Industrial machineries wear down by nature after certain years of operation, which could lead to critical break-down and a costly repair. However, a regular overhauling process can stop the system from wearing down and malfunctioning. Overhauling consists of disassembling, inspecting and reassembling the components to ensure that each part is in serviceable condition. The main workflow of the overhaul process can be generalized to removing, disassembly, cleaning, inspection, replace/repair, reassembly, installing and testing.
Inspection is the core of the overhauling process, in which all components must be inspected for damages and anomalies, and repaired or replaced if needed. The components consist of parts of various sizes such as rotors and covers, as well as small parts and fasteners such as bolts, screws, washers, and nuts that hold different parts together. Therefore, the inspection involves identifying damages, sorting and classifying the fasteners, and placing them into compartments for further reuse, which are performed manually in overhaul plants by technicians. The technicians are more likely to be exposed to a broad range of occupational hazards, noise, vibration and different kinds of radiation, and chemicals. Some inspection tasks can be tedious and time-consuming, which increases the errors, especially under time pressure. Moreover, the inspection tasks have their specific challenges: 1) the number and the similarity of the fasteners and small parts that must be manually sorted, classified and packaged, 2) characteristics and nature of metallic parts and surfaces, and 3) different range of damages which need to be inspected. These challenges introduce opportunities to use machine learning, computer vision and automation for inspection tasks in overhaul processes. To assist technicians in overhauling processes, we conducted an experimental research on developing supporting systems using deep learning and computer vision, including damage detection, sorting, classification, and packaging of the fasteners used in the machineries. We used a deep learning based approach to sort, classify and place the components inside compartments. Deep learning based methods are characteristically data-centric algorithms and their performance depends on the availability of a dataset of images. We collected 7 experimental different datasets for fastener damage detection, surface damage detection, sorting, single-view classification, multi-view classification, classification with mixed real and synthetic images, and bin picking . Having an accuracy of over 99\% in damage detection and classification, over 99\% in sorting, over 99\% in classification and over 70\% in bin picking, our approaches imply a possible usage in overhaul processes. Using these applications reduces the the time and human error of the overhauling process.
«
Industrial machineries wear down by nature after certain years of operation, which could lead to critical break-down and a costly repair. However, a regular overhauling process can stop the system from wearing down and malfunctioning. Overhauling consists of disassembling, inspecting and reassembling the components to ensure that each part is in serviceable condition. The main workflow of the overhaul process can be generalized to removing, disassembly, cleaning, inspection, replace/repair, reas...
»