Industrial Structures such as pipes develop issues over time which are primarily due to ageing, corrosion, cracks, or other forms of mechanical defects which could potentially lead to leakages or further uncontrollable fatal occurrences. As a result, inspections of such industrial structures are therefore of extreme importance. While prevailing inspection techniques are usually sensor based, time-consuming and labour-intensive, the application of robotic systems significantly reduces the human effort required for the same level of inspection. In this paper, implementation of  a vision-based approach and a suitable low-cost solution was developed for detecting defects in industrial structures such as pipelines and hollow cylindrical structures. The camera embedded in the robot is used to identify defects such as cracks, corrosion and blockage in real-time through a systematic algorithm of dataset training (1500 datasets) representing these conditions. This paper also focuses on the modelling and analysis of the dataset training of a pipeline inspection robot. The machine vision system thus developed was capable of effectively detecting defects even in low light conditions with a mean average precision of 91%.