Multimodal Thermogram–XCT Registration for AM Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing additive manufacturing (AM) methods face challenges in accurately registering thermal imaging-based in-situ monitoring data with X-ray computed tomography (XCT) reference data due to irregular microsize and different data formats, leading to incomplete defect detection and quality evaluation of components.
Innovation Solution
A multimodal fusion-based registration method integrates thermogram and XCT data, preprocessing, image registration, and a machine learning model with a performance evaluation function to enhance accuracy and consistency, incorporating preliminary and non-rigid registrations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single sensor-based registration method is used, then the registration speed is fast and surface quality feedback is obtained quickly, but the defect detection completeness and quality evaluation accuracy are insufficient
Solution Approach 1:
The patent combines multiple sensor types (thermal imaging sensor and X-ray computed tomography sensor) into an integrated registration system. The thermal imaging data provides surface temperature distribution during processing, while XCT data provides internal structural information. By merging these different modalities, the system achieves comprehensive defect detection and accurate registration that neither sensor could achieve alone, resolving the contradiction between measurement precision and device complexity.
Solution Approach 2:
The patent creates a composite data structure that integrates thermogram data and XCT reference data with different formats and dimensions. This composite approach allows the system to leverage the complementary strengths of both data types - the real-time surface information from thermal imaging and the internal defect information from XCT - achieving high-precision registration and comprehensive quality evaluation.
2Adaptability or versatility
If different measurement methods are used to collect data, then comprehensive defect detection is achieved, but different data formats and dimensions make registration more challenging
Solution Approach 1:
The patent introduces a machine learning model as an intermediary that learns the complex mapping relationship between thermogram data and XCT reference data. This intermediary automatically handles the challenges of different data formats and dimensions by learning feature representations and transformation relationships, making the registration process robust to modality differences while maintaining comprehensive defect detection capability.
Solution Approach 2:
The patent transforms the registration problem from a direct spatial alignment task to a learned parameter optimization problem. By using machine learning to automatically learn transformation parameters and feature mappings, the system adapts to different data formats and dimensions without requiring manual calibration or complex preprocessing, thus maintaining versatility while reducing registration difficulty.
3Productivity
If single sensor registration is used, then timely collection of surface layer data is ensured, but continuous changes in pore status and molten pool status cannot be comprehensively detected
Solution Approach 1:
The patent implements continuous multi-modal data collection during the L-PBF process. The thermal imaging sensor continuously monitors surface temperature and molten pool dynamics, while the XCT sensor provides continuous internal structural information. This continuous multi-source data collection ensures both timely detection and comprehensive monitoring of pore status and molten pool status throughout the entire printing process, maintaining both productivity and reliability.
Solution Approach 2:
The patent establishes a feedback mechanism where the registered multi-modal data is used to continuously evaluate defect development and molten pool status. The machine learning model processes the integrated thermal and XCT data to provide real-time feedback on quality parameters, enabling dynamic adjustment and comprehensive monitoring that enhances reliability without sacrificing data collection speed.
Data Source
AI summary
A multimodal fusion-based precise registration method for additive manufacturing (AM), includes: simultaneously collecting thermogram data and X-ray computed tomography (XCT) reference data; preprocessing the collected data; performing image registration; establishing a machine learning model; completing training to obtain a trained machine learning model; registering a thermogram dataset and an XCT reference dataset by using the trained machine learning model, to obtain a pre-registration result; and evaluating the machine learning model using a performance evaluation function; and if the evaluation is successful, using the pre-registration result as a final registration result; or if the evaluation is unsuccessful, re-training the machine learning model. The present disclosure integrates different types of sensor data and evaluates the machine learning model using the performance evaluation function, so as to register the thermogram dataset and the XCT reference dataset by using a high-precision machine learning model.