Experimental Data and Operator Knowledge used for Classifying Welding Quality

M. Kristiansen and O. Madsen (Denmark)


Knowledge acquisition, automation, artificial neuralnetworks, Bayesian network, welding, process-planningmodels.


Automation of many manufacturing processes requires a process-planning model. These models are hard and time consuming to construct because they are often non-linear and the physics of the process is not completely understood. In the literature these models are usually constructed from experiments. However, the operator possesses a lot of knowledge about the process, but it is rarely used because it is hard to formalize. This paper presents the results of a feasibility study in the use of the operator knowledge for building a process-planning model. A method is presented in which operator knowledge in the construction of process-planning models is used. The method consists of three steps: 1) Generating knowledge from operator interviews, 2) Construction of training data, 3) Construct process-planning models. Use of operator knowledge compared with use of experiments shows that the same precision of the process-planning modes is achieved but the time consumption is reduced considerably.

Important Links:

Go Back