Polynomial Chaos-Kriging approaches for an efficient probabilistic chatter prediction in milling. (October 2020)

• Record Type: Journal Article
• Authors: Totis, G.; Sortino, M.
• Is Part Of: International journal of machine tools & manufacture. Volume 157(2020)
• Publication Date: 2020-10
• Subjects: Milling -- Dynamics -- Chatter -- Probabilistic -- Polynomial Chaos-Kriging -- Monte Carlo; Machine-tools -- Periodicals; Manufacturing processes -- Periodicals; Machines-outils -- Périodiques; Fabrication -- Périodiques; Electronic journals; 621.902
• DOI: 10.1016/j.ijmachtools.2020.103610
• Languages: English
• ISSNs: 0890-6955

Abstract

Abstract: After more that 60 years of investigation, chatter vibrations in metal cutting are still a major cause for poor surface finish and machine tool damage. In order to avoid undesired machining conditions, chatter prediction algorithms may be applied to draw stability charts that allow a preliminary identification of the safe areas. Nevertheless, the stability boundaries are sensitive to the variations and uncertainties of the dynamic milling model coefficients. Thus, the accuracy and reliability of the obtained predictions can be inadequate for many industrial applications. For solving this problem, robust methods were recently devised that are fast but usually too conservative. On the other side, probabilistic approaches were also developed to estimate the probability of instability for a given combination of cutting parameters, by taking into account the statistical distributions of model coefficients. Probabilistic approaches allow a less conservative, risk-aware selection of stable cutting conditions. Unfortunately, their application is still very limited due to the required large amount of computational power and time. In this work, three novel probabilistic methods based on Polynomial Chaos and Kriging metamodels (PCE, KRI and PCK) were compared to state of the art probabilistic algorithms (MC, MC-SPA, DRM-SPA, RCPM). The numerical analysis and the experimental validation proved that MC-SPA, DRM-SPA, RCPM and PCE are too rough and thus needless for industrial applications. On the contrary, KRI and in some cases also PCK showed an excellent accuracy together with significantly shorter elaboration time than that required by the reference Monte Carlo (MC) technique. Graphical abstract: Highlights: Probabilistic approaches for chatter prediction in milling are inaccurate or slow Polynomial Chaos-Kriging approaches are applied for the first time in this context Kriging has the same accuracy but it is two times faster than Monte Carlo Experimentally unstable points are better predicted by the novel probabilistic lobes The novel probabilistic lobes are less conservative than the robust stability lobes