SUMMARY

Wire arc additive manufacturing (WAAM) is an additive process which utilizes common welding equipment to construct metallic parts from wire feedstock. The process is highly complex and needs intensive monitoring improvements in order for it to be adopted by industry. A key element of WAAM process monitoring is in controlling contact tip to workpiece distance (CTWD). CTWD influences bead geometry, weld penetration, and even weld stability. Due largely to errors in process modeling and thermal conditions, CTWD can vary throughout the WAAM process and lead to unsatisfactory part geometry. Welding sound is an underutilized source of weld information that can provide a wealth of knowledge about the welding process. Advances in machine learning and computational processing are increasing the reliability of sound as an information source. In this thesis, machine learning models including artificial neural networks (ANN), and convolutional neural networks (CNNs), and waveform convolutional neural networks are utilized to predict CTWD from welding sound. Waveform CNN models showed the highest accuracy of 95\% success when classifying CTWD to within 2.54 mm. Mel spectrum coefficients were shown to have increased accuracy when compared to STFT spetrograms. In addition, models were evaluated on an edge device and edge deployment considerations briefly discussed.