This paper describes the effort of implementation of object detection architectures into data mining for physical asset management purpose. Data mining in asset management often relates to the activity of recovering information from engineering diagrams in PDF format such as Piping and Instrumentation Diagram (PID). The existing study around the world revolves around the basic component detections without the aim to produce a practical methodology for usage in industry. This study started with how the final output should be according to normal industry practice and standards such as ISO14224. It is hypothesized that a good pre-trained model for infrastructure detection on PID can be developed to suit industrial needs. Three different deep learning architectures were used in the study are Faster R-CNN, YOLOv3 and Yolov5. YOLOv3 and Faster R-CNN provides much consistent training results compared to Yolov5, hence they are better suited for further development.