Controlling 1000 amps using neural networks

The continued effort to improve working conditions and efficiency in fusion welding has increased automation and taken the operator further from the workpiece. This inherently has increased the demand for improved monitoring and control systems to cope with the increase in throughput. The paper describes an application of two network architectures to control submerged arc welding-a high current, low voltage automatic joining process. A logical discriminator, implemented in hardware is used to identify time/amplitude return echoes derived from ultrasonic interrogation of the arc vicinity and a Kohonen feature map is used to classify arc sound.