Abnormal Sound Detection Using Spectrogram Image Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting assembly defects in sound-related electronic products rely on human auditory evaluation, which is subjective and prone to errors, and fail to classify abnormal sound signals, leading to increased repair time and occupational hazards.
Innovation Solution
An abnormal sound detection method and apparatus using image recognition on spectrograms generated from sound signals, employing a neural network model with a bidirectional long short-term memory layer, max pooling, flatten, and fully connected layers to classify defect categories, reducing reliance on human evaluation and providing accurate defect identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human auditory evaluation is used to detect assembly defects, then detection capability is provided, but detection accuracy deteriorates due to subjective factors and repair time increases due to lack of classification
Solution Approach 1:
The patent replaces the mechanical human auditory evaluation system with an automated sound detection system comprising a sound detection model, spectrogram generation module, and image recognition model. This substitution eliminates subjective human factors while providing both defect detection and classification capabilities, thereby improving detection accuracy and reducing repair time through automated defect category identification.
Solution Approach 2:
The patent introduces spectrograms as an intermediary representation that bridges sound signals and defect classification. The sound detection model first converts audio signals into spectrograms, which are then processed by the image recognition model to identify defect categories. This intermediary step enables both detection and classification functions that were previously unavailable in human-based systems.
2Reliability
If human listeners are used for defect detection, then detection function is provided, but reliability deteriorates due to occupational injuries and subjective variability
Solution Approach 1:
The patent replaces human listeners with an automated sound detection system that processes audio signals through sound detection models and image recognition algorithms. This eliminates exposure to harmful sound levels while maintaining consistent, objective detection results free from human fatigue and subjective variability.
Solution Approach 2:
The system performs self-diagnosis by automatically detecting and classifying defects without human intervention. The sound detection model and image recognition model work together to identify defect categories, enabling the system to serve itself in terms of quality control and reducing reliance on human operators exposed to harmful conditions.
3Loss of information
If existing models are used to detect abnormal sounds, then detection capability is provided, but information completeness deteriorates due to lack of defect classification
Solution Approach 1:
The patent enhances the functionality of the detection system by making it perform multiple tasks: both detecting abnormal sounds and classifying defect categories. The image recognition model is trained to identify various defect types, enabling the system to provide comprehensive information about the nature of defects, which directly improves repair efficiency by eliminating the need for additional diagnostic steps.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method accurately classifies abnormal sound signals, reducing repair time, improving defect detection accuracy, and minimizing occupational injuries by automating the identification of defect categories, thereby enhancing the efficiency of machine repair processes.
Implementation Method 1
the step of converting the abnormal sound signal into the spectrogram includes executing fast Fourier transform on the abnormal sound signal for generating the spectrogram
Data Source
AI summary
An abnormal sound detection method and apparatus are provided. First, an abnormal sound signal is received. Next, the abnormal sound signal is converted into a spectrogram. Afterwards, image recognition is performed on the spectrogram for obtaining a defect category corresponding to the abnormal sound signal.