Welding Audio Defect Detection with Self-Supervised Multi-Resolution AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI-based methods for detecting welding defects in metal-to-metal welding processes are limited by the need for large amounts of labeled audio data, which is difficult and time-consuming to annotate, leading to inefficiencies and potential missed defects in the manufacturing process.
Innovation Solution
A method that trains a foundation model using unlabeled audio data in a self-supervised manner with reconstruction loss, followed by fine-tuning with a limited amount of labeled data, enabling the detection of welding defects through multi-resolution audio analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If large amounts of labeled audio data are used for training AI-based defect detection, then detection accuracy is improved, but data annotation time and cost increase significantly
Solution Approach 1:
The patent applies preliminary action by using self-supervised pre-training on unlabeled audio data before fine-tuning with labeled data. The foundation model learns general welding audio patterns and defect characteristics from abundant unlabeled data during pre-training, which prepares the model to achieve high detection accuracy with significantly less labeled data required in the subsequent fine-tuning stage, thereby reducing overall annotation time and cost
Solution Approach 2:
The patent implements self-service through self-supervised learning where the model learns from unlabeled welding audio data without requiring manual annotations. The model automatically identifies patterns and features in the raw audio data through reconstruction tasks and contrastive learning, eliminating the need for time-consuming human annotation while still acquiring useful defect detection capabilities
2Measurement precision
If traditional single-resolution audio analysis is used, then processing speed is maintained, but detection of subtle defects is insufficient
Solution Approach 1:
The patent applies segmentation by dividing the audio analysis into multiple resolution levels: coarse-grained analysis for overall welding process monitoring and fine-grained analysis for detailed defect detection. The model processes audio features at different temporal and frequency resolutions, allowing it to capture both broad patterns and subtle defect characteristics that would be missed in single-resolution analysis
Solution Approach 2:
The patent implements another dimension by introducing multi-resolution analysis across temporal and frequency dimensions. Instead of analyzing audio at a single resolution, the model examines features at multiple scales - from broad temporal patterns to fine frequency details - adding dimensional depth to the analysis that enhances defect detection precision without requiring fundamentally new analysis methodologies
Data Source
AI summary
According to at least one embodiment, a method, a computer system, and a computer program product for multi-resolution audio defect detection in welding is provided. The present invention may include receiving unlabeled and labeled audio data; formatting the received unlabeled and labeled audio data; training a foundation model using the formatted unlabeled audio data in a self-supervised manner with a reconstruction loss; training the foundation model using the formatted labeled audio data with a classification loss; and performing multi-resolution audio defect detection on welding audio data using the trained foundation model.