Many researchers considered looking at tool wear as a related to the cutting parameters such as cutting speed, feed rate and depth of cut. Some of these then go on looking at cutting forces for predicting tool wear. They analysed the cutting force signals recorded by a dynamometer and characterized the tool wear by time domain and frequency domain via Neural Networks (NN) and Fast Fourier Transform (FFT).
Lin and Lin (1996), Dong et al. (2006) claimed that cutting forces in machining
operations are actually related to tool wear and can be used in estimating the tool wear. In addition, they deduced that the results from the neural networks were found to be quite similar to the experimental results and the proposed model was more accurate in predicting flank wear.
Sarhan and El-Zahry (2011) used the FFT analysis. They found that the magnitudes
of the cutting force and surface roughness changed with the flank wear at different rates.
Čuš and Župerl (2011) engineered a system for monitoring tool condition in real time
based on an NN and Adaptive Neuro-Fuzzy Inference System. They conclude that the (ANFIS) system proposed model could predict flank wear for different cutting conditions with high accuracy.
Other researchers have engineered a statistical approach to analyse and identify the most significant cutting force signals recorded by a dynamometer for the tool wear monitoring system.
In Choudhury and Rath (2000), a series of experiments were conducted according to the design of experiment Analysis of variance (ANOVA). The proposed model represented the relationship between flank wear, tangential cutting force coefficient and the cutting parameters.
A tool wear monitoring strategy for end milling operations when cutting steel with High-Speed Steel (HSS) tool has been presented (Sarhan et al. 2001). The cutting force signals were obtained using a sensitive strain gauge dynamometer. Signals were fed into
13
a FORTRAN program to plot responses in both the time and frequency domains. The results indicated that the cutting forces were more sensitive to the change of flank wear and increased significantly when the tool wears.
Three in-process tool wear monitoring systems based on cutting force were developed and tested (Chen 2003). These systems are using multiple linear regression, artificial neural networks, and a statistics assisted fuzzy-nets based in-process tool wear prediction system. This study demonstrated that the average peak cutting forces in the Y direction (the direction that is perpendicular to the table feed) is the most efficient cutting force representation for tool wear monitoring.
Nouri et al. (2015) employed a statistical method based upon a Cumulative Sum
(CUSUM) control chart to detect the transition of the force model coefficient. They summarised that this approach could be used in real-time to monitor the wear of the tool and identify the transition point from the gradual wear region to the failure region.
A combination of signal processing techniques for estimation of tool wear in real time based on cutting force signals has been presented (Bhattacharyya et al. 2007). Discrete Wavelet Transform, Time Domain Average and Linear Filtering were adopted for extracting relevant features from cutting force signals when milling C-60 mild steel with a single cutting tool insert in face milling operations. Tests producing four different datasets were carried out covering a wide range of machining parameters. They summarised that the proposed model gives satisfactory prediction results by both laboratory and industrial implementations. It is important to note that in that technique, one insert tool was used, which simplified the algorithms, and the tool wear relationships established may not be applicable to multi-toothed cutters.
An approach for fault detection and a diagnosis based on an observer model of an uncertain linear systems was developed (Huang et al. 2007). They designed a model by using the observed variables and cutting force. Four sets of cutting tests were conducted under different working conditions defined by controlled variations to cutting speed, depth of cut, and feed rate. The results indicated that this approach could
14
be used for the detection of failures arising from sensor or actuator functions. However, it is hard to detect tool wear in industry.
The tool wear during end milling of AISI-D2 Steel was monitored using the resultant cutting force (FR) measured by a dynamometer (Chandgude and Sadaiah 2014). A series of experiments were conducted with TiAlN-coated flat solid carbide tools to determine the relationship between flank wear and cutting force as well as the cutting parameters such as cutting speed, feed rate, axial and radial depth of cut. They considered that the cutting force measurement method was the best method for in- process condition monitoring and observed that the resultant force increased as the tool wear progressed consistently.
Although many investigators agreed that cutting force is a reliable and sensitive method to estimate or detect tool wear. There continues to be disagreement on which force component is the most sensitive. Moreover, in some cases, it is hard to separate between cutting force increment due to other disturbances from those resulting from tool wear. For example, a sudden change from hard spots or inclusions in the workpiece material or unexpected changes of the depth of cut. Therefore, sensors for the spindle motor current and power signals are free from such limitations and have the potential of being effective indicators for indirect cutting force measurement.