Force modeling of inhomogeneous material using unsupervised learning and model identification

In this paper a force modeling method for inhomogeneous materials is introduced. This modeling method is based on samples during haptic operations, for instance presses. Using biomimetic unsupervised learning, the model is primarily identified, including the distribution and material parameters of the inhomogeneous regions, and in this learning the parameter initial estimation, the principal component analysis, the cluster analysis and the quadratic discriminant analysis are applied. Then the material parameters and boundaries of the different regions are accurately optimized using the Gauss-Newton algorithm. Further more the modeling method is tested and verified by a set of simulations. In addition, the suggestions and prospect of the modeling method are also given.