Screw Tightening Defect Detection Through Unsupervised Waveform Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing screw tightening technologies struggle to detect defects such as screw floating, foreign object biting, and screw oblique insertion, as they are based on specific threshold values and feature extraction methods, making it difficult to determine normal completion regardless of defect type.
Innovation Solution
A defect determination apparatus that utilizes unsupervised learning to generate a trained model from normal screw tightening data, allowing it to determine normal completion by analyzing torque and rotation speed waveforms, regardless of defect type.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If threshold determination using physical quantities such as torque and rotation speed is used, then defective screw threads and bottoming can be detected, but foreign object biting and screw floating cannot be detected
Solution Approach 1:
The patent applies universality by creating a single defect determination system that can detect multiple types of defects (defective screw threads, bottoming, foreign object biting, and screw floating) using a unified approach based on waveform feature extraction and machine learning, rather than requiring separate detection methods for each defect type
Solution Approach 2:
The patent changes the detection parameters from simple threshold-based physical quantity comparisons to complex waveform feature quantities extracted from torque and rotation speed waveforms. By analyzing multiple features including area, standard deviation, and other waveform characteristics, the system achieves comprehensive defect detection across all defect types
2Measurement precision
If dedicated threshold values are provided for each defect type, then specific defects can be detected, but defects without threshold values cannot be detected
Solution Approach 1:
The patent substitutes the mechanical threshold determination system with a machine learning-based classification system. Instead of manually setting and managing threshold values for each defect type, the system uses trained models that automatically classify defects based on extracted waveform features, eliminating the need for complex threshold management
Solution Approach 2:
The system implements self-service by using unsupervised learning to automatically identify defect patterns and determine classification criteria without requiring manual intervention for each defect type. The machine learning model autonomously learns from training data and applies the learned patterns to detect and classify various defects
3Measurement precision
If feature quantities are extracted from torque and rotation speed waveforms, then defective screw threads and foreign object biting can be detected, but the system becomes complex and cannot detect screw floating
Solution Approach 1:
The patent applies segmentation by dividing the defect detection process into distinct stages: waveform acquisition, feature extraction, and classification. By organizing the complex feature extraction process into structured segments with clear functions, the system manages complexity while maintaining high detection accuracy across multiple defect types
Data Source
AI summary
A defect determination apparatus according to an aspect of the present disclosure includes: a physical quantity acquisition circuitry which, in operation, acquires a target physical quantity, the target physical quantity being a physical quantity generated in a power source during a new screw tightening; a model acquisition circuitry which, in operation, acquires a first trained model obtained through unsupervised learning of a physical quantity generated in the power source during a past screw tightening that has been normally completed; and a determination circuitry which, in operation, determines whether the new screw tightening has been normally completed by applying the target physical quantity to the first trained model.