Machine vision based adaptive online condition monitoring for milling cutter under spindle rotation

Tool condition monitoring (TCM) is an important guarantee for quality evaluation of products and parameter optimization of machining operations. The direct methods of TCM have made significant progress in condition recognition and wear measurement. However, these methods based on a single image that reflects the tool condition inevitably bring downtime to the machine tool. Moreover, a single image cannot reflect the tool wear characteristics integrity because the morphology of tool wear is complex. Regarding the issue above, the aim of this paper was to adaptively online monitoring for milling cutters. Firstly, tool condition image sequence (TCIS) is proposed in successive images to express and enhance tool wear characteristics from multiple angles. Secondly, the time-sequential gradient map between adjacent images is constructed based on histograms of oriented gradient. It is used to capture the initial frame of TCIS. Then, the subsequent images are encoded into the classification model. A logistic regression algorithm is applied to train the classification model to capture the end frame of TCIS. Finally, the tool wear area is located by balancing the rectangular box of wear area and benchmarks of wear measurement and is tracked based on the motion model and the local search. In the experiment of accelerating milling cutter life and three different failure phenomena, the recognition accuracy in the initial and end frame of TCIS is 100%. The average measurement accuracy of flank wear based on the proposed method in two experiments is up to 97.02% and 94.71%, respectively. These operations are automated online and provide complete data support for TCM.