Our objective is to determine how experienced welders have learned to make certain adjustments to the welding parameters based on auditory feedback of the arc sound, and how this feedback control loop can be applied to welding automation. In this paper, we first introduce a technique for sound signal processing, namely Computational Auditory Scene Analysis (CASA). Auditory Scene Analysis has its background in the study of how the human auditory system processes sound into meaningful elements in perception; CASA has the deeper goal of developing computer systems capable of processing real-world sound scenes into an abstract representation of the sound sources as perceived by a human listener. As far as we are aware, the CASA technique has not yet been used in welding applications, and is utilized here to extract characteristics of arc sound in both time and frequency domains, from which a set of features is formed.
A fuzzy-logic (FL) intelligent controller has been designed and implemented, which consists of an if-then rule base and Mamdani’s fuzzy reasoning mechanism. This captures the knowledge of several experienced welders who were asked how they would adjust the welding parameters while listening to the sound of a weld. Membership functions of the FL controller were generated based on the arc sound features. Experiments were conducted to re-shape the membership functions and to tune the controller. Results are presented that demonstrate the ability of using acoustic feedback to maintain proper control of the weld process.