| Summary: | 碩士 === 元智大學 === 工業工程與管理學系 === 106 === The purpose of this research is to develop an effective and efficient machine vision method to inspect a machined surface. The target machined surface to inspect has a circular shape. The tool-mark patterns on the machined surface follow the same circular shape. Many machine vision methods offer solutions for tool-mark inspection, but these methods are designed for linear tool-mark patterns. The methods used to solve for linear tool-mark patterns are not effective on circular shaped tool-mark patterns. In this research, two morphology-based methods are presented to solve the circular tool-mark surface defect inspection problem.
The morphology methods are utilized to remove the background and highlight defective regions. Morphology especially works in separating regular background patterns from irregular objects. Traditional morphology uses rectangular structuring elements for morphological operations. The first method, named Rectangular morphology, converts the circular tool-mark pattern to a horizontal tool-mark pattern by transforming the original image from it Cartesian coordinates to its equivalent Polar coordinate. Following the Polar conversion, traditional morphology operations and noise removal techniques are used to separate the defect areas from the background. The second method, named Circular morphology, conserves the image in its original coordinates with the circular tool-mark pattern. It uses arc-shaped structural elements for the morphology operations to remove the background and highlight the defects in the original image.
The experimental results show that the two presented morphological approaches can achieve high defect detection rates while reducing the false alarms. The Rectangular morphology method has to deal with more noise removal steps due to noise and interpolated artifacts created by the Polar conversion. This causes some false alarms present in the final result. The Circular morphology method does not require Polar image conversion. This translates into less noise and none false alarms in the final result.
|
|---|
