Skewness-Based Pattern Vector Similarity Detection for Abnormal Sound
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting similarity between standard and input patterns, particularly in the context of abnormal sound detection in concrete structures, face challenges due to non-monotonic changes in kurtosis values, leading to inaccurate detection of shape differences and abnormal sounds.
Innovation Solution
The method involves calculating a skewness geometric distance by evaluating the magnitude of shape changes in a reference pattern as a variable of skewness, rather than kurtosis, and using optimized weighting vectors to create weighted pattern vectors, which are then normalized and combined to improve similarity detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If kurtosis is used as the variable to evaluate shape changes in pattern detection, then the detection method is simple, but the detection accuracy deteriorates due to non-monotonic changes in kurtosis values
Solution Approach 1:
The patent changes the evaluation parameter from kurtosis to skewness. Skewness provides monotonic changes that accurately reflect shape differences between patterns, resolving the non-monotonic behavior of kurtosis while maintaining computational simplicity. The skewness value monotonically increases or decreases with shape changes, enabling reliable detection accuracy.
2Measurement precision
If optimized weighting vectors are introduced to create weighted pattern vectors, then the similarity detection accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary optimization of weighting vectors using training data before actual pattern recognition. The optimal weighting vectors are pre-computed and stored, so during runtime only simple vector operations are needed. This preliminary action separates the complex optimization process from the runtime detection, improving accuracy without significantly increasing operational complexity.
Data Source
AI summary
The present invention provides a method for obtaining an accurate detected value of a similarity, such as an hitting sound. The method includes the steps of: creating original standard/input pattern vectors each having a feature quantity of an hitting sound; creating a skewness-weighting vector and a kurtosis-weighting vector based on a reference pattern vector of a reference shape; calculating a skewness-weighted standard pattern vector and a kurtosis-weighted standard pattern vector by product-sum operation using component values of the skewness-weighting vector and the kurtosis-weighting vector and a component value of the original standard pattern vector; creating a dual and weighted standard pattern vector based on these vectors and similarly creating a dual and weighted input pattern vector; creating dual and selected standard/input pattern vectors based on the dual and weighted standard/input pattern vectors; and setting an angle between the dual and selected standard and input pattern vectors as a geometric distance value between the original standard and input pattern vectors.