| A study was conducted on the possibility of implementing diverse sensors for closed loop control of penetration depth in laser welding. In order to control the process, a monitoring scheme was developed, and due to the specific nature of the problem, custom monitoring software was developed. The sensors that were used range from an inexpensive light diode to an expensive laser beam monitoring system and imaging radiometer. The corresponding depth of penetration was recorded by cutting the welds along the middle and measuring the depth of the fusion zone.; Based on recorded signals and measuring depth of penetration, the neural network was designed to predict the depth of penetration. For a specific task, the Radial Basis Function network was more suitable than the widely used Multi-layer Perceptron network. To close the control loop, a fuzzy-logic controller was implemented to generate control signals for laser power and welding speed.; Finally, as an alternative to destructive testing to train the neural network, the finite element model was developed using commercially available software. In the model, major thermal effects were attributed to heat transfer by conduction, convection and radiation. A moving laser beam was modeled as a set of moving heat inputs. To make the model more accurate, material loss due to the evaporation was modeled, as well. |
