Please cite this article in press as: Corne R, et al. Study of spindle power data with neural network for predicting real-time tool wear/breakage during inconel drilling. J Manuf Syst (2017), http://dx.doi.org/10.1016/j.jmsy.2017.01.004 ARTICLE IN PRESS G Model JMSY-524; No. of Pages 9 Journal of Manufacturing Systems xxx (2017) xxx–xxx Contents lists available at ScienceDirect Journal of Manufacturing Systems j ourna l h omepage: www.elsevier.com/locate/jmansys Study of spindle power data with neural network for predicting real-time tool wear/breakage during inconel drilling Raphael Corne a,b , Chandra Nath a,∗ , Mohamed El Mansori b , Thomas Kurfess a a School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, GA, USA b Arts et Métiers ParisTech, Aix-en-Provence, 13617, France a r t i c l e i n f o Article history: Received 27 September 2016 Received in revised form 30 November 2016 Accepted 13 January 2017 Available online xxx Keywords: Spindle power data Digital manufacturing Neural network Wear prediction Drilling Superalloys a b s t r a c t Digital manufacturing systems are determined to be a major key to enhance productivity and quality mainly due to real-time process monitoring and control capability with instant data processing. During machining, such systems are anticipated to excerpt reliable data within a short time-lapse, monitor tool wear progress, anticipate its wear and breakage, alert the machinist in real time to avoid unexpected failure of tool or machine, and help obtaining quality products. This is vital, especially, when drilling Ni-/Ti-based superalloys because catastrophic failure and premature breakage of tools occur in random manner due to aggressive welding and chipping of tool including the rake and/or flank faces and tool corner. Nowadays, spindle power data are easy to collect directly from modern machine tools and can be made available in production floor for such real-time data processing. This work aims to evaluate spindle power data for real-time tool wear/breakage prediction during drilling of a Ni-based superalloy, Inconel 625. Experiments were performed by varying speed and feed. Spindle power data were collected from the power meter (also called load meter) to feed into the neural network (NN) for functional processing. To understand the reliability of the spindle power data, force data were also collected and compared. The results show that the trends of these two different types of data over cutting time are similar for any feed and speed combinations. The error in NN prediction from actual wear was found to be between 0.8–18.4% with power data as compared to that between 0.4–17.9% with force data. Findings suggest that spindle power data integrated with the artificial intelligence (NN) system can be used for real-time tool wear/breakage monitoring and process control, thus appreciate digital manufacturing systems. © 2017 Published by Elsevier Ltd on behalf of The Society of Manufacturing Engineers. 1. Introduction In recent years, the manufacturing world has made considerable progress towards digital intelligent manufacturing technologies and systems – from data processing to shop floor data mining to final products [1]. Many different intelligent hardware and soft- ware, including sensors and data acquisition (DAQ) systems are being developed to improving both manufacturing productivity and product quality [2,3]. Nonetheless, a syndrome called “DRIP – Data Rich Information Poor” is still common to engineers and machine operators at the manufacturing shop floor [4]. A major- ity of data are analyzed offline by researchers or engineers rather than machine operators, which slows down productivity. Analy- ∗ Corresponding author. Current address: Hitachi America Ltd. R&D Division, Farmington Hills, MI 48335, USA. E-mail addresses: raph.corne@gmail.com (R. Corne), nathc2@asme.org (C. Nath), mohamed.elmansori@ensam.eu (M. El Mansori), kurfess@gatech.edu (T. Kurfess). ses are performed mainly by developing analytical force and wear models or statistical tool life models based on experimental data, such as force, wear, and surface metrology. However, in a produc- tion line, e.g., in machining, there is a serious need of intelligent manufacturing systems that can automatically collect reliable data during machining, monitor tool wear progress based on the col- lected data, anticipate wear and breakage, alert the machinist in real time to avoid unexpected tool failure, and help to obtaining quality products. Drilling is counted to be one-third of all machining operations [5] and usually comes last in the entire process. It is therefore crit- ical for this process to be as precise and optimized as possible. Drill wear monitoring is very important since worn tools affect hole quality (e.g., cylindricity, circularity, surface roughness), and prod- uct life [6]. The problem becomes more critical when machining superalloys, such as, Ni- and Ti-based alloys that possess unique physical and metallurgical properties [7–9]. Due to mainly poor machinability, significant chip welding at the cutting edge, and http://dx.doi.org/10.1016/j.jmsy.2017.01.004 0278-6125/© 2017 Published by Elsevier Ltd on behalf of The Society of Manufacturing Engineers.