Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Learning by doing - Data-driven optimization of arc-evaporated TiAlN-coating for cutting tools

Holmquist, Peter LU (2021) KASM10 20211
Centre for Analysis and Synthesis
Abstract
The purpose of this study was to investigate the applicability of data-driven experimental design for the optimization of a hard coating for cutting tools in metal machining. The complexity of metal machining leads to difficulty in theoretically modeling and understanding the interactions and correlations between parameters and performance. Because of this and the importance of incremental performance gains, new and innovative ways to simplify or improve the optimization of a component such as the coating of the cutting tool is of utmost importance. In the context of machine learning, the optimization of black-box, expensive to evaluate functions is commonplace. These methods could help simplify and improve the optimization of parameters... (More)
The purpose of this study was to investigate the applicability of data-driven experimental design for the optimization of a hard coating for cutting tools in metal machining. The complexity of metal machining leads to difficulty in theoretically modeling and understanding the interactions and correlations between parameters and performance. Because of this and the importance of incremental performance gains, new and innovative ways to simplify or improve the optimization of a component such as the coating of the cutting tool is of utmost importance. In the context of machine learning, the optimization of black-box, expensive to evaluate functions is commonplace. These methods could help simplify and improve the optimization of parameters in metal cutting to great effect.
The study focuses on a multilayered TiAlN-coating produced by cathodic arc evaporation where three parameters are varied: aluminum content, thickness of the coating and the substrate bias during the arcevaporation. The algorithm Bayesian optimization is used to optimize the parameters to maximize the tool life during high-speed turning of 316L stainless steel. To evaluate the effectiveness of the algorithm, more traditional methods of optimization such as response surface design is used as comparison. Selected coatings are analyzed further with SEM and EDS to understand the tool wear and its dependence on
the parameters.
The benefit of data-driven optimization is shown over traditional experimental designs in the context of metal machining. Although the function can be approximated by a second order polynomial with interaction effects (R2 = 0.83, Q2 = 0.80) which would indicate that response surface methodology would be effective, the difference between the two methods is clear. If the process had been slightly more complex, the advantages had been even more apparent.
The tool wear shows two competing macroscopic wear mechanisms in notch wear and crater wear. The best performing coatings show very light signs of crater wear compared to the rest of the coatings as well as having a slow rate of notch wear. It is suggested that the difference in crater wear is the result
in a difference in dominating wear mechanism. The coatings exhibiting more crater wear is suspected to be dominated by crater wear as a result of their poor cohesion and toughness.
The notch wear is also investigated and show a pattern of a low, constant wear rate until a "critical point" which occurs at different points for different coatings. A theory is presented which connects the critical point to the plastic deformation of the substrate but this would have to be proven by further investigations. (Less)
Popular Abstract (Swedish)
Data-driven optimering en succé för skärande bearbetning

Optimeringsmetod från maskininlärning mycket lovande vid användning för komplexa processer inom skärande bearbetning

En otrolig mängd av industriell tillverkning idag förlitar sig på ett till synes enkelt koncept, att forma metall genom svarvning eller fräsning. Skärande bearbetning letar sin väg in i många industrier och alla sätt att öka produktivitet eller effektivitet är viktigt då det kan leda till stora besparingar i tid och kostnad. Däremot är förslitningen av skärverktyg ökänt svårt att förutspå och behovet av effektiv optimering av verktygen är stort. Den här studien undersökte appliceringen av en metod använd vid maskininlärning för att optimera en slitstark... (More)
Data-driven optimering en succé för skärande bearbetning

Optimeringsmetod från maskininlärning mycket lovande vid användning för komplexa processer inom skärande bearbetning

En otrolig mängd av industriell tillverkning idag förlitar sig på ett till synes enkelt koncept, att forma metall genom svarvning eller fräsning. Skärande bearbetning letar sin väg in i många industrier och alla sätt att öka produktivitet eller effektivitet är viktigt då det kan leda till stora besparingar i tid och kostnad. Däremot är förslitningen av skärverktyg ökänt svårt att förutspå och behovet av effektiv optimering av verktygen är stort. Den här studien undersökte appliceringen av en metod använd vid maskininlärning för att optimera en slitstark beläggning för ett skärverktyg.
Metoden som heter Bayesisk optimering lyckades maximera livstiden för ett specifikt belagt verktyg vid metallsvarvning. Studien fungerar som ett bevis för potentialen som metoden kan ha för att optimera de komplexa processerna vid skärande bearbetning.
Bayesisk optimering är en algoritm som sekventiellt föreslår nya kombinationer av parametrar att mäta vid. Genom att utnyttja de tidigare mätpunkterna kan algoritmen maximera den förväntade förbättringen jämfört med den nuvarande bästa mätpunkten. På detta vis har varje ny mätpunkt en hög sannolikhet att komma närmre det riktiga optimumet.
I studien applicerades Bayesisk optimering på en beläggning för ett skärverktyg. Tre parametrar för beläggningen varierades och verktygets prestanda utvärderades genom att svarva rostfritt stål och mäta livstiden för verktyget. Efter optimeringen genomförts jämfördes resultaten med statistiska optimeringsmetoder som används inom skärande bearbetning idag. Resultatet var positiva och visar att metoden är lovande som ett generellt optimeringsverktyg men särskilt för komplexa processer.
Användandet av Bayesisk optimering inom skärande bearbetning kan ses som ett naturligt steg i utvecklingen av effektiva verktyg för att optimera skärverktyg. Detta är alltid viktigt att förbättra och små, stegvisa förbättringar kan göra stor skillnad på grund av hur utbrett skärande bearbetning är som tillverkningsprocess. (Less)
Please use this url to cite or link to this publication:
author
Holmquist, Peter LU
supervisor
organization
course
KASM10 20211
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Bayesian optimization, Data-driven optimization, TiAlN, Metal machining, Metal cutting, Cutting tools, Hard coatings, Materials chemistry
language
English
id
9051782
date added to LUP
2021-06-16 15:24:43
date last changed
2021-06-16 15:24:43
@misc{9051782,
  abstract     = {{The purpose of this study was to investigate the applicability of data-driven experimental design for the optimization of a hard coating for cutting tools in metal machining. The complexity of metal machining leads to difficulty in theoretically modeling and understanding the interactions and correlations between parameters and performance. Because of this and the importance of incremental performance gains, new and innovative ways to simplify or improve the optimization of a component such as the coating of the cutting tool is of utmost importance. In the context of machine learning, the optimization of black-box, expensive to evaluate functions is commonplace. These methods could help simplify and improve the optimization of parameters in metal cutting to great effect. 
The study focuses on a multilayered TiAlN-coating produced by cathodic arc evaporation where three parameters are varied: aluminum content, thickness of the coating and the substrate bias during the arcevaporation. The algorithm Bayesian optimization is used to optimize the parameters to maximize the tool life during high-speed turning of 316L stainless steel. To evaluate the effectiveness of the algorithm, more traditional methods of optimization such as response surface design is used as comparison. Selected coatings are analyzed further with SEM and EDS to understand the tool wear and its dependence on
the parameters.
The benefit of data-driven optimization is shown over traditional experimental designs in the context of metal machining. Although the function can be approximated by a second order polynomial with interaction effects (R2 = 0.83, Q2 = 0.80) which would indicate that response surface methodology would be effective, the difference between the two methods is clear. If the process had been slightly more complex, the advantages had been even more apparent.
The tool wear shows two competing macroscopic wear mechanisms in notch wear and crater wear. The best performing coatings show very light signs of crater wear compared to the rest of the coatings as well as having a slow rate of notch wear. It is suggested that the difference in crater wear is the result
in a difference in dominating wear mechanism. The coatings exhibiting more crater wear is suspected to be dominated by crater wear as a result of their poor cohesion and toughness.
The notch wear is also investigated and show a pattern of a low, constant wear rate until a "critical point" which occurs at different points for different coatings. A theory is presented which connects the critical point to the plastic deformation of the substrate but this would have to be proven by further investigations.}},
  author       = {{Holmquist, Peter}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Learning by doing - Data-driven optimization of arc-evaporated TiAlN-coating for cutting tools}},
  year         = {{2021}},
}