Modelling of FDM process using genetic programming with classifiers for model selection

Fused deposition modeling (FDM) is one of the important rapid prototyping (RP) processes that build parts of complex shapes by sequential deposition of material on a layer by layer basis. Properties such as compressive strength, tensile strength, wear and dimensional accuracy of the FDM fabricated prototype, which depend on the parameter settings of RP machines, have been modeled and studied using computational intelligence methods such as genetic programming (GP), artificial neural network (ANN), regression analysis and fuzzy logic. There is extensive literature that describes the improvement of performance of GP but less attention has been paid to the problem of model selection in GP. In the context of symbolic regression problems, the best GP model selected based on highest accuracy from the training data samples may not always give satisfactory performance on the testing data. It is observed that there are still many models in the population whose performance on testing data are better than the best model with a little compromise on the training error. This reflects the problem of model selection in GP. To tackle this problem to some extent, model selection criteria such as AIC and BIC from statistical learning theory can be used. Alternatively, we have proposed a methodology to improve the model selection in GP by integrating classifiers such as ANN, support vector machines and naïve Bayesian classifier in the framework of GP. The methodology is applied on the compressive strength data obtained from the FDM process. In comparison with the standardized GP approach, the proposed methodology with classifiers is able to evolve the GP model that gives better performance on the testing data.