surrogate of the simulation process. Instead of running actual simulations for every design evaluation, the trained model provides approximate predictions that mimic simulation results but execute orders of magnitude faster, enabling rapid iterative design
Solution Approach 2:
The system implements a dual feedback mechanism: the machine learning model provides immediate approximate feedback for rapid iteration, while periodic full simulation feedback refines the model's predictions and validates design decisions, creating a layered feedback system that balances speed and accuracy
Data Source
AI summary
A machine learning model processes a current partial design of a technical system and a candidate component for a next design step of designing the technical system. The model computes a probability distribution, which is a probability distribution over changes of a design KPI if the candidate component is added to the current partial design, with the design KPI describing a property of the technical system, and a predicted impact value predicting an absolute value of the design KPI or a change of the design KPI if the candidate component is added to the current partial design. These predictions (for partial designs that cannot be processed by a simulation environment due to their incompleteness) can drastically shorten the feedback loop between engineers in charge of designing a new technical system/product and a simulation environment used for estimating the performance characteristics of the product.