Abstract
Ensuring surface quality in industrial production is crucial to guarantee product functionality, safety and customer satisfaction. This thesis work presents a comprehensive analysis of surface quality control methods in industry, focusing on the comparison between manual and automated control approaches based on Deep Learning (DL) architectures. Manual inspection, often performed by skilled operators, relies on human visual inspection skills honed over time. However, this approach can be repetitive and labor-intensive, resulting in potential fatigue and decreased detection reliability. Manual inspection can struggle to adapt to the inspection pace of production machines, resulting in undesirable variations in defect detection. Automated inspection systems, on the other hand, thanks to artificial intelligence (AI) and in particular DL architectures and machine vision (MV) technologies, offer an attractive alternative. These systems utilize high-speed cameras, image-processing algorithms and advanced DL models such as the architecture neuronal network Denoising Autoencoder (DAE). Such systems are used for unsupervised learning tasks, in particular for feature learning and data reconstruction. The advantages of automatic control include high accuracy, efficiency, non-contact measurement and real-time monitoring capabilities, making it suitable for modern industrial environments. DL architectures have demonstrated outstanding performance in face recognition, speech processing and natural language understanding. In the field of surface quality control, they have the potential to revolutionize defect detection through machine learning from training data without manual feature design. This industry-oriented research analyses the challenges of applying DL algorithms to surface quality control, such as the need for diverse defect datasets. Recent advances and applications are presented that demonstrate the versatility and robustness of DL-based methods in the detection of various surface defects in different industries, including steel production, semiconductor manufacturing and textile processing. Through a comparative analysis, I examined the strengths and limitations of manual and automated surface quality inspection. The results reveal that automated inspection systems offer greater efficiency, consistency and adaptability to different inspection scenarios, while the flexibility and adaptability of manual inspection remain valuable for some applications. Overall, this applied research aims to shed light on the evolution of surface quality control methodologies and the transformative impact of automated inspection powered by AI and MV. As industries strive to meet ever higher quality standards, the integration of automated systems is a compelling solution for the efficient, reliable and adaptive detection of surface defects in modern production environments.