Intelligent gas metal arc welding process control system: An application of artificial neural network fuzzy inference, and expert system technologies

More and more, manufacturing companies are seeking cost-effective approaches to improve productivity and quality not only because of stronger competition, but also to ensure their company's survival. Intelligence-based, cost-effective engineering tools in the future will depend on the abilities of comprehension, reasoning and learning to achieve higher productivity, quality and reliability of process monitoring and control.; Recently, much research has proposed monitoring the integrity of weld quality and controlling the arc welding processes based on the geometry of the resulting weld bead. In this research, the geometry of the weld bead along with the absence of penetration is represented as a quality indicator. Furthermore, the GMAW (Gas Metal Arc Welding) process is the most widely metal joining process and real-time control of the welding process offers significant potential for reducing or eliminating the unnecessary manufacturing cost and lost man-hours spent repairing defective parts. Therefore, the primary objective of this dissertation is to develop an on-line intelligent GMAW process control methodology with the following unique features: (1) welding quality monitoring and diagnosis; (2) defect recognition and reasoning; (3) welding parameter correction; (4) fast computational algorithm; and (5) new arc sound signal for incremental learning. The implementation concept consists of an integrated learning classification reasoning, and verification algorithm that applies AI (Artificial Intelligence) techniques (artificial neural network, fuzzy logic, and knowledge-based system) to increase the reliability of the intelligent GMAW process control system(GMAW-IPCS). The modified CMAC neural network mimics the computational architecture of the human brain to achieve the intelligent capabilities (learning and pattern recognition), while the fuzzy logic and expert system endows the ability of reasoning and verification to verify and recommend control decisions.; The implementation results show that the presented methodology can be effectively used to estimate the weld bead geometry and spatter, identify the condition of the weld, stabilize the arc, and assure good weld quality. This new approach also has great potential for real-world process control, including processes other than welding, and lends itself to parallel processing.