استفاده از تئوری مجموعه راف برای تشخیص نقص شیشه های خودرو

A technique based on rough set theory is investigated for identifying defects on a backlight (a rear window of a vehicle with a defrost circuit). Since replacement of defective backlights result in a significant financial loss, automobile manufacturers are trying to remove defective backlights during the manufacturing process. Therefore, an automated inspection system based on infrared (IR) imaging techniques has been developed to detect backlight defects such as missing lines or hotspots, where the most challenging task is identifying hotspots from their artifacts. Feature selection techniques based on rough set theory are explored in this paper and are used to extract a feature vector, which increases inspection accuracy as well as reduces computational complexity. The theory is also applied to generate decision rules, which can be simply added to existing inspection systems to assist the operators in their decision making process. The proposed inspection system is expected to provide more reliable fault detection with low rate of false alarms than currently available systems.

In automotive industries, a defective backlight is considered as a significant source of financial loss as well as a source for assembly delays and warranty violations. Therefore, automobile manufacturers have been trying to screen the defective part in its manufacturing stage. As a result, an automated inspection system, which analyzes an infrared (IR) image of a heated backlight, has been developed and tested successfully [2]. Even though the automated system replaces manual inspection successfully, enhancing the inspection accuracy is still required to reduce the rate of false alarms. The accuracy problem is focused on isolating a hotspot (a spot which produces more heat than its neighbors) correctly from artifacts. Although image-processing techniques may be satisfactory for hotspot detection, they generally impose a computational burden. Therefore, an innovative approach to deal with both aspects simultaneously is pursued in this paper. Rough set theory, has been gaining attention due to its mathematical abilities to deal with uncertainty in data sets [4], [5] and [6]. Many applications have proven its usefulness especially in removing redundancies and extracting hidden relationships among data items [3] and [8]. As a result, the theory has been successfully applied for feature extraction and rule generation [7] and [9]. In this paper, rough set theory is applied to an automotive glass inspection system. Conventional image-processing techniques [1] are also used to reduce search space before feature extraction, and the rough set approach selects a feature vector from the initial feature set. Finally, the rough set method extracts a minimal set of decision rules, which will be implemented in the automotive glass inspection system to increase the accuracy and to decrease the rate of false alarms.

This paper presents an application of rough set theory to the detection of defects on an automotive glass. The theory provides sound mathematical tools to reduce redundancies in a data set as well as to identify hidden relationships among data items; redundant features are removed from the initial feature set through a reduction process, and a minimal set of decision rules is obtained. Image-processing techniques are also utilized before feature extraction to reduce the search space. As a result, a new backlight inspection system is expected to reduce the rate of false alarms without extra computational burden.