The purpose of this paper was to monitor the occurrence of clogging of powder-feeding nozzles and the type of nozzle clogging during laser-directed energy deposition (LDED).
The experiments used four powder-feeding nozzles in LDED, focusing on establishing a normal condition and simulating 15 fault scenarios specific to the nozzles. The discrimination of the state of the powder-feeding nozzles was realized with the assistance of acoustic emission sensors and combined with machine learning algorithms. Throughout this study, the time–frequency domain signals and their characteristics of acoustic emission signals during normal settling and various types of powder-feeding nozzle faults were compared and analyzed. The applicability of acoustic emission in the diagnosis of LDED powder flow faults was explained in conjunction with machine learning.
This study demonstrated that the integration of acoustic emission sensors with machine learning can be used to detect the presence of faults in powder-feed nozzles and locate the faulty nozzles during the LDED process. Acoustic emission was found to be more sensitive to changes in the number of faults in the powder-feed nozzle than to changes in the location of the faulty nozzle.
It was found that the diagnosis of powder-feeding nozzle faults during LDED could be achieved using an acoustic emission sensor. By integrating an acoustic emission sensor with a machine learning algorithm, a prediction accuracy of up to 100% was achieved in differentiating between normal and abnormal nozzle conditions.
