Wavelet-Based Invariant Feature Extraction for Universal Fault Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fault detection mechanisms in manufacturing are costly, time-consuming, and limited in detecting a wide range of defects due to reliance on complex techniques, intense manual efforts, and AI models that require large datasets and are specific to fixed product sizes and orientations.
Innovation Solution
A method and system using wavelet coefficients to generate invariant features, grouping them by product type, and quantizing these features to create representative coefficient signatures for universal fault detection across various product characteristics and orientations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI-based deep learning models are used for fault detection, then detection accuracy is improved, but computational resource consumption and training time increase significantly
Solution Approach 1:
The patent extracts only the essential invariant features from product images using wavelet transformation, rather than using entire images for training. This extraction of key features reduces the data dimensionality and computational burden while maintaining detection accuracy, directly addressing the contradiction between accuracy and computational resource consumption
Solution Approach 2:
The patent transforms images into the wavelet domain and extracts invariant features that are parameter-independent (invariant to scale, rotation, and translation). This parameter transformation allows the system to achieve high detection accuracy without requiring extensive training data for different product variations, thereby reducing both computational resources and training time
2Adaptability or versatility
If AI models are trained with extensive datasets covering all fault types, then detection coverage is improved, but training time and data requirements increase
Solution Approach 1:
The patent creates a universal fault detection mechanism using invariant features that work across different product types, sizes, and orientations. The wavelet-based invariant features capture essential fault characteristics that are independent of product-specific parameters, allowing a single trained model to generalize to multiple product variants without requiring separate training datasets for each, thus improving detection coverage while reducing training time
Solution Approach 2:
The patent performs wavelet transformation and invariant feature extraction as preliminary processing steps before fault detection. This preliminary action transforms the data into a form where fault characteristics are highlighted and made invariant to product variations, enabling the model to learn general fault patterns more efficiently and apply them across different product types without extensive retraining
3Measurement precision
If complex fault detection mechanisms are used, then detection accuracy is improved, but system complexity and manual effort increase
Solution Approach 1:
The patent replaces complex manual inspection mechanisms with an automated wavelet-based feature extraction system. The wavelet transformation and invariant feature extraction automatically identify fault characteristics without requiring manual annotation or complex preprocessing, thereby maintaining high detection accuracy while reducing system complexity and manual effort
Solution Approach 2:
The patent segments the image processing task into distinct stages: wavelet transformation, invariant feature extraction, and fault detection. This segmentation allows each stage to focus on a specific function, simplifying the overall system architecture while maintaining accuracy. The wavelet coefficients are computed once and reused for multiple feature extractions, reducing redundant computations and system complexity
4Adaptability or versatility
If fault detection is performed for all product variations, then detection universality is improved, but computational resources per product increase
Solution Approach 1:
The patent transforms product images into the wavelet domain and extracts invariant features that are independent of product parameters such as scale, rotation, and translation. This parameter transformation allows the same feature extraction and detection pipeline to be applied universally to all product variations without requiring parameter-specific adjustments or additional computational resources, achieving both universality and computational efficiency
Data Source
AI summary
Disclosed herein is method and fault detection system for detecting faults in one or more products. In an embodiment, method comprises generating plurality of wavelet coefficients corresponding to transformed images of each of the one or more products and determining a set of invariant features from the plurality of wavelet coefficients. Further, a dynamic set of invariant features is generated by grouping invariant features into a set of groups based on type of the one or more products. Subsequently, the dynamic set of invariant features is quantized based on a predetermined quantization threshold and a representative coefficient signature is associated for each group in the dynamic set of invariant features. Finally, faults in the one or more products are detected by comparing coefficient signatures associated with the one or more products with the representative coefficient signature of each group in the dynamic set of invariant features. In an embodiment, the present disclosure helps in accurate detection of faults in one or more products irrespective of type and characteristics of one or more products.