Machine Learning Control for Real-Time Additive Manufacturing Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Additive manufacturing processes face challenges in rapidly optimizing and adjusting process control parameters in response to changes, leading to suboptimal quality and yield due to the lack of real-time adaptive control capabilities.
Innovation Solution
Implementing a method that uses machine learning algorithms to predict and adjust process control parameters in real-time based on input design geometry, training data sets including simulation, characterization, and inspection data, allowing for iterative improvements in process control during free form deposition or joining processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional additive manufacturing processes are used without real-time adaptive control, then the process is simpler to operate, but the manufacturing precision and quality are suboptimal due to inability to rapidly optimize process control parameters
Solution Approach 1:
The machine learning model is trained in advance using simulation data, process characterization data, and inspection data to learn the relationships between process control parameters and part quality. This preliminary training enables the system to rapidly predict optimal parameters during actual manufacturing without requiring complex real-time adjustments, thus improving manufacturing precision while keeping the real-time control system relatively simple.
Solution Approach 2:
The system implements a feedback loop where in-process inspection data and process characterization data are continuously fed back to the machine learning model. The model uses this feedback to dynamically adjust process control parameters in real-time, enabling adaptive optimization that improves part quality while maintaining manageable system complexity through automated closed-loop control.
2Manufacturing precision
If real-time adaptive control using machine learning is implemented, then manufacturing precision and part quality improve, but the device complexity and computational requirements increase
Solution Approach 1:
The machine learning model performs computationally intensive training and optimization work in advance, before actual manufacturing begins. By pre-learning the complex relationships between process parameters and quality outcomes from simulation and historical data, the system transfers computational complexity from real-time operation to offline preparation, thus improving real-time manufacturing precision without proportionally increasing real-time device complexity.
Solution Approach 2:
The system uses simulation data to create virtual copies of the manufacturing process for training the machine learning model. This allows the model to learn from extensive simulated scenarios without requiring equivalent computational resources during actual manufacturing, effectively using a simplified digital twin to guide the physical process with improved precision while managing computational complexity.
3Measurement precision
If iterative training with process characterization data is performed, then the machine learning model accuracy improves, but the loss of time for data collection and processing increases
Solution Approach 1:
The system implements continuous collection of process characterization data and in-process inspection data throughout manufacturing operations. Rather than performing discrete, time-consuming batch training sessions, the system continuously feeds new data into the machine learning model, enabling iterative improvement of prediction accuracy without significant interruptions to production, thus balancing model accuracy improvement with minimal time loss.
Solution Approach 2:
The machine learning model is initially trained in advance using available simulation and historical data before manufacturing begins. This preliminary training provides a functional model that can immediately guide production, allowing iterative refinement with actual process data to occur in parallel with manufacturing operations rather than sequentially, thereby reducing the overall time loss associated with data collection and model improvement.
Data Source
Figure 1
Figure 2
Figure 3A~3C
AI summary
Disclosed herein are machine learning-based methods and systems for automated object defect classification and adaptive, real-time control of additive manufacturing and/or welding processes.