Characterisation of cutting tool wear by monitoring system vibration response

Konsistentna izdelava komponent, ki izpolnjujejo visoke standarde kvalitete, zahteva visoko zanesljivost proizvodnega procesa. V primeru mehanskega odvzemanja materiala visoka zanesljivost zahteva spremljanje stanja obrabe orodja za odrezavanje, saj le–ta vpliva na kvaliteto površine obdelovanca. Preko spremljanja obrabe orodja lahko izvajamo proces prediktivnega vzdrževanja in s tem zmanjšamo izmet v fazi proizvodnega procesa. Velik potencial na tem področju predstavlja uporaba posrednih metod, saj je moč v realnem času detektirati stanje rezilnega orodja. Glavni izziv na področju razvoja posrednih metod merjenja obrabe orodja predstavlja karakterizacija povezave med obrabo in merjeno veličino. V sklopu obravnavanega dela smo razvili metodo posrednega merjenja obrabe preko spremljanja vibracijskega odziva orodja. Merjene signale smo v skladu z metodo MSSA (ang. Multi-channel Singular Spectrum Analysis) za razcep časovnih vrst razčlenili na posamezne komponente. Na osnovi izbranih cenilk porazdelitve MSSA komponent vibracijskega odziva smo z uporabo nevronskih mrež preko regresijskega pristopa in klasifikacije identificirali stopnjo obrabe rezilnega orodja. Pri tem smo pokazali, da cenilke parametrov porazdelitve MSSA komponent omogočajo natančnejšo identifikacijo obrabe kot cenilke merjenih signalov.Consistent production of components meeting high-quality standards requires high production process reliability. In the case of mechanical material removal, high reliability requires monitoring the state of wear of the cutting tool, as it affects the surface quality of the workpiece. A predictive maintenance process can be implemented by monitoring tool wear in order to reduce the number of low-quality pieces. The use of indirect methods carries significant potential in this field, as it enables detecting the cutting tool\u27s condition in real-time. The main challenge in developing indirect methods for monitoring tool wear is characterizing the relationship between wear and the measured quantity. In this work, we developed a method of indirect wear measurement by monitoring the tool\u27s vibration response. The measured signals were split into individual components by the MSSA (Multi-channel Singular Spectrum Analysis) method for time series decomposition. Based on selected estimators of the distribution of MSSA components of the vibration response, the degree of wear of the cutting tool was identified using neural networks through a regression approach and classification. It was demonstrated that the distribution parameter estimators of MSSA components enable more accurate identification of wear than estimators of the measured signals