Fuzzy logic control of CO2 laser welding

Abstract

CO2 laser penetration welding is well known as a complex phenomena which makes modeling by mathematical techniques difficult and computationally intensive. Expertise in this field generally is acquired with experience and to some extent is empirical and intuitive in nature. To embed this knowledge in a closed loop control system an inference process is required that can encode the knowledge (in the form of rules) from an expert into a computer numerical algorithm. To this end an expert CO2 laser welding system is being developed which makes use of a video camera system programmed as a fuzzy controller to adjust in real time (up to rate of 28 Hz) welding speed to achieve and guarantee full penetration in mild steel.In the development system a Lumonics VFA 1700 CO2 laser is controlled through the CNC interface by an external 80386DX based PC. The PC ISA bus connects a digital servo motion controller and a frame grabber with on board digital signal image (DSP) processing capabilities. A standard RS-170 video camera with simple filtering provides images of the weld pool and plasma generated from the interaction to the frame grabber. The on board TI320C30 DSP processor has been programmed to first extract numerical information from the images suitable as system inputs to the controller. Further, the DSP processor has been programmed as a complete fuzzy logic controller (FLC). Input fuzzification, rule evaluation and output defuzzification processes all occur on the image processing board working independently from the host PC. In this paper we demonstrate using only the simplest image processing capabilities of the system and a minimum number of expert rules how the controller can recognize good penetration and correct for poor penetration and burn through conditions.CO2 laser penetration welding is well known as a complex phenomena which makes modeling by mathematical techniques difficult and computationally intensive. Expertise in this field generally is acquired with experience and to some extent is empirical and intuitive in nature. To embed this knowledge in a closed loop control system an inference process is required that can encode the knowledge (in the form of rules) from an expert into a computer numerical algorithm. To this end an expert CO2 laser welding system is being developed which makes use of a video camera system programmed as a fuzzy controller to adjust in real time (up to rate of 28 Hz) welding speed to achieve and guarantee full penetration in mild steel.In the development system a Lumonics VFA 1700 CO2 laser is controlled through the CNC interface by an external 80386DX based PC. The PC ISA bus connects a digital servo motion controller and a frame grabber with on board digital signal image (DSP) processing capabilities. A standard RS-170 video...