Monitoring the extrusion state of fused filament fabrication using fine-grain recognition method

In the fused filament fabrication (FFF) printing process, the polymer extrusion that is extruded from the nozzle is the key element, which is an important stage between the raw material and the production molding. The stability of its state will directly affect the quality of the printed product. In this work, an effective method is provided to monitor and identify the state of the extrusion during the production process. First, four basic extrusion state patterns of extrusion are analyzed, revealing extrusion recognition as a fine-grained task. Then, a targeted, weakly supervised data-enhanced deep learning recognition network is established, which utilizes attention maps from deeper, high-level features during the training to adjust the region of interest of the input images to improve the recognition granularity. In the next step, the experiments are established to compare the performance of this work's method with that of traditional deep learning, and the accuracy of this work's method reaches 98%, which significantly exceeds that of the traditional deep learning method, illustrating the effectiveness of this work's method for extrusion recognition in printing scenarios. Finally, the method of this work is applied in the field production data for process monitoring verification, and the results show that the method of this work can effectively and agilely recognize the extrusion pattern at the nozzle. This method can provide a potential monitoring tool for extrusion of FFF.