Image Processing Based Machine Vision System For Datum Setting In Computer Numeric Control Machine

Datum setting is a preliminary process that is done prior to any drilling or milling process in Vertical Machining Centers. It is performed either manually using a trial-and-error method or with the assistance of touch trigger probes. In the manual method, the operator brings the cutting tool closer to the workpiece surface until a feather touch is achieved at the desired reference plane. This method is time consuming and prone to errors. The alternative for this would be the usage of touch trigger probes. However, touch trigger probes are expensive. This research work presents an Image Processing based Machine Vision mechanism for datum setting in Computer Numeric Control machine. The Datum Setting System network (DSS-net), a single stage object detection algorithm for tool detection employs perimeter crossing as the detection principle to indicate zero point on the reference plane. The DSS-net is trained on a manually curated dataset using several YOLO variants by varying batch sizes and workers in the GPU. Finally it is tested on CNC image sequences and achieves a 97.3% mean Average Precision (mAP) and F1-score of 99.9% for tool detection. DSS-net’s performance and its average inference rate of 50.85 frames per second meet the necessary standards for datum setting in CNC machines.