Automated Deform Detection For Automotive Body Panels Using Image Processing Techniques

• Main Author: Edris, Muhammad Zuhair Bolqiah
• Format: Thesis
• Language: English; English
• Published: 2016
• Subjects: Q Science (General); QA Mathematics
• Online Access: http://eprints.utem.edu.my/id/eprint/18376/; https://plh.utem.edu.my/cgi-bin/koha/opac-detail.pl?biblionumber=100216

The demand for automotive industry has been rapidly increasing as the number of consumer increases. In order to ensure the quality of their product, the manufacturers need to minimalize any deformation that occurs to their products. Early deformation detection on the automotive body panels manufactured must be conducted in order to rectify the problem. Automated deformation detection was designed to replace manual labour and this technique is found to be more accurate and effective. This thesis proposed a method to detect the deformation that occurs on the automotive body panel surface while in assembly lines. Three-dimensional data is acquired from the body panels as an input for deformation detection system. The data is converted in two-dimensional data image by using scatter data interpolation. Gradient filtering is used to identify the gradient energy value yield from the surface by using two types of kernels. Background illumination correction is implemented in order to reduce unwanted regions in the image. The prepared images undergo segmentation stage by recognizing the deformation in each threshold value by using Artificial Neural Network. The threshold value has been assigned with range between 0.0001 until 0.2000 where the threshold value is increased by 0.0001 in iteration.The Gabors’ Wavelet is used to extract the features of the segmented candidates and as the input for the artificial neural network. A fuzzy logic decision rule is used to classify the types of deformations that have been obtained from the artificial neural network outputs.The depth of the deformation is then computed by subtracting the maximum and minimum values of the segmented candidates. Several test units were purposely built with deforms in order to test the proposed method. The mean accuracy of the NN recognition with Gabors’ Features Extraction was recorded at 99.50 %. The segmentation on flat surface was recorded with lowest accuracy percentage of 68.81 %, then followed by the car door and the curved surface with accuracy percentage recorded at 70.39 % and 79.03 % respectively. The detection accuracy percentage was found to be 100 % where all the deformed location was able to be detected.