Methods are provided for utilizing machine learning operations configured for use in processing missing pieces of visual data in image data to predict potential location of defects and/or damage in storage device disks. These predictions can allow for a sufficient ability to categorize disks during storage device quality inspections. This can allow for quality inspections to conclude before all areas of the disk surface are scanned. Because less surface area of the disks within the storage device require scanning, the time required for quality inspection scanning prior to deployment can be greatly reduced. Additionally, the partial scans occurring prior to deployment may be supplemented or updated after deployment through the performance of a dense scan. These secondary scans can be configured to scan all previously unscanned areas during storage device downtimes or in response to an environmental trigger such that the storage device will not exhibit any loss in performance.