Estimate of cutting forces and surface roughness in end milling of glass fiber reinforced plastic composites using fuzzy logic system

1 Introduction

Fiber reinforced plastic (FRP) composite materials, which are produced by different methods, are widely used in aircraft, aerospace and other engineering fields, due to their excellent mechanical properties [1]. These mechanical properties are light weight, high specific strength, good corrosion resistance, low cost, low thermal conductivity, etc. [2, 3]. However, the machinability of FRP composites is fairly difficult, because of nonhomogeneous and anisotropic properties [4, 5].

Especially, drilling and milling as secondary operations in FRP composite materials are needed to assemble the structures [5, 6]. The most frequently used method in manufacturing is milling, which is the machining operation of materials. Some problems occur with materials and tools during milling. These problems are tool wear, surface roughness, and delamination [7, 8]. One of the major problems is surface roughness due to the heterogeneity of fiber and matrix [9]. Surface roughness has a great effect on dimensional precision, performance of mechanical items, and manufacturing costs [5]. It is the effect of the cutting parameters, tool properties, and structure of composite materials [10, 11].

While numerous studies have been conducted with drilling of glass fiber reinforced plastic (GFRP) composite materials, only a few studies have been conducted on cutting forces and surface roughness in milling of GFRP composite materials. Azmi et al. [8] performed an experimental study on end milling of GFRP composites, using uncoated tungsten carbide tools. They investigated the machinability of GFRP composites in terms of tool wear, tool useful life, machining quality, and machining forces. As a result, they determined that when feed rates and cutting speeds were increased, tool life decreased. Variations of machining forces during end milling of GFRP composites were attributed to the growth of tool wear and tool/fiber contact during machining. In another study, Azmi et al. [9] carried out the GFRP composites end milling machinability with regard to surface roughness, life span of tool, and the forces present in machining. The experiments were performed under varying experimental parameters and their levels were found out in accordance with the Taguchi design experimental method. They noticed that the most important parameter affecting surface roughness, tool life, and machining forces was the feed rate. Palanikumar [10] developed the fuzzy logic for modeling machining parameters in machining GFRP by means of poly-crystalline diamond tools. While input parameters were selected as cutting speed, feed, and depth of cut for fuzzy logic, output parameters were selected for surface roughness via criteria including, for example, arithmetic average height (Ra) and maximum profile height (Rt). Consequently, he determined that a rise in cutting speed led to a decrease in the surface roughness; however, an increase in feed rate raised the surface roughness. He also notified that feed rate was the basic parameter affecting the surface roughness parameters that followed cutting speed, while the depth of cut was the least effective parameter influencing the parameters of surface roughness (Ra and Rt) in GFRP composites machining. Erkan et al. [12] milled GFRP composites experimentally to minimize the damage on the machined surfaces. They also developed the artificial neural network to predict the damage factor. As a result, they demonstrated that the damage factor dropped with an increase in cutting speed and feed rate; nevertheless, it concluded that the damage factor became smaller with increases in cut depth and flutes number. A similar study was performed by Razfar and Zadeh [13]. However, they investigated the surface roughness using a neural network and a genetic algorithm. Raj and Perumal [14] investigated the delamination factor of GFRP composites and surface roughness that were in use in the Ti-Namite carbide K10 end mill, solid carbide K10 end mill, and Tipper carbide K10 end mill.

When the experimental study performed was examined, all parameters seemed to be counted as individual. Therefore, an experimental study was conducted on the effects on thrust forces and surface roughness in milling glass fiber reinforced (GFRP) composite materials. In the experimental study, parameters such as feed rate, cutting speed, and tool geometry were taken into account. The results obtained from the experimental study were implemented to a fuzzy logic model. Thus, fuzzy logic was used to predict the surface roughness and thrust forces for various cutting parameters of GFRP composites without experiment.

2 Experimental study

In the experimental study, GFRP composite materials produced by hand layup methods were used. This material, provided by Izoreel firm (zmir, Turkey), consisted of a density of 1.9 g/cm³, angle among fibers of 90°, mat shapes and 18 layers. The module of elasticity was 20 N/mm², the tensile strength was 300 N/mm², the bend strength was 400 N/mm², and the pressure strength was 500 N/mm².

A Hummer VMC-1000 CNC milling machine was used for the milling. This machine has 15 kW of engine power and 8000 rpm of maximum speed. In the experimental study, a 10 mm diameter cemented carbide end mill, which had different numbers of flutes (2, 3, and 4), was used. In addition to this, during milling, the machining forces were evaluated by means of a Kistler 9257B dynamometer and its Kistler 8 channel amplifier.

The 100×100×10 mm sized specimen is fixed over dynamometers in the Hummer VMC-1000 CNC machine (Figure 1). Experiments are performed at different cutting parameters. However, in the experiments, the cutting depth and machining direction are taken as 2 mm and X, respectively. At each test, the milling was changed to avoid tool wear due to the absence of cutting fluids. The cutting parameters are given in Table 1.

Figure 1

Experimental setup.

Considering the processing quality and tolerance, surface roughness is the most important feature of machinability materials. Therefore, the surface roughness was measured at different cutting parameters. Taylor Hobson’s Surtonic 3+ surface roughness device was used for measurements. The measurement sampling length was chosen as 1.6 mm. Measurement processes were performed parallel to the axis channel. Three surface roughness values of machined surfaces were measured; the average surface roughness was found by taking their averages.

2.1 Experimental results

With the intention of achieving the best performance milling process, chip removing and cutting parameters, such as cutting speed and feed rate, should be chosen carefully. For this purpose, a number of experiments were performed to explore the effect on thrust forces and surface roughness using different numbers of flutes, cutting speed, and feed rate. Because of the tiny chips formed during milling, however, they were not investigated in this paper. By contrast, in evaluating the surface roughness, the surface roughness and thrust forces values of cutting tools were obtained depending on cutting parameters made in comparison with each other. Milling forces in x, y, z directions of each cutting parameter were recorded on a computer during the experimental study. Milling forces obtained from three flutes, 30 m/min cutting speed and 0.15 mm/rev feed rate are given in Figure 2. Thrust forces were calculated depending on milling forces in x, y, z directions. Thrust forces are given in Eq. (1):

(1)F=Fx2+Fy2+Fz2 (1)

Figure 2

Cutting forces for three flutes, 30 m/min and 0.15 mm/rev.

3 Fuzzy logic system

3.1 The structure of the fuzzy logic

Fuzzy logic is a method used to model the actual, ambiguous, and indefinite data which frequently occur. With this modeling, relationships between input and output membership functions are established. Membership functions depend on the philosophy of the neighborhood of numbers. If a situation is represented by a number in the decision process, the acceptability of the situation in question will be provided in that number’s realization. However, the numbers which are closer to the number in question will not be perceived as a part of the decision process. In fact, in a certain confidence coefficient, it will be statistically wrong to suggest that these numbers are members of different populations. In this case, it is possible to talk about the neighborhood numbers serving the same basic aim. If A is an element of the set in question in R∈(-∞, +∞), μ_A(x) membership function occurs in R→[0, 1] interval. In other words, if the set A is A=[a₁, a₃] interval, the membership function μ_A(x) can be generally shown with the following expression:

(2)μA(x)={0,x<a11,a1≤x≤a30x>a3 (2)

μ_A(x) triangular membership function is defined with the following statement:

(3)μA(x)={0,      x<a1x-a1a2-a1, a1≤x≤a2a3-xa3-a2,   a2≤x≤a3 0,     x>a3 (3)

According to Eq. (3), the set must be A=(a₁, a₂, a₃). Here a₂ can be defined as a normal value membership. Fuzzy logic at this point assumes that the values closer to a₂, depending on an α coefficient, will be represented with the meaning attributed to this value. In other words, the uncertainty in a₂ can be tolerated with an α coefficient, that can be assumed or found out according to distribution. The neighborhood in question is shown in Figure 3. The α value is called as cutting coefficient in the fuzzy logic terminology. The a1α and a3α numbers are the lower and upper values of the intervals forming the neighborhood of the a₂ normal value.

Figure 3

The neighborhood of numbers.

In other words, all numbers in the a1α and a3α interval have the same meaning with the a₂ normal value. a1α and a3α values are calculated with the following formulas:

(4)a1α-a1a2-a1=α (4)

(5)a3-a3αa3-a2=α (5)

From these formulas, Aα=[a1α, a3α] interval for ∀α∈[0, 1] can be created. a1α and a3α values are shown with the following formulas:

(6)a1α=α(a2-a1)+a1 (6)

(7)a3α=a3-(a3-a2)α (7)

These topics related to fuzzy logic and systems are discussed by Zadeh [15] and Mendel [16] in detail.

3.2 Application of the fuzzy logic model to GFRP composites

Modeling with fuzzy logic of GFRP composites was created by the input and output parameters. While input parameters were selected as cutting speed, feed rate, and number of flutes, output parameters were selected as surface roughness and thrust force. Structure of the fuzzy logic model depending on input and output parameters is given in Figure 4.

Figure 4

Fuzzy logic model structure.

After the structure of fuzzy logic model was created, the membership function of the input and output linguistic parameters were determined [17]. The membership function of the input parameters were selected as low, medium, and high in different number ranges. The different number ranges input parameters are given in Figure 5. The membership function of the output parameters were selected as lowest, lower, low, medium, high, higher, and highest (Figure 6).

Figure 5

The membership functions and the number range of input parameters.

Figure 6

The membership functions and the number range of output parameters.

The results depending on the experimental study were described in the fuzzy logic as rules. Some of these rules are given in Table 2.

Table 2

Fuzzy logic rules in linguistic form.

These rules were produced for predicting values of fuzzy logic for surface roughness and thrust force on GFRP composite materials.

3.3 Fuzzy logic results

The structure and membership functions of the fuzzy logic model and its rules are the most important to obtain from values close to real results. If these parameters are determined well, the fuzzy logic model will estimate the values close to real results. For estimation of fuzzy logic, an example is given in Figure 7.

Figure 7

The predicted result of fuzzy logic.

In Figure 7, when input parameters were determined as cutting speed 60 m/min, feed rate 0.15 mm/rev, and number of flutes three, output parameters were found by fuzzy logic as surface roughness 2.95 μm and thrust force 65 N. Different examples were made.

The forces and surface roughness values obtained from both the fuzzy logic model and experimental study, depending on cutting parameters, are given in Table 3.

Table 3

Cutting parameters with the values both obtained from the experimental study and the predicted fuzzy logic model.

Cutting parameters				The values obtained from experimental study					The predicted values of fuzzy logic
EN	FN	CS (m/min)	FR (mm/rev)	Fx (N)	Fy (N)	Fz (N)	F (N)	SR (μm)	F (N)	SR (μm)
1	2	30	0.1	146.25	41.35	-12.74	152.52	4.17	145	4.45
2	2	30	0.15	179.18	57.6	-18.78	189.15	4.52	185	4.45
3	2	30	0.2	191.74	69.34	-26.8	205.65	4.98	225	4.95
4	2	60	0.1	159.19	44.27	-12.56	165.71	3.45	185	3.45
5	2	60	0.15	187.5	59.9	-20.33	197.88	3.89	185	3.95
6	2	60	0.2	205.3	72.51	-28.5	219.55	4.18	225	4.45
7	2	90	0.1	194.2	50.16	-22.12	201.8	2.9	225	2.95
8	2	90	0.15	215.6	61.6	-24.1	225.52	3.29	225	3.45
9	2	90	0.2	254.61	75.15	-27.79	266.92	3.69	270	3.95
10	3	30	0.1	48.89	30.97	-9.051	58.58	3.04	65	2.95
11	3	30	0.15	59.11	41.02	-13.63	73.23	3.36	65	3.45
12	3	30	0.2	66.82	49.51	-18.43	85.18	3.84	105	3.95
13	3	60	0.1	53.86	27.34	-8.68	61.02	2.36	65	2.45
14	3	60	0.15	62.5	42.13	-13.81	76.63	2.96	65	2.95
15	3	60	0.2	71.92	52.44	-19.27	91.07	3.22	105	3.45
16	3	90	0.1	70.02	33.79	-14.76	79.14	2.28	65	2.45
17	3	90	0.15	71.73	39.1	-14.69	83.01	2.58	105	2.45
18	3	90	0.2	78.95	47.21	-19.14	93.96	3.1	105	3.45
19	4	30	0.1	15.58	19.39	-6.153	25.62	2.06	25	1.95
20	4	30	0.15	19.22	28.78	-11.1	36.34	2.24	25	2.45
21	4	30	0.2	17.59	30.55	-10.94	36.91	2.4	25	2.45
22	4	60	0.1	15.95	19.26	-6.424	25.82	1.96	25	1.95
23	4	60	0.15	20.22	28.53	-9.979	36.36	2.1	25	2.45
24	4	60	0.2	20.71	37.09	-14.05	44.74	2.24	65	2.45
25	4	90	0.1	19.53	20.35	-6.548	28.96	1.8	25	1.95
26	4	90	0.15	24.24	29.74	-8.988	39.41	1.88	25	1.95
27	4	90	0.2	27.5	36.41	-13.75	47.66	1.92	65	1.95

CS, cutting speed; EN, experiment number; F, thrust force; FN, flutes number; FR, feed rate; SR, surface roughness.

4 Results and discussion

With the intention of obtaining the best performance in the milling process, the number of flutes end mill and cutting parameters, like cutting speed and feed rate, must be selected carefully for chip disposal. For this purpose, a number of experiments were performed to determine the effect on thrust forces and surface roughness of cutting speed and feed rate, using various numbers of flutes end mill. The data obtained from the experiments were implemented in the fuzzy logic model. The thrust forces and the surface roughness for both experiments and the fuzzy logic model agreed with each other. The effects on both thrust forces and surface roughness, of feed rates, are given in Figure 8 for 60 m/min cutting speed and three number of flutes.

Figure 8

The effect on thrust forces and surface roughness of feed rates.

As the feed rate increased, thrust forces and surface roughness increased in both the experimental study and the fuzzy logic model. While minimum thrust force error between experimental studies with the fuzzy logic model was found to be about 0.065–0.1% at feed rate, maximum thrust force error was found to be about 0.15–0.2% at feed rate. The values of both experimental and the fuzzy logic model for the surface roughness were found to be similar. The fuzzy logic model was found to be very close to the experimental study for 0.15 feed rates. The error was determined as 0.0034%. The higher the feed rate, the more chips were removed by the cemented carbide tool. It was thought that this process was the cause of increasing both thrust force and surface roughness. Increasing the feed rate also caused deterioration of cutting tools [8].

The effects on both thrust forces and surface roughness of cutting speeds are given in Figure 9 for 0.15 mm/rev feed rate and three numbers of flutes. As the cutting speed was increased, thrust forces increased, but surface roughness decreased.

Figure 9

The effect on thrust forces and surface roughness of cutting speeds.

The minimum thrust force was obtained at 30 m/min of cutting speed. While the thrust force value obtained from the experimental study was found to be 73.23 (N), the estimated thrust force value of fuzzy logic was calculated to be 65 (N). For the surface roughness, values at 60 (m/min) cutting speed for the experimental study and the fuzzy logic model were found to be very close.

When the cutting speed increased, the cutting tool flutes were faster in contact with the work-piece. It was thought that the interaction between the cutting tool and the work-piece was to cause the increasing of thrust force due to a reaction consideration. Increase in cutting speed was also the cause of cutting tool wear [8]. By contrast, as the cutting speed was increased, more heat occurred. This caused softening of the work-piece material. Therefore, it was thought that when the cutting speed was increased, the surface roughness improved.

The effects on both thrust forces and surface roughness of number of flutes are given in Figure 10 for 0.15 mm/rev feed rate and 60 (m/min) cutting speed.

Figure 10

The effect on thrust forces and surface roughness of number of flutes.

When the number of flutes was increased, both thrust forces and surface roughness decreased. For both the experimental study and the fuzzy logic model, the thrust forces were found parallel to all flutes. For the surface roughness, however, three flutes were found to be much closer with the experimental study and the fuzzy logic model, as seen in Figure 10. When the number of flutes was increased, the chips removed from the work-piece per flute became fewer. Therefore, as the number of flutes was increased, thrust force and surface roughness automatically reduced.

As a result, for the both experimental study and the fuzzy logic model, the maximum thrust force was approximately worked out as 265 N, when the number of flutes was two, the cutting speed was 90 m/min, and the feed rate was 0.2 mm/rev. The minimum thrust force was approximately worked out as 25 N, when the number of flutes was four, the cutting speed was 30 m/min and the feed rate was 0.1 mm/rev. By contrast, maximum surface roughness was approximately worked out as 4.95 μm, when the number of flutes was two, the cutting speed was 30 m/min, and the feed rate was 0.2 mm/rev. Minimum surface roughness was approximately worked out as 1.85 μm, when the number of flutes was four, the cutting speed was 90 m/min, and the feed rate was 0.1 mm/rev.

5 Conclusions

In the studies carried out, the following results were obtained:

1. As the feed rate was increased, thrust forces, and surface roughness increased for both the experimental study and the fuzzy logic model.

2. When the cutting speed increased, thrust forces increased, and surface roughness decreased.

3. When the number of flutes was increased, both thrust forces and surface roughness decreased.

4. The most important factor effecting thrust force and surface roughness was found to be the number of flutes. Later, it was affected by feed rate.

It is concluded that to obtain a good surface and ideal cutting parameters in the milling of GFRP composite materials, a high cutting speed and number of flutes, and a low feed rate are needed.