Additive Manufacturing Parameter Modeling for Quality Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current additive manufacturing (AM) processes, such as directed energy deposition (DED), face challenges in efficiently configuring and monitoring process parameters due to stochastic variations, leading to suboptimal product quality and resource inefficiency.
Innovation Solution
The implementation of a method that uses machine learning models and mathematical models to automate the selection and monitoring of AM process parameters, allowing for real-time adjustments and optimization of the manufacturing process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional sensor-based monitoring systems are implemented to ensure quality control in additive manufacturing, then product quality and process stability are improved, but system complexity and computational resource requirements increase significantly
Solution Approach 1:
The patent extracts only the essential features from complex sensor data by using trained predictors to identify critical parameters. Instead of processing all sensor data, the system extracts specific meaningful features that indicate process quality, thereby reducing computational complexity while maintaining monitoring effectiveness
Solution Approach 2:
The system performs preliminary training of machine learning predictors using historical process data before actual manufacturing. This pre-training phase establishes the relationship between process parameters and output quality, enabling the system to make accurate predictions during production without requiring complex real-time computations
2Manufacturing precision
If extensive sensor systems and computational models are deployed to determine optimal process parameters, then manufacturing precision is improved, but the time and resources required for process setup and optimization increase dramatically
Solution Approach 1:
The patent implements preliminary training of predictors using historical data before actual manufacturing processes. This pre-computed knowledge base allows the system to quickly determine optimal parameters for new manufacturing tasks without requiring extensive real-time experimentation, significantly reducing setup time
Solution Approach 2:
The system creates virtual models and digital twins of the manufacturing process through trained predictors. These digital copies allow for rapid simulation and optimization of process parameters without physical trial-and-error, reducing the time required for process setup and optimization
3Measurement precision
If large computational systems are used to process all sensor data in real-time, then monitoring accuracy is improved, but energy consumption and computational resource usage increase
Solution Approach 1:
The system extracts only the most relevant features from sensor data using trained predictors, processing only essential information rather than analyzing complete raw datasets. This selective feature extraction maintains monitoring accuracy while significantly reducing computational energy requirements
Solution Approach 2:
The patent implements partial processing by focusing computational resources on analyzing only critical process parameters and deviations from target values. Instead of processing all sensor data equally, the system applies computational analysis selectively to the most important measurements, reducing overall energy consumption
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for performing an automated additive manufacturing (AM) process comprising obtaining a set of first AM factors associated with performing a first AM process and obtaining a set of first characteristics of a first AM output, wherein the first AM output is generated by performing the first AM process using the set of first AM factors. A predictor is trained using the set of first AM factors and the set of first characteristics, which is a machine learning model, and a mathematical model is generated based on extracting internal parameters of the predictor. The mathematical model comprising represents an estimated relationship between first AM factors and first characteristics. A set of target characteristics is obtained and, using the mathematical model, a set of second AM factors is identified. The set of second AM factors are then used to perform a second AM process, generating a second AM output.