Automated fusion welding processes play a fundamental role in modern manufacturing industries. The proliferation of joint geometries together with the large permutation of associated process variable configurations has given rise to research into complex system modelling and control strategies. Many of these techniques have involved monitoring of not only the electrical characteristics of the process but visual and acoustic information. Acoustic information derived from certain welding processes is well documented as it is an established fact that skilled manual welders utilise such information as an aid to creating an optimum weld. The experimental investigation presented in this thesis is dedicated to the feasibility of monitoring airborne acoustic emissions of Submerged Arc Welding (SAW) for diagnostic and real time control purposes. The experimental method adopted for this research takes a cybernetic approach to data processing and interpretation in an attempt to replicate the robustness of human biological functions. A custom designed audio hardware system was used to analyse signals obtained from bead on mild steel plate fusion welds. Time and frequency domains were used in an attempt to establish salient characteristics or identify the signatures associated with changes of the process variables. The featured parameters were voltage / current and weld travel speed, due to their ease of validation. However, consideration has also been given to weld defect prediction due to process instabilities. As the data proved to be highly correlated and erratic when subjected to off line statistical analysis, extensive investigation was given to the application of artificial neural networks to signal processing and real time control scenarios. As a consequence, a dedicated neural based software system was developed, utilising supervised and unsupervised neural techniques to monitor the process. The research was aimed at proving the feasibility of monitoring the electrical process parameters and stability of the welding process in real time. It was shown to be possible, by the exploitation of artificial neural networks, to generate a number of monitoring parameters indicative of the welding process state. The limitations of the present neural method and proposed developments are discussed, together with an overview of applied neural network technology and its impact on artificial intelligence and robotic control. Further developments are considered together with recommendations for future areas of research.