Acoustic Defect Detection Model Training With Outlier Sample Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The inconsistency in human judgment and the difficulty in cultivating experts with 'golden ears' lead to high leak rates and increased costs in product quality inspection, particularly in sound component assembly, due to inconsistent marking standards and personnel fatigue, which affects the accuracy of defect detection.
Innovation Solution
A two-stage model training method is employed, involving a first classification model trained using contrastive learning to identify outliers, followed by a second classification model established on an optimized sample set, utilizing spectrogram conversion and data augmentation to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human inspection with golden ear is used, then detection accuracy is improved, but device complexity and cost increase due to difficulty in cultivating experts
Solution Approach 1:
The patent replaces the mechanical system of human ear inspection with an automated acoustic detection system that uses signal processing and machine learning algorithms. The system converts acoustic signals into spectrograms and uses classification models to automatically identify defective components, eliminating the need for human experts with golden ears while maintaining high detection accuracy.
Solution Approach 2:
The patent creates a digital copy of the expert inspection process by training machine learning models on labeled acoustic data. The classification models learn to replicate the decision-making process of human experts, enabling automated detection that mirrors expert judgment without requiring actual experts to perform the inspection.
2Ease of operation
If human marking is performed, then initial detection is achieved, but measurement precision deteriorates due to inconsistent standards and personnel fatigue
Solution Approach 1:
The system enables self-service by allowing the machine to automatically label and classify acoustic data without human intervention. The classification models independently process spectrograms and assign labels to components, eliminating the need for human operators to perform repetitive marking tasks that lead to inconsistency and fatigue.
Solution Approach 2:
The patent implements a feedback mechanism where the system continuously learns from labeled data to improve its classification accuracy. The model is trained on marked data, performs detection, and the results can be used to refine the model further, creating a self-improving system that maintains consistent and accurate marking without human fatigue.
3Productivity
If automated machine inspection is used, then productivity is improved, but measurement precision deteriorates due to inability to detect subtle acoustic anomalies
Solution Approach 1:
The patent transforms the acoustic signal inspection from the time domain to the frequency domain by converting audio signals into spectrograms. This dimensional transformation allows the system to analyze acoustic anomalies across multiple frequency dimensions simultaneously, enabling the automated system to detect subtle anomalies that would be invisible in traditional time-domain analysis while maintaining high inspection speed.
Solution Approach 2:
The system changes the parameters of analysis by using different spectrogram representations and adjusting model hyperparameters to optimize detection sensitivity. By transforming the input data into different feature spaces and adjusting classification thresholds, the automated system achieves both high productivity and precise detection of subtle acoustic anomalies.
4Measurement precision
If more training data is used, then model accuracy is improved, but loss of time increases due to extensive data processing
Solution Approach 1:
The patent extracts only the most relevant features from the acoustic data by using spectrogram transformation and feature selection techniques. Instead of processing all raw audio data, the system extracts key frequency and temporal features that are most indicative of defects, reducing processing time while maintaining model accuracy by focusing computational resources on the most informative data elements.
Data Source
AI summary
A method for establishing a defect detection model and an electronic apparatus are provided. A first classification model is established based on a training sample set including a plurality of training samples. The training samples are respectively input to the first classification model to obtain a classification result of each training sample. A plurality of outlier samples that are classified incorrectly are obtained from the training samples based on the classification result. A part of outlier samples that are classified incorrectly is deleted from the training samples, and the remaining training samples are used as an optimal sample set. A second classification model is established based on the optimal sample set so as to perform a defect detection through the second classification model.