Abstract-This paper presents a new output partitioning approach with the advantages of constructive learning and output parallelism. Classification error is used to guide the partitioning process so that several smaller sub-dimensional data sets are divided from the original data set. When training each sub-dimensional data set in parallel, the smaller constructively trained sub-network uses the whole input vector and produces a portion of the final output vector where each class is represented by one unit. Three classification data sets are used to test the validity of this algorithm, while the results show that this method is feasible.

Intelligent procedure expert system was developed to select appropriate GTAW procedure in this paper. First, the function design and implementation methods of the welding procedure expert system were introduced. The expert system can present the welding procedure card, multimedia display of welding process, and output function to makes the data sharing more convenient. Then, the database design of the welding procedure expert system based on C/S mode was presented where the expert knowledge was stored. At last, the neural network model was established to realize procedure selection based on the neural network learning ability and the welding case from the database. With the BPNN model, the welding parameters can be obtained based on the input welding conditions.

No abstract

No abstract

In this paper, an output partitioning algorithm is proposed to improve the performance of neural network (NN) learning. It is assumed that negative interaction among output attributes may lower training accuracy when we have only one single network to produce all the outputs. Our output partitioning algorithm partitions the output space into multiple groups according to correlation, with strong correlation within each group. After partitioning, each group employs a learner to train itself. The training results from each group are integrated to produce the final result. According to our experimental results, the accuracy of NN is improved