| ptimization and reliability of machining processes are vitally important in today's automated manufacturing systems. To eliminate defective part production on-line monitoring of tool wear and tool fracture is essential. These two problems have to be solved before we achieve the goals of reliable unmanned operation and optimal cutting conditions. Having information about tool wear several control actions can be taken such as implementation of tool position compensation and exchange of the worn out tool for a sharp one. Real time information about tool condition might be used by a process optimizer to recompute the process cutting parameters. Optimization could be done by minimizing important performance indices such as: surface roughness, dimensional accuracy, or process cost.;A neural network method appropriate for on-line estimation of flank wear of screw-on coated inserts during end milling operations is developed. The network consists of three types of networks: Kohonen feature maps, radial basis function networks, and a recurrent neural network. The estimation is done for continuous flank wear rather than for binary (i.e. fresh or worn out) estimations. A reliable and consistent tool wear monitoring scheme is achieved by simultaneously using data from an acoustic emission sensor and from a dynamometer. The neural network is trained for the working range of a specific coated insert end mill. The model includes feature data acquired while varying three milling cutting variables: surface speed, chip load, and width of cut. A reduction of the required training data is achieved by selecting a subset of the possible combinations of the mentioned cutting conditions. The fraction is selected by means of the design of a one-third three level fractional factorial design. The performance of the network is tested with two sets of data gathered at cutting conditions other than the ones used for training. The network's capabilities are evaluated for interpolation and extrapolation inside and outside the volume of training. It is found that the network performs satisfactorily, the resulting error mean is... |
