This thesis presents a physics-guided neural network model to predict the frequency response function (FRF) at the tool tip and a Bayesian learning method to reach the optimum cutting conditions with the highest material removal rate by a few cutting tests.
The spindle assembly is one of the weakest parts in the machine tool and contributes to the chatter. The undesired vibrations lead to a poor surface finish and can damage the tool, tool holder, and spindle. The FRF of the spindle up to the flange is obtained experimentally by three impact modal tests and the inverse receptance coupling method. The Timoshenko beam based finite elements model is used to calculate the free-free FRF of arbitrary tools and tool holders. Then, the spindle, tool, and tool holder are coupled using the receptance coupling method to obtain the tool tip FRF. The modal parameters are extracted from the FRF to build a dataset of different tools and tool holders geometry with the corresponding modal parameters. A deep neural network (DNN) is then trained using the simulated dataset and fine-tuned using the experimental impact modal tests to overcome the inaccuracies and uncertainties in the model and measurements. The model’s predictions are verified with experimental impact modal tests and showed computationally efficient predictions.
This thesis also presents a hybrid method to find the stable combination of cutting conditions with the highest material removal rate (MRR). The FRF is used to calculate the analytical stability lobes using the zero-order stability solution. Then, the cutting conditions with the highest analytical MRR are found, and the Bayesian algorithm is applied to search for the optimum cutting conditions with few cutting tests. The method is verified on a milling machining center.