A Hybrid Approach Integrating Deep Learning and Knowledge Graph for Detection and Recognition of Machine Parts

While machine learning powers impressive computer vision systems, they lack the human advantage of general world knowledge. This means they struggle to interpret visual data with humans' same richness of understanding. Accurate machine part detection is critical for intelligent manufacturing and automation. This paper employs a drilling machine as an example to demonstrate the approach to machine part identification. The model uses a specialized computer vision technique called Named Part Recognition (NPR) to recognize image components even with minimal training data. The model shows impressive accuracy in identifying parts of a drilling machine. Detecting machine parts is pivotal for identifying faulty components and averting malfunctions. It consistently detects the correct parts with a success rate between 89.6% and 92.5% (precision). Additionally, it rarely mistakes other objects for parts (high specificity exceeding 95%). This balanced performance is further highlighted by the overall accuracy consistently hovering around 95.5%.