Model-Based Defect Detection on Metal Strips
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting periodic defects on metal strips, such as autocorrelation methods, face challenges with process-related fluctuations, are not effective for small defects, and struggle to detect multiple periodic errors reliably.
Innovation Solution
A model-based estimation method, specifically a maximum likelihood method, is used to convert time signals into quality functions, enhancing detection sensitivity and stability by accounting for uncertainties like slip and signal-to-noise fluctuations, and allowing for the identification of periodic signals even in noisy conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Difficulty of detecting and measuring
If autocorrelation method is used to detect periodic defects, then detection capability for periodic signals is improved, but detection reliability deteriorates due to process-related fluctuations
Solution Approach 1:
The patent changes the fundamental parameter of the detection method from autocorrelation to model-based estimation (maximum likelihood method). This parameter change transforms the detection approach from one sensitive to process fluctuations to one that explicitly accounts for uncertainties like slip and signal-to-noise variations, thereby improving reliability while maintaining periodic signal detection capability
Solution Approach 2:
The patent substitutes the autocorrelation method (a statistical signal processing approach) with a model-based estimation method (maximum likelihood method). This substitution replaces a method that passively correlates signals with one that actively models the expected signal characteristics and estimates parameters, making the detection more robust against process-related fluctuations
2Difficulty of detecting and measuring
If autocorrelation method is used, then periodic signals can be recognized, but detection precision deteriorates for small defects
Solution Approach 1:
The patent changes the detection parameter from autocorrelation coefficient to maximum likelihood estimate. This parameter change enables the detection system to achieve higher precision for small defects because the maximum likelihood method optimizes the estimation based on the actual probability distribution of the signal, making it more sensitive to small periodic variations that autocorrelation might miss
Solution Approach 2:
The patent applies preliminary action by using a model-based approach that incorporates prior knowledge about the expected signal characteristics (periodicity, amplitude relationships) before performing the actual detection. This preliminary modeling allows the system to be pre-tuned for detecting small defects, improving precision before the measurement is even taken
3Difficulty of detecting and measuring
If autocorrelation method is used, then periodic defects can be detected, but detection stability deteriorates when detecting multiple periodic errors
Solution Approach 1:
The patent applies segmentation by separating the detection of multiple periodic defects into independent parameter estimation problems. The maximum likelihood method allows each defect's parameters (frequency, amplitude, phase) to be estimated separately based on the overall signal model, preventing the detection stability from deteriorating when multiple defects are present simultaneously
Solution Approach 2:
The patent changes from a method that processes multiple defects as a mixed signal (autocorrelation) to one that estimates individual defect parameters simultaneously (maximum likelihood). This parameter change in the detection approach maintains stability by treating each defect as a distinct parameter to be optimized, rather than as interfering periodic components
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
The invention shows an apparatus (1) and a method for detecting at least one periodically occurring defect (8) on an object (3), in particular on a metal strip, in which at least one time signal (7) is recorded from the moving object (3) using a measuring method, the time signal (7) is converted into a quality function (9) using a stochastic method and one or more periodic signals (7') in the time signal (7) are inferred on the basis of this quality function (9) in order to thereby detect at least one periodically occurring defect (8) on the object (3). In order to obtain advantageous method properties, it is proposed to convert the time signal (7) recorded by means of the measuring method, in particular the electromagnetic measuring method, into the quality function (9) using a model-based estimation method, in particular using a maximum likelihood method.