Accurate Three-Dimensional (3D) Measurement of Highly Specular Surfaces for Quality Control Program of Large-scaled Production Line

Dawda, Arpita Rajkumar
Nguyen, Minh
Huang, Loulin
Item type
Degree name
Doctor of Philosophy
Journal Title
Journal ISSN
Volume Title
Auckland University of Technology

In the production line, inspection and quality control are essential to maintain the quality of the products. It assures confidence in the manufacturer and provides satisfaction to the customer. Product inspection is an indispensable tool of the modern manufacturing process. It helps in maintaining the quality of the product and reduces manufacturing costs by eliminating scrap losses. Several non-destructive examinations (NDE), or non-destructive testing (NDT), are used to analyze materials for inherent flaws such as fractures, dents or cracks. Same as the manufacturing process, the inspection process should also be automatic.

Machine vision emerged as an important new technique for industrial inspection and quality control in the early 1980s. It is proven to be an accurate and inexpensive inspection tool for high volume, labour-intensive and repetitive inspection operations in automotive and manufacturing industries. Machine vision provides the technology and methods for imaging-based automatic inspection and analysis.

However, the reflective surface of the object puts some limitations on traditional methods of machine vision. Generally, the inspection of the reflective surface is performed in a dark environment, as the ambient lighting condition of the working environment makes the reflective surface look highly specular.

This research mainly focuses on overcoming the limitation of traditional machine vision methods. A novel three-dimensional (3D) measuring system is developed to inspect a product with a highly specular surface accurately. This technique aims to combine the concepts of stereo vision and laser triangulation for the 3D reconstruction of the product. This method provides a simple but accurate solution to inspect the reflective surface. The main advantage of this system is that it works robustly even in the presence of ambient light.

The thesis briefly explains the effect of background on the accuracy of the inspection. Also, a thorough comparison of red and blue light lasers in terms of accuracy is described. In addition, the difficulties induced by the nature of the surface in ambient lighting conditions are evaluated. An algorithm is invented to overcome these difficulties.

Along with accurate measurements, it is also essential to detect defects such as dents, bumps, cracks, and scratches present in a product. As these defects are palpable and are not visible by the camera, it is tough to detect them using vision-based inspection techniques in ambient lighting conditions. This thesis focuses on three types of defects: Dents, Bumps, and Scratches. With the proposed 3D measurement system, we can detect the defects of size 0.02mm accurately.

Artificial intelligence (AI) has many applications in the production industries. One of the applications is to inspect the products for defects. However, AI is not used to reconstruct a 3D model of the product with reflective surfaces accurately. In this research, we propose to use machine learning-based techniques for the accurate 3D reconstruction of the product. The one-dimensional (1D) data of the projected laser line is used to train machine learning (ML) and deep learning (DL) models. These models are trained to detect the projected laser line accurately in the presence of ambient light. The detected laser line plays a vital role in creating an accurate 3D model of the product.

Finally, we compare different machine learning-based techniques with the abovementioned stereo-laser technique based on accuracy.

Three-Dimensional(3D) measurement , Quality control , Highly specular surfaces , Ambient lighting conditions , Defects detection , Stereo vision , Laser line projection , Deep learning , Machine learning , One-Dimensional(1D) convolutional neural network , Supervised regression , Recurrent neural network
Publisher's version
Rights statement