Abstract
Wire arc additive manufacturing (WAAM) is a metal additive process that allows for constructing diverse parts using an electric arc and metal wire feed stock. It provides higher deposition rates and larger build areas that make it an attractive technology for industry. An important element of WAAM process control is maintaining specific contact tip to work-piece distance (CTWD). Cumulative errors caused by modeling inaccuracies and thermal conditions can create an increase in CTWD during processing. In the present study, acoustic signals were investigated as a non-invasive form of monitoring this phenomenon and were compared with traditional current-based sensing approaches. Experiments were conducted across a range of controlled CTWD conditions. From the acoustic signals, 35 different statistical features were extracted and machine learning strategies were utilized to correlate them to the CTWD. Random forests composed of 50 trees each were chosen to classify the CTWD into three different levels of CTWD granularity. The ability of random forest algorithms to detect varying levels of CTWD was investigated and the models showed limited performance especially for high granularity CTWD predictions when compared to current sensing approaches. The implications for these results in rapid implementation for high-level process monitoring of process condition are briefly discussed.
