Methods For Detection Of Ss Plate Surface Flaws And Character Recognition Using Thermal Images

Abstract

To ensure the quality of industrial manufacturing processing, newlineparticularly for stainless steel plates, defect assessment is an essential newlinestage. However, this labor-intensive and unpredictable technique is newlinetypically carried out manually in industry. It is essential to enable a newlinemachine to examine automatically surface imperfections from newlinestainless steel plates using computer vision technologies to replace the newlinemanual effort. It is critical to have an effective method of constantly newlineinspecting the materials flowing through the manufacturing line in large newlinevolume of metal sheet processes. The inspection mostly comprises of newlinedefect detection and serial number tracking. This thesis investigates the newlinefeasibility of developing an automatic inspection system on stainless newlinesheet by assessing several machine learning methods for defect newlineidentification and automated serial number detection (OCR). newlineInitially, infrared thermal imaging for surface roughness newlinemeasurement of metal is discussed. The thermal image of metal is newlinepreprocessed to remove noise followed by thresholding and detecting newlineporosity for metal grooves. Furthermore, Dyadic Wavelet Transform newline(DyWT) is applied for discontinuous metal groove edge detection. The newlineobtained DyWT parameters are applied to the regression model to newlineestimate the metal surface roughness. The accuracy, sensitivity, and newlinespecificity metrics are used to evaluate the output of the predicted routine. newlineSecondly anAFRCNN is applied for automated detection of surface newlinedefects in SS plate.This methodology uses different algorithms like RPN newlineand Faster R-CNN to identify the fault that occurs during manufacturing. newlineHere, the damaged plates are identified using Region Proposal Network newlinevi newline(RPN) and Fully Convolutional Neural Network (FCNN) functioning as a newlinecombined process under Faster R-CNN. Then, the number corresponding newlineto the particular plate is recognized using the standard ALPR approach newlinewith the support of character recognition technique. Character recognition newlineis applied after segmentation to identify the alphabets