J Intell Manuf (2012) 23:797–809 DOI 10.1007/s10845-010-0436-x Tool wear estimation using an analytic fuzzy classifier and support vector machines Danko Brezak · Dubravko Majetic · Toma Udiljak · Josip Kasac Received: 10 October 2008 / Accepted: 10 July 2010 / Published online: 23 July 2010 © Springer Science+Business Media, LLC 2010 Abstract A new type of continuous hybrid tool wear esti- mator is proposed in this paper. It is structured in the form of two modules for classification and estimation. The classifica- tion module is designed by using an analytic fuzzy logic con- cept without a rule base. Thereby, it is possible to utilize fuzzy logic decision-making without any constraints in the number of tool wear features in order to enhance the module robust- ness and accuracy. The final estimated tool wear parameter value is obtained from the estimation module. It is structured by using a support vector machine nonlinear regression algo- rithm. The proposed estimator implies the usage of a larger number and various types of features, which is in line with the concept of a closer integration between machine tools and different types of sensors for tool condition monitoring. Keywords Tool wear estimation · Fuzzy logic · Support vector machines · Dynamic feature selection Introduction The development of production systems toward higher levels of automation and flexibility is an on-going process stimu- lated by a requirement for a continuous improvement in the quality of new products while simultaneously maintaining a high productivity level. The trend is especially marked in the field of machine tools which constantly undergo differ- D. Brezak (B ) · D. Majetic · J. Kasac Department of Robotics and Production Systems Automation, Faculty of Mechanical Engineering and Naval Architecture, University of Zagreb, Zagreb, Croatia e-mail: danko.brezak@fsb.hr T. Udiljak Department of Technology, Faculty of Mechanical Engineering and Naval Architecture, University of Zagreb, Zagreb, Croatia ent design and control system modifications. In that sense, the development of monitoring systems capable of identify- ing the machining process dynamics and the condition of all machine modules in the real time has became one of the most important imperatives. The primary segment of the over- all monitoring process is tool condition monitoring (TCM) since tool wear is the main generator of random process dis- turbances with a direct influence on the safety, quality and productivity of the machining process. Additionally, a con- tinuous estimation of chosen wear parameters is also essen- tial for the realization of tool wear regulation which would improve tool efficiency, i.e. extend tool life or increase pro- ductivity in the scheduled tool change period in the mass production environment (Landers et al. 2002; Liang et al. 2004). In that sense, TCM systems for a continuous estima- tion of wear parameters is expected to be utilized mainly in situations where the variability of process parameters is rel- atively low and the influence of tool efficiency maximization on the overall productivity is big. Research efforts in the field of TCM systems have inten- sified in past years. They have resulted in a number of various solutions usually based on different types of compu- tational intelligence algorithms. The ones most commonly used are artificial neural networks (ANN) due to their fea- tures, such as abilities of identification of complex systems and processes, parallel data processing, noise suppression characteristics, and adaptability to varying machining con- ditions and tool wear dynamics (Wang et al. 2001). Among a number of neural network types, the most frequently used are Multilayer Perceptron Neural Network—MLP NN (Sick 2002), commonly trained by the Error-Back Propagation algorithm (Huang & Chen 2000; Tandon & El-Mounayri 2001; Chen & Chen 2004; Ghosh et al. 2007; Alonso & Salgado 2008). Besides them, TCM systems built on Radial Basis Function NN (Srinivasa et al. 2002), Adaptive 123