Computer Vision Methods for Autonomous Remote Sizing in Manufacturing

In the grand scheme of Industry 4.0, the employment of modern intelligent digital technology has been utilised to facilitate industrial production, leveraging automation to elevate production efficiency. Building upon this, Industry 5.0 takes a step forward, accentuating the concept of human-machine symbiosis. It directs its focus on augmenting human performance within the industry, mitigating errors made by workers, and honing the overarching performance of human-machine systems. Across various manufacturing domains, an escalating demand for this level of automation has been noticed. One such area is the speciality steel industry, whose tasks are the primary consideration of this dissertation. Speciality steel rolling forms the backbone of industrial sectors as diverse as aerospace and oil and gas. The key to the sustained survival of steel plants hinges on the digitalisation of the rolling process. Despite this, a significant number of steel rolling plants in the present day continue to place a heavy reliance on human operators to oversee and regulate the manufacturing process. With a view to securing the safety of workers in high-risk factory environments and optimising the control of steel production, this dissertation puts forth machine vision approaches. These are aimed at supervising the direction of hot steel sections and remotely gauging their dimensions, both conducted in real-time. This dissertation further contributes a novel image registration approach founded on extrinsic features. This approach is then amalgamated with frequency domain image fusion of optical images. The resultant fused image is designated to evaluate the size of high-quality hot steel sections from a remote standpoint. With the integration of the remote imaging sizing module, operators can stay abreast of the section dimensions in real time. Concurrently, the mill stands can be pre-adjusted to facilitate quality assurance. The efficacy of the developed approaches has been tested over real data, delivering an accuracy rate exceeding 95%. This suggests that the approach not only ensures worker safety but also contributes significantly to the enhancement of production control and efficiency in the speciality steel industry.