Marjan Korosec

Intelligent tool path generation for milling of free surfaces using neural networks

Abstract The presented paper has an intention to show how with the help of Artificial Neural Network (ANN), the prediction of milling tool-path strategy could be made in order to establish which milling path strategy or their sequence will show the best results (will be the most appropriate) at free surface machining, according to set technological aim. In our case the best possible surface quality of machined surface was taken as the primary technological aim. Configuration of used Neural Network (NN) is presented, and the whole procedure is shown on an example of mould, for producing light switches. The verification of machined surface quality, according to average mean roughness, Ra, is also being done, and compared with the NN predicted results.

Identification and optimization of key process parameters in noncontact laser scanning for reverse engineering

This Extended Technical Note shows that the final accuracy level of reverse engineered surfaces depends on scanning distance and scanning angle of the laser beam, hence it also depends on the morphology of the scanned objects. On scanning strongly curved objects, such as the ones with free form surfaces, the distance and impact angle of the laser beam are changing continuously during the scanning process. On the basis of these, two critical parameters are specified for the design model, which make it possible to predict these two factors in advance, depending on the morphology of the scanned object. First, a mathematical-statistical design model of the scanning process is generated, which relies on ANOVA (analysis of variance) and DOE (design of experiments). In the next step, a fitness function is optimized by the genetic algorithm (GA) program. This optimization improves the accuracy of the final scanned surfaces, in terms of the minimum standard deviation values of reverse engineered 3D surface model. The proposed approach was confirmed in a case study, which is presented at the end of this Technical Note.

Determination of flow stress properties of machinable materials with help of simple compression and orthogonal machining test

Abstract This paper describes a possible method for predicting values of orthogonal metal cutting properties such as shear angle, cutting force etc., on a basis of the well known Hollomon equation, using a simple compression test in order to avoid any cutting experiments. There are two possibilities: the flow stress properties can be obtained from an independent material test; or by measuring the active and passive cutting forces from the orthogonal machining test itself. This paper is concerned with a material flow stress equation, including the effects of strain (ϵ), strain rate ( ϵ ) and temperature (T), which is one of the five equations that have to be solved in simulation analysis with the finite element method. In finding a solution for those five equations, it is necessary to dispose of flow stress properties by rearrangement of the Hollomon equation and so making it usable for cutting process investigation. The rearrangement is described in this paper.

Technological information extraction of free form surfaces using neural networks

The aim of this paper is to show how to predict the accurate machining technology for the particular free form NURBS or B-spline surface. Since that kind of a surface is very hard to describe in an analytical manner, the topological and geometrical information about the surface was acquired with the help of self-organized neural networks (NNs) and first- or second-order statistic parameters. It is proved that the most significant parameter in this process is the curvature, especially when rapid changes of curvature on a free form surface occurred. As the Gaussian distribution of surface curvatures and slope gradient data were presumed, the mean and variance was used for one-dimensional data presentation, and the Hebbian output data vector was used to assess probability, density function and distribution of the presented data. For collecting the maximum amount of surface information, the principal component analysis method inside the Hebbian NN was used.

Neural Network Based Quality Increase Of Surface Roughness Results In Free Form Machining

This paper concerns with free form surface reorganization and assessment of free form model complexity, grouping particular surface geometrical properties within patch boundaries, using self organized Kohonen neural network (SOKN). Neural network proved itself as an adequate tool for considering all topological non-linearities appearing in free form surfaces. Coordinate values of point cloud distributed at a particular surface were used as a surface propertys descriptor, which was led into SOKN where representative neurons for curvature, slope and spatial surface properties were established. On a basis of this approach, surface patch boundaries were reorganized in such a manner that finish machining strategies gave best possible surface roughness results. The patch boundaries were constructed regarding to the Gaussian and mean curvature, in order to achieve smooth transition between patches, and in this way preserve or even improve desired curve and surface continuities, (C2 and G2). It is shown that by reorganization of boundaries considering curvature, slope and spatial point distribution, the surface quality of machined free form surface is improved. Approach was experimentally verified on 22 free form surface models which were reorganized by SOKN and machined with finish milling tool-path strategies. Results showed rather good improvement of mean surface roughness profile Ra for reorganized surfaces, when comparing to unorganized free form surfaces.