Low Cost effective Real Time Detection of Fabric Defects with CCD Linear Image Sensor

Abstract

In this study, real-time detection of defects of fabrics produced on knitting machines was carried out by using CCD (Charge Coupled Device) linear image sensor. To simulate knitting machines, a cylindrical drum rotated with a direct current motor was designed, and work was carried out with faulty and faultless fabrics wound on this drum. A lowcost linear line camera sensor containing 1500 active pixels was chosen for this study. The driver circuit of the selected image sensor requires interdependent and very precise timer signals. For this reason, a development board containing a 32-bit microcontroller with an operating frequency of 100 MHz with a superior price/performance ratio was used to control the sensor. In addition, a light source with a wavelength between 500 nm and 600 nm, to which the sensor is maximum sensitive, was used. Linear dash camera sensor has analog output. For this reason, every image pixel detected by the sensor is converted into digital. Since the microcontroller used in the study supports direct memory access, the data converted from analog to digital are automatically saved in the memory. Thus, both the maximum speed supported by the development board in analog-digital conversion has been reached, and it has been possible to perform the control in real time by processing the previous line data in this hardware-driven processing time period. Since two-dimensional image analysis algorithms require very high processing power and it is not possible for microcontrollers to perform this process in real time, image analysis was performed for one dimension, ie row. The fabric defect was detected by monitoring the statistical parameters, the drum was stopped by the control circuit of the motor, and the machine operator was simulated to be informed of the error with the audible warning system at the same time.