Power Beam Process Control Using Discriminative Quality Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing power beam processes, such as electron beam welding, require significant time and cost for machine-specific validation and calibration due to limited understanding of key parameters affecting weld quality, leading to inefficiencies in repeatability and consistency.
Innovation Solution
A method that determines key discriminative features through descriptive analytics to classify power beam processes into quality indicators, allowing for predictive modeling and control without reliance on periodic tests, using a control system and computer-readable storage medium to facilitate accurate and efficient process control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine-specific validation and calibration are performed through process control test pieces, then the power beam process quality is ensured, but the time and operational cost increase significantly
Solution Approach 1:
The patent replaces the mechanical/physical system of process control test pieces with a data-driven predictive model. Instead of physically testing components to validate parameters, the system uses machine learning models trained on historical process data to predict optimal parameters and assess quality risks, thereby eliminating the time-consuming physical validation process while maintaining quality assurance
Solution Approach 2:
The patent performs preliminary validation through predictive modeling before actual production. By training the model on historical data and using it to predict outcomes for new parameter sets, the system validates parameters computationally in advance, avoiding the need for time-consuming physical test pieces during production setup and changes
2Reliability
If machine-specific validation and calibration are performed through process control test pieces, then the power beam process quality is ensured, but the operational cost increases significantly
Solution Approach 1:
The patent replaces the costly physical validation process using process control test pieces with a computational predictive model. The model uses historical process data to predict quality outcomes, eliminating the need for expensive physical testing materials and reducing operational costs while maintaining quality assurance
Solution Approach 2:
The patent creates a virtual copy of the validation process through predictive modeling. Instead of physically validating parameters on actual test pieces, the system uses a computational model that replicates the validation function, thereby eliminating the need for expensive physical test materials while maintaining validation effectiveness
3Stability of the object's composition
If comprehensive parameter verification is performed using process control test pieces, then the power beam process consistency is improved, but the productivity decreases
Solution Approach 1:
The patent replaces the time-consuming physical verification process with rapid computational prediction. The predictive model can assess parameter consistency instantly using historical data patterns, eliminating the need for slow physical testing while maintaining the ability to detect inconsistencies and ensure process stability
Solution Approach 2:
The patent enables continuous parameter verification through the predictive model, which can instantly assess new parameter sets against learned patterns from historical data. This continuous computational validation replaces discrete, time-consuming physical tests, allowing the production process to maintain momentum without interruption for verification
4Manufacturing precision
If each machine performing the power beam process is validated individually, then the process accuracy is ensured, but the device complexity increases
Solution Approach 1:
The patent creates a universal predictive model that can be applied across multiple machines performing the same power beam process. The model learns from aggregated historical data and can predict outcomes for different machines, reducing the need for separate validation systems for each machine while maintaining process accuracy through the generalizable insights from the model
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
The present application relates to a method (200) for controlling a power beam process. The method (200) includes : carrying out a plurality of test power beam processes (202) using a power beam on one or more test components and determining a plurality of power distributions (204) corresponding to the plurality of test power beam processes (202). The method (200) further includes determining a plurality of beam parameters (206), generating derived features (208) based on the plurality of beam parameters (206), and determining a plurality of process characteristics (210) of each test power beam process (202). The method (200) further includes generating a comprehensive dataset (214), dividing the comprehensive dataset (216) into a test dataset and a training dataset, and determining a plurality of key discriminative features (222) from the plurality of derived features of a set of training power distributions of the training dataset. The present application relates also to corresponding computing device and non-transitory computer-readable storage medium.