*2.1.1. Analysis of cutting force and torque signals in CFRP/CFRP stack drilling*

Advanced methodologies for tool condition monitoring were developed with reference to an extensive experimental campaign of drilling of CFRP/CFRP stacks for aeronautical industry assembly [21–26]. The CFRP/CFRP stacks employed in the tests were made of two superimposed symmetrical and balanced laminates of 5 mm each. The laminates consisted of 26 unidirectional prepreg plies made of Toray T300 carbon fibers and CYCOM 977-2 epoxy matrix, with stacking sequence [±45<sup>2</sup> /0/90<sup>4</sup> /0/90/0<sup>2</sup> ] s and a 0°/90° fiberglass fabric (80 g/m<sup>2</sup> ) on the top and bottom. Following vacuum bag molding and autoclave curing, the laminates displayed an uneven surface finish on the bag side. To test the hardest condition, as requested by the industry, CFRP/ CFRP stack drilling was performed by facing the laminates with their uneven sides [21].

The drilling tests were performed on a CNC drilling center with a 2-flute 6.35 mm diameter with 125° point angle twist drill made of tungsten carbide (WC).

To assess the impact of the cutting parameters on the machinability of the CFRP stack laminates as concern tool wear and hole quality, different cutting conditions were adopted for the experimental drilling tests: feed = 0.11, 0.15, and 0.20 mm/rev; spindle speed = 2700, 6000, and 9000 rpm. For each cutting condition, 60 holes were realized with the same drill bit in order to evaluate the tool wear development.

During the experimental stack drilling tests, a multiple sensor monitoring system was utilized, consisting of a Kistler-9257A piezoelectric dynamometer to acquire the thrust force along the z-direction, Fz, and a Kistler-9277A25 piezoelectric dynamometer to acquire the cutting torque about the z axis, and T. A National Instruments NI USB-6361 DAQ board was employed to digitalize the analogue signals acquired by the force and torque sensors at 10 kS/s.

**Figure 1** shows an example of thrust force signal acquired during the CFRP/CFRP drilling tests. Both the force and torque sensor signals displayed high frequency oscillations as those highlighted in **Figure 2**. This characteristic, not observed when drilling homogeneous and isotropic materials, is associated to the anisotropic nature of the CFRP laminates, which exhibit superior mechanical properties along the fiber directions.

As shown in earlier studies on cutting of FRP composites, the fiber orientation with regard to the cutting direction regulates the mechanism of chip formation and the quality of the cut surface [5]. Based on the angle formed between the cutting edge and the carbon fibers, also known as fiber cutting angle, different cutting modes can be identified [27].

Experimental studies have been performed with the aim to model the relationship between fiber cutting angle and cutting force in drilling of unidirectional CFRP laminates, obtaining the calculation of force coefficients as a sine wave function of 2θ, where θ is the fiber cutting angle [28]. In [27], in order to investigate the effect of local feed force on delamination in drilling of unidirectional CFRP laminates, a model was proposed by resolving cutting forces into steady and transient components, the first associated to the progressive engagement of the cutting edge, and the second associated to the cutting modes. Drilling of multidirectional CFRP laminates is far more complex: different cutting modes take place at the same time along the cutting edge, due to the diverse fiber orientations of the multiple plies simultane-

**Figure 1.** Thrust force sensor signal in CFRP/CFRP stack drilling (6000 rpm–0.15 mm/rev).

Drilling of Fiber-Reinforced Composite Materials for Aeronautical Assembly Processes

http://dx.doi.org/10.5772/intechopen.80466

71

**Figure 2.** Enlarged thrust force signal: high frequency oscillations (6000 rpm–0.15 mm/rev).

ously cut by the drill cutting edge.

In CFRP drilling, the fiber cutting angle changes during a half revolution of the drill, and the interaction mechanism between the tool and the laminate varies accordingly, causing diverse conditions of mechanical loading and resulting surface quality [27]. As a matter of fact, loading is influenced by cutting edge geometry and cutting mode and therefore varies during a drill revolution, as observed in the sensor signals acquired in CFRP drilling, which display high amplitude oscillations.

Drilling of Fiber-Reinforced Composite Materials for Aeronautical Assembly Processes http://dx.doi.org/10.5772/intechopen.80466 71

**Figure 1.** Thrust force sensor signal in CFRP/CFRP stack drilling (6000 rpm–0.15 mm/rev).

from the force and torque sensor signals, it is possible to develop an artificial neural network (ANN)-based cognitive paradigm for pattern recognition with the aim to find correlations

Advanced methodologies for tool condition monitoring were developed with reference to an extensive experimental campaign of drilling of CFRP/CFRP stacks for aeronautical industry assembly [21–26]. The CFRP/CFRP stacks employed in the tests were made of two superimposed symmetrical and balanced laminates of 5 mm each. The laminates consisted of 26 unidirectional prepreg plies made of Toray T300 carbon fibers and CYCOM 977-2 epoxy matrix, with

tom. Following vacuum bag molding and autoclave curing, the laminates displayed an uneven surface finish on the bag side. To test the hardest condition, as requested by the industry, CFRP/ CFRP stack drilling was performed by facing the laminates with their uneven sides [21].

The drilling tests were performed on a CNC drilling center with a 2-flute 6.35 mm diameter

To assess the impact of the cutting parameters on the machinability of the CFRP stack laminates as concern tool wear and hole quality, different cutting conditions were adopted for the experimental drilling tests: feed = 0.11, 0.15, and 0.20 mm/rev; spindle speed = 2700, 6000, and 9000 rpm. For each cutting condition, 60 holes were realized with the same drill bit in order to

During the experimental stack drilling tests, a multiple sensor monitoring system was utilized, consisting of a Kistler-9257A piezoelectric dynamometer to acquire the thrust force along the z-direction, Fz, and a Kistler-9277A25 piezoelectric dynamometer to acquire the cutting torque about the z axis, and T. A National Instruments NI USB-6361 DAQ board was employed to

**Figure 1** shows an example of thrust force signal acquired during the CFRP/CFRP drilling tests. Both the force and torque sensor signals displayed high frequency oscillations as those highlighted in **Figure 2**. This characteristic, not observed when drilling homogeneous and isotropic materials, is associated to the anisotropic nature of the CFRP laminates, which exhibit

As shown in earlier studies on cutting of FRP composites, the fiber orientation with regard to the cutting direction regulates the mechanism of chip formation and the quality of the cut surface [5]. Based on the angle formed between the cutting edge and the carbon fibers, also

In CFRP drilling, the fiber cutting angle changes during a half revolution of the drill, and the interaction mechanism between the tool and the laminate varies accordingly, causing diverse conditions of mechanical loading and resulting surface quality [27]. As a matter of fact, loading is influenced by cutting edge geometry and cutting mode and therefore varies during a drill revolution, as observed in the sensor signals acquired in CFRP drilling, which display

digitalize the analogue signals acquired by the force and torque sensors at 10 kS/s.

known as fiber cutting angle, different cutting modes can be identified [27].

and a 0°/90° fiberglass fabric (80 g/m<sup>2</sup>

) on the top and bot-

between the extracted sensor signal features and tool condition [21, 22, 24].

*2.1.1. Analysis of cutting force and torque signals in CFRP/CFRP stack drilling*

stacking sequence [±45<sup>2</sup>

70 Characterizations of Some Composite Materials

evaluate the tool wear development.

high amplitude oscillations.

/0/90<sup>4</sup>

/0/90/0<sup>2</sup> ] s

with 125° point angle twist drill made of tungsten carbide (WC).

superior mechanical properties along the fiber directions.

**Figure 2.** Enlarged thrust force signal: high frequency oscillations (6000 rpm–0.15 mm/rev).

Experimental studies have been performed with the aim to model the relationship between fiber cutting angle and cutting force in drilling of unidirectional CFRP laminates, obtaining the calculation of force coefficients as a sine wave function of 2θ, where θ is the fiber cutting angle [28]. In [27], in order to investigate the effect of local feed force on delamination in drilling of unidirectional CFRP laminates, a model was proposed by resolving cutting forces into steady and transient components, the first associated to the progressive engagement of the cutting edge, and the second associated to the cutting modes. Drilling of multidirectional CFRP laminates is far more complex: different cutting modes take place at the same time along the cutting edge, due to the diverse fiber orientations of the multiple plies simultaneously cut by the drill cutting edge.

Because of the overlap of different cutting modes in both space and time, the high amplitude oscillations that can be observed in the thrust force and torque signals acquired during drilling of the multidirectional CFRP laminates are the sum of multiple waves having a phase difference dependent on the different fiber orientations (e.g., 0, 45, 90, −45°) and having an amplitude related to the number of plies with same fiber orientations concurrently cut by the drill cutting edge.

*2.1.2. Sensor signal feature extraction and selection in the frequency domain and neural network pattern recognition for tool wear estimation*

With the purpose to explore the complex frequency content of the thrust force and torque sensor signals acquired in the multidirectional CFRP/CFRP stack drilling experimental tests, advanced signal feature extraction in the frequency domain was carried out in [21] with the final objective to find correlations between the extracted frequency domain sensor signal features and the tool wear state. The fast Fourier transform (FFT) algorithm was applied to translate the signals into the frequency domain. Following this transformation, a number of significant peaks were detected at frequencies equal to 1, 2, 3, 4, 5, and 6 times the revolution frequency of the tool. For instance, on the FFT of the thrust force signals acquired in the CFRP/ CFRP stack drilling tests carried out at 6000 rpm and 0.15 mm/rev, in which the revolution frequency is 6000 rpm/60 = 100 Hz, the tallest frequency peaks were found at 100, 200, 300, 400, 500, and 600 Hz, that is, 1x-6x the revolution frequency, as visible in **Figure 3**.

out at 6000 rpm and 0.15 mm/rev, suggesting a potential correlation of the frequency peak

**Figure 4.** Amplitude of the peaks detected in the frequency transform of the thrust force signal vs. hole number

Drilling of Fiber-Reinforced Composite Materials for Aeronautical Assembly Processes

http://dx.doi.org/10.5772/intechopen.80466

73

A statistical feature selection procedure based on the Pearson's correlation coefficient was applied to evaluate the statistical correlation between the extracted features and the output

Substantial correlations with tool wear state were identified for the following set of features: Fpeak2x, Fpeak4x, Fpeak6x, Tpeak2x, and Tpeak4x. The subset composed the first three features extracted from the thrust force signal displayed the highest correlation with tool wear. The selected features were used to build two different feature pattern vectors, FPV<sup>1</sup>

, to feed cognitive paradigms for the estimation of tool wear. FPV<sup>1</sup>

feature pattern vector containing the five features extracted from the thrust force and the

The FPVs were fed to three-layer cascade-forward backpropagation artificial neural networks (ANN) to find correlations between the extracted and selected frequency domain sensor signal features and tool wear state through pattern recognition [29]. ANN training was performed using the Levenberg-Marquardt optimization algorithm. Each FPV was associated to its corresponding flank wear (VB) to create input-output vectors for ANN learning. For each drilling condition, 60 input-output vectors (i.e., one for each drilled hole) were built to create the related ANN learning set. Training and testing were performed with different ANN configurations by varying the number of hidden layer nodes. In particular, the number of hidden

Cross validation of the NN was carried out by means of the leave-k-out method with *k* = 1. The global pattern recognition performance was evaluated by merging the recognition rates

and 5-10-15 for FPV<sup>2</sup>

of three features extracted from the thrust force signal, and FPV<sup>2</sup>

layer nodes was varied between 3, 6, and 9 for FPV<sup>1</sup>

1, 2, and 3 times the number of input features.

achieved across all trials.

was made

was a sensor fusion

, that is, equal to the

amplitude with tool wear progression.

tool flank wear, VB.

(6000 rpm–0.15 mm/rev) [21].

and FPV<sup>2</sup>

torque signals.

As regards the torque signals, the tallest frequency peaks were found at 1x-4x the revolution frequency (100, 200, 300, and 400 Hz).

This seemed to confirm the tight connection between the peaks of the signal frequency transform and the effect of the cutting angle variation during the drilling process due to the several fiber orientations in the multidirectional laminates.

The amplitudes of the identified force and torque frequency peaks with increasing number of holes were investigated, showing a notable growth of some of the peaks, as shown in **Figure 4** with reference to the thrust force signals of the experimental drilling tests carried

**Figure 3.** Single-sided amplitude spectrum of thrust force signal (6000 rpm–0.15 mm/rev) [21].

Drilling of Fiber-Reinforced Composite Materials for Aeronautical Assembly Processes http://dx.doi.org/10.5772/intechopen.80466 73

Because of the overlap of different cutting modes in both space and time, the high amplitude oscillations that can be observed in the thrust force and torque signals acquired during drilling of the multidirectional CFRP laminates are the sum of multiple waves having a phase difference dependent on the different fiber orientations (e.g., 0, 45, 90, −45°) and having an amplitude related to the number of plies with same fiber orientations concurrently cut by the

With the purpose to explore the complex frequency content of the thrust force and torque sensor signals acquired in the multidirectional CFRP/CFRP stack drilling experimental tests, advanced signal feature extraction in the frequency domain was carried out in [21] with the final objective to find correlations between the extracted frequency domain sensor signal features and the tool wear state. The fast Fourier transform (FFT) algorithm was applied to translate the signals into the frequency domain. Following this transformation, a number of significant peaks were detected at frequencies equal to 1, 2, 3, 4, 5, and 6 times the revolution frequency of the tool. For instance, on the FFT of the thrust force signals acquired in the CFRP/ CFRP stack drilling tests carried out at 6000 rpm and 0.15 mm/rev, in which the revolution frequency is 6000 rpm/60 = 100 Hz, the tallest frequency peaks were found at 100, 200, 300,

*2.1.2. Sensor signal feature extraction and selection in the frequency domain and neural* 

400, 500, and 600 Hz, that is, 1x-6x the revolution frequency, as visible in **Figure 3**.

**Figure 3.** Single-sided amplitude spectrum of thrust force signal (6000 rpm–0.15 mm/rev) [21].

As regards the torque signals, the tallest frequency peaks were found at 1x-4x the revolution

This seemed to confirm the tight connection between the peaks of the signal frequency transform and the effect of the cutting angle variation during the drilling process due to the several

The amplitudes of the identified force and torque frequency peaks with increasing number of holes were investigated, showing a notable growth of some of the peaks, as shown in **Figure 4** with reference to the thrust force signals of the experimental drilling tests carried

drill cutting edge.

72 Characterizations of Some Composite Materials

*network pattern recognition for tool wear estimation*

frequency (100, 200, 300, and 400 Hz).

fiber orientations in the multidirectional laminates.

**Figure 4.** Amplitude of the peaks detected in the frequency transform of the thrust force signal vs. hole number (6000 rpm–0.15 mm/rev) [21].

out at 6000 rpm and 0.15 mm/rev, suggesting a potential correlation of the frequency peak amplitude with tool wear progression.

A statistical feature selection procedure based on the Pearson's correlation coefficient was applied to evaluate the statistical correlation between the extracted features and the output tool flank wear, VB.

Substantial correlations with tool wear state were identified for the following set of features: Fpeak2x, Fpeak4x, Fpeak6x, Tpeak2x, and Tpeak4x. The subset composed the first three features extracted from the thrust force signal displayed the highest correlation with tool wear. The selected features were used to build two different feature pattern vectors, FPV<sup>1</sup> and FPV<sup>2</sup> , to feed cognitive paradigms for the estimation of tool wear. FPV<sup>1</sup> was made of three features extracted from the thrust force signal, and FPV<sup>2</sup> was a sensor fusion feature pattern vector containing the five features extracted from the thrust force and the torque signals.

The FPVs were fed to three-layer cascade-forward backpropagation artificial neural networks (ANN) to find correlations between the extracted and selected frequency domain sensor signal features and tool wear state through pattern recognition [29]. ANN training was performed using the Levenberg-Marquardt optimization algorithm. Each FPV was associated to its corresponding flank wear (VB) to create input-output vectors for ANN learning. For each drilling condition, 60 input-output vectors (i.e., one for each drilled hole) were built to create the related ANN learning set. Training and testing were performed with different ANN configurations by varying the number of hidden layer nodes. In particular, the number of hidden layer nodes was varied between 3, 6, and 9 for FPV<sup>1</sup> and 5-10-15 for FPV<sup>2</sup> , that is, equal to the 1, 2, and 3 times the number of input features.

Cross validation of the NN was carried out by means of the leave-k-out method with *k* = 1. The global pattern recognition performance was evaluated by merging the recognition rates achieved across all trials.

The ANN performance was evaluated in terms of mean square error (MSE), that is, the variance of the differences between the VB values predicted by the ANN and the target VB values. The very low MSE values, between 9.82E-07 and 1.57E-05, confirmed the robust correlation between the sensor signal features extracted in the frequency domain and the tool wear, due to the interaction between the drill bit and the anisotropic CFRP material.

Through pattern recognition based on the frequency domain signal features extracted from the multiple sensor monitoring signals, the ANN accurately reconstructed the tool wear curve. **Figure 5** presents the reconstructed tool wear curve for the drilling test performed at 6000 rpm and 0.11 mm/rev, obtained by applying the leave-k-out method with *k* = 1 to the ANN with three hidden layer nodes trained with the three features extracted from the thrust force signal (MSE = 1.57E-05). **Figure 6** presents the reconstructed tool wear curve for the same drilling test obtained by applying the leave-k-out method with *k* = 1 to the ANN with 5 hidden layer nodes trained with thefive features extracted from the thrust force and the torque signal (MSE= 3.79E-06).

For all drilling conditions, the sensor fusion pattern vector FPV<sup>2</sup> , including frequency domain features coming from both thrust force and torque signals, provided better results than FPV<sup>1</sup> .

> and an innovative geometry step drill bit. The traditional tool was a 2-flute 6.35-mm diameter twist drill with 125° point angle and 30° helix angle tungsten carbide (**Figure 7a**), while the innovative step drill bit was a 2-flute 6.35-mm diameter drill bit with 120° point angle and 20° helix angle made of tungsten carbide (WC) (**Figure 7b**). Different cutting parameters were used for the experimental drilling tests: three feed values (0.11, 0.15, and 0.20 mm/rev) and

> **Figure 6.** Reconstructed tool wear curve obtained by the ANN trained with the five features extracted from the thrust

Drilling of Fiber-Reinforced Composite Materials for Aeronautical Assembly Processes

http://dx.doi.org/10.5772/intechopen.80466

75

A statistical approach in the time domain was applied for feature extraction from the thrust force and torque sensor signals acquired during the drilling tests. The following five statistical signal features were extracted: arithmetic mean, variance, skewness, kurtosis, and signal

assess its correlation with tool wear (0 < *r<sup>s</sup>* < 0.3, weak correlation, 0.3 < *r<sup>s</sup>* < 0.7, moderate cor-

strong correlation with tool wear: thrust force average (Fz\_av), torque average (Tav), thrust force

Every 10 holes, a magnified picture of the tool was acquired through an optical measuring machine (Tesa Visio V-200) to measure the flank wear [21]. A 3rd order polynomial interpolation of the VB values was applied to reconstruct the tool wear curves for each drill bit under

The graphical analysis of the selected sensor signal features, with particular reference to the thrust force average (**Figures 10** and **11**), which is the most correlated feature, shows that the behavior of the features and the behavior of tool wear (**Figure 8** and **9**) display the same increasing trend with increasing number of holes. The average thrust force is higher for the

The selected statistical features were employed to construct, for each drilling condition, 60 sensor fusion feature pattern vectors (SFPVs), that is, one for each drilled hole. Each SFPV was built by combining the selected statistical signal features and the corresponding hole number, *n*. Each SFPV was associated to its matching flank wear value (VB) to create input-output

, was calculated to

coefficient, five features showed a

energy. For each extracted feature, the Spearman correlation coefficient, *r<sup>s</sup>*

variance (Fz\_var), torque variance (Tvar), and thrust force kurtosis (Fz\_kurt).

relation, and 0.7 < *r<sup>s</sup>* < 1, strong correlation). Based on the *r<sup>s</sup>*

three spindle speeds (2700, 6000, and 7500 rpm).

force and the torque signal (6000 rpm–0.11 mm/rev).

the tested drilling conditions (**Figures 8** and **9**).

innovative drill bits.

The accurate tool wear curve reconstruction achieved by the ANN can be effectively utilized to determine the end of useful tool life through on-line diagnosis during CFRP/CFRP stack drilling, identifying the transition of the tool flank wear between the second and the third phase of the wear curve.
