A real-time approach for failure detection in material extrusion process based on artificial neural network

This paper aims to detect the printing failures (such as warpage and collapse) in material extrusion (MEX) process effectively and timely to reduce the waste of printing time, energy and material.

The approach is designed based on the frequently observed fact that printing failures are accompanied by abnormal material phenomena occurring close to the nozzle. To effectively and timely capture the phenomena near the nozzle, a camera is delicately installed on a typical MEX printer. Then, aided by the captured phenomena (images), a smart printing failure predictor is built based on the artificial neural network (ANN). Finally, based on the predictor, the printing failures, as well as their types, can be effectively detected from the images captured by the camera in real-time.

Experiments show that printing failures can be detected timely with an accuracy of more than 98% on average. Comparisons in methodology demonstrate that this approach has advantages in real-time printing failure detection in MEX.

A novel real-time approach for failure detection is proposed based on ANN. The following characteristics make the approach have a great potential to be implemented easily and widely: (1) the scheme designed to capture the phenomena near the nozzle is simple, low-cost, and effective; and (2) the predictor can be conveniently extended to detect more types of failures by using more abnormal material phenomena that are occurring close to the nozzle.