Chatter Identification using Multiple Sensors and Multi-Layer Neural Networks

Milling is a core process in many manufacturing industries. The productivity of any milling operation is reflected by the capacity to remove material as fast as possible. However, the material removal rate is highly restricted due to chatter and to all the negative effects that come together with this phenomenon. Researchers have used several methodologies to control it. However, to be able to control chatter, first it has to be detected. For this, different sensors have been used according to the specific conditions and objectives of each particular research. This study discusses the suitability of several sensors for chatter detection by carefully analyzing and understanding the vibrations involved in the phenomena. A series of experiments were performed on a piece of aluminum7075 fixed in an overhang position, making the cutting process vulnerable to vibrations, therefore facilitating the natural appearance of chatter. Since chatter is a resonant phenomenon excited by the tooth passing frequency of the tool against the work piece, it becomes relevant to study the cutting tool and work piece vibration frequencies. Several Multi-Layer Neural Networks, were created and analysed according to the input of the different signals and cutting conditions used during experimentation, in order to evaluate which sensor or combination of sensors could provide a reliable source of information for monitoring systems or complex deep learning approaches. Finally, a conclusion was reached considering not only the accuracy of the ANN’s but the practicability and ease to use of each sensor.