Additive Manufactured Component Fatigue Prediction by Localized ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting the fatigue life of additive manufactured components are hindered by the complexity of interactions between multiple parameters, including roughness and porosities, which vary across the component, and lack of empirical rules specific to additive manufacturing, making it difficult to develop accurate mathematical models that account for localized material properties.
Innovation Solution
A machine learning approach using a Gaussian Progress Regression with Squared Exponential covariance function is employed to predict fatigue life, allowing for localized parameter consideration and zone-specific calculations without a priori assumptions, integrated with a zoning concept and durability solver to account for varying properties across complex components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional empirical rules from casting or forging are used to compensate for artefacts, then some guideline on how to compensate for certain artefacts is provided, but these rules are empirically derived for conventional manufacturing and may not be transferable to additive manufacturing, and they only consider artifacts such as surface roughness or porosities without considering print orientation
Solution Approach 1:
The patent transforms the approach by changing from using conventional empirical rules to implementing a machine learning model that specifically learns the relationships between printing parameters, material properties, and fatigue life. The system processes multiple parameters including print orientation, surface roughness, porosity, and laser parameters to predict fatigue life, thereby adapting the methodology to additive manufacturing-specific characteristics
Solution Approach 2:
The patent replaces conventional empirical rule-based mechanical approaches with a data-driven machine learning system. Instead of relying on established rules from casting or forging, the system uses trained algorithms to predict fatigue life based on actual additive manufacturing data, substituting mechanical reasoning with computational intelligence
2Reliability
If a mathematical model is developed to describe how multiple parameters interact, then fatigue life prediction becomes more comprehensive, but the large number of parameters and interactions makes it difficult to develop, define and calibrate the model
Solution Approach 1:
The machine learning model performs self-calibration by automatically learning the relationships between multiple parameters and fatigue life from training data. The system defines its own mathematical relationships through the training process, eliminating the need for manual model definition and calibration by researchers or engineers
Solution Approach 2:
The patent transforms the complex multi-parameter problem into a manageable solution by using machine learning to automatically process and integrate numerous parameters. The system handles printing parameters, material properties, artefact characteristics, and their interactions simultaneously through the training process, converting a complex calibration problem into a data-driven learning process
3Productivity
If printing parameters are introduced into a damage approach, then part-level fatigue can be predicted, but it is not possible to account for artefacts such as surface roughness or porosities, and variations throughout the part are not included
Solution Approach 1:
The patent segments the component into multiple zones with different material properties and artefact characteristics. The machine learning model predicts localized material properties for each zone, allowing the system to account for variations throughout the part including surface roughness, porosity, and print orientation-specific features in each segmented region
Solution Approach 2:
The patent implements local quality by assigning different material properties to different zones within the component. The machine learning model predicts zone-specific material properties based on local printing parameters and artefacts, enabling accurate fatigue life prediction that accounts for spatial variations in surface roughness, porosity, and other artefacts across the component
Data Source
AI summary
A method and a system for fatigue life prediction of additive manufactured components accounting for localized material properties. The method and the system is employed for prediction of fatigue life properties of an additive manufactured element, with a data collection step in which several data points for maximum stress vs. cycles to failure for different given processing steps of the element are collected, with a training step in which a Machine Learning system is trained with the collected data, and with an evaluation step in which the trained Machine Learning system is confronted with actual processing steps and used to predict the fatigue life properties of the element.