MACHINE SETUP ABNORMALITY DETECTION USING MACHINE VISION AND DEEP LEARNING

Machine tools can perform manufacturing operations autonomously with minimal involvement of machine operators. However, significant human expertise is required while machine setup to minimize machine faults, downtime, and scrap parts for achieving better productivity. The machine setup involves activities such as the presence/absence of a correct cutting tool in the spindle, adequate placement of workpiece on the fixture, ensuring proper coolant flow, clogging of chips, etc. The present machine shops employ skilled human operators to perform these tasks with a checklist for ensuring completeness. The outcomes of the process are dependent on the judgment, consistency, and skills of the operator. Machine vision is considered a low-cost, low-error, and higher consistency substitute for skilled humans in industrial processes involving repetition or fatiguing tasks. The existing machine vision systems developed for dimension measurement or surface defect detection application cannot detect machine setting abnormalities due to the distinct requirements. The present work explores the development of a robust algorithm by augmenting machine vision with the deep learning algorithm You Only Look Once (YOLO) to directly extract features from the captured images. The object detection algorithm based on YOLO-v2 and ResNet-50 is implemented to detect and segregate the elements of interest from different shopfloor images. The algorithm is trained using labeled image datasets generated for several machine setup abnormalities. An interactive Graphical User Interface (GUI) is developed and integrated with the model to implement the proposed framework on the manufacturing shop floor. The developed system was implemented to detect machine setup abnormalities by considering different case studies. The study demonstrated robust detection abilities of the algorithm, offering a potential solution to minimize dependence on human operators for machine setup abnormalities.