Sensors and Actuators A, 37-38 (1993) 589-595 589 Experimental comparison of sensors for tool-wear monitoring on milling L J Barrios, A Rulz, D Guinea, A Ibafiez and P Bustos Inst~ruto de Automatlca Industrral (CSIC), N-III Km 22,800 desvro a La Poveda, Arganda de1 Rey, 28500 MadrId (Spam) J Etxeberrla TEKNIKER, Av Otaola 20, 20600 .&bar Gwpuzcoa (Spam) Abstract This paper compares spmdle motor current, vlbratlon and acoustic enuss~on sensors for tool-wear momtormg, the methodology and the results are also gven Companson of the data IS not restrrcted to raw ones, but It IS apphed to a broad set of denved features by means of previously developed procedures The analysis of vanance 1s carried out for the best found features Correspondmg models are bmlt to explam the measurements vanabdlty, and sigmficatlve dtierences are tested Some quahty indexes are specially mtroduced for the assessment Rnally, sensors are classified accordmg to its sensltlvlty to the tool wear From the point of view of the application the observed dependence of measurements on the used tool and on the cuthng position, and the discovery of mteractions between this last factor and tool wear, are very remarkable facts Introduction It 1s a well-known fact that the nonavalab&y of effective sensors systems to allow unmanned machmmg 1s a crucial problem to reach full spreadmg of com- puter-integrated manufactunng To solve this problem many kinds of sensors [I] and signal-processmg techmques to extract mformatlon from raw data [2] have been studied for various func- tions Tool-wear estunation m real time has been shown as one of the most cntlcal methods and it has been extensively studied Selection of sensors and signal-processing techniques are mainly based on researcher’s experience and mtu- ltlon Many procedures have been proposed, but no rigorous comparison of sensors has been found m the literature Further, although machmmg processes are characterized by a great vanabdlty, the research reports seldom include statistical tests It is therefore not possl- ble to know the confidence level when one claims that a sensor 1s sensltlve to a process vanable This paper presents a comparison of sensors momtor- mg three different process variables (1) acoustic emls- slon, (n) vlbratlons and (m) spmdle motor current The aim 1s to analyse the most appropnated vanable to estimate tool wear, m the analysed condltlons m mllhng The expetlmental design techmques accordmg to the obJectives have been applied Experimental design Destgn and model A model of nonrephcated factonal design was ap- plied Selected factors were tool-wear state, tool and cutting posltlon on the machme table We are mainly interested to know the effect of the tool-wear state on sensor signals However, tool and cutting posltlon were also important as well as their adverse reactlon on the mam factor In addition, we use this nonrephcated factorial design to look for possible mteractlons be- tween the factors Three levels were defined for tool-wear state low (O-O 2mm), medium (04-O 6mm) and high (0 8- 1 0 mm) Five tools were used, each one considered as a level of the second factor Finally, three levels of the cutting posltlon on the machme table were also defined inner, central and exterior Signal features, zyxwvutsrqponmlkjihgfedcbaZYXW used as responses m the analysis, ~111 be presented m the paragraph on samphng The following model 1s proposed for each featuref A,& = fl + =, + 8, + l-k + @&, + (ar)& + (pr),k + u& whereA;,, represents the value of the feature at level I of wear state, level zyxwvutsrqponmlkjihgfedcbaZYXWVUT J of tool and level k of cuttmg posltlon The global effect ,u measures the average response The mam effects a,, B, and r, measure the contnbutlons of wear state at level I, tool at level J and cuttmg position 0924~4247/93/%6 00 @ 1993 - Elsev~er Sequoia All nghts reserved