Neural Networks For Diagnostics Of Metal Cutting Machine Modules

Abstract

The work is devoted to solving the problem online diagnostics of machine tools modules using data-based models. The authors propose a diagnostic method that includes models based on long short-term neural memory networks as a repository of frequency reference values. Data for training neural networks is a frequency spectrum reflecting the oscillations of the tool and the workpiece normal to surfaces caused by the presence of a manufacturing defect in the module element of a metalworking machine. Neural network model with long short-term memory are used for approximation the nonlinear frequency characteristics. For classification of module defects proposed a second neural network that compare the neural network model of the reference spectrum with the spectrum obtained from the actual quality parameters of the part in real time, determine the sources of defects. To evaluate the effectiveness of the method, a series of experiments were carried out with the definition of defective machine modules. An experimental result of the application of proposed method is given. Keywords-cutting machining, operational diagnostic, long short term memory, deep neural networksI.