Keywords

1 Introduction

The performance of cutting tools is directly related to the macro and micro geometry of the tool, cutting materials and coating systems. Recent studies show that the micro geometry of the tool has a very important influence on the cutting process [1, 2]. Furthermore, extending the lifetime of the cutting edge is possible with a specific shape of the cutting edge [3,4,5,6]. In this sense, achieving an effective geometry is the main concern of researchers and the manufacturing industry.

There are several cutting-edge preparation processes that have evolved with the emergence of new technologies, each of which has its advantages and limitations [2]. One process that has become relevant and very useful in recent years is the “Dragging” or “Drag Finishing” abrasive preparation process [7, 8]. Its low cost and versatility to work with the grinding of complex geometries are advantages that make it relevant and useful [9]. However, in the specialized literature there is little information on the working methodology, its limitations, its precision, and its influential parameters in obtaining the cutting microgeometry. Consequently, the need to find the process parameters to find the sharpest cutting-edge sharpness is vital.

Measurement and test results are always subject to a certain uncertainty. Traditionally, accuracy, linearity and stability were characteristics considered in the evaluation of measurement systems. However, it is now recognized that important properties such as repeatability and reproducibility R&R need to be included in the measurement system as an indicator of a reliable process [10, 11]. In this sense, their usefulness is relevant for DF process evaluation [11, 12].

Due to the rise of Industry 4.0, the inclusion of Machine Learning for process monitoring is a field of high importance and demand in the industry. Artificial neural networks are a type of artificial intelligence that allow the prediction of variables. Authors such as [13, 14] used the artificial neural network model to study and predict tool wear from machine working parameters and observed the effectiveness of the algorithms. Therefore, seeing similarity in the investigated object of study, the neural network could predict the evolution of cutting-edge preparation.

This work describes the use of the DF cutting edge preparation process and all its characteristics to allow a better understanding of the process, to verify its accuracy, reproducibility, limitations, and influential parameters applied to carbide tools with the abrasive material Alumina (Al2O3) and Silicon Carbide (SiC) in two grain sizes.

2 Methodology

The An OTEC-DF drag finishing machine was used with abrasive grains of Silicon Carbide (SiC) and Alumina (Al2O3) in two different sizes 24 and 46 according to the sieve aperture given by ASTM D E 11-70 (Fig. 1a and 1b). The tool tested (Fig. 2), was a triangular insert for turning operation without K10-Co7% Tungsten Carbide K10-Co7% coating. The experiment consisted of varying three drag depth (ID) levels, three abrasive mixing ratios (RM) with two grit sizes (GZ) and drag duration times (DT) to achieve expected edge radii ranges (Fig. 1). The variants are shown in Table 1. Subsequently, the ER was measured using a TMAlicona InfiniteFocusG5 edge profile measuring device with a 10X lens as well as the surface roughness with a 20X lens.

Fig. 1.
figure 1

Experimental scheme and description of parameters.

Full size image

The R&R analysis was performed by measuring 24 randomly chosen parts with 3 technicians with 3 repetitions each giving a total of 72 measurements each. Finally, the prediction of the edge radius artificial neural network ANN. The experimental data set used for the training stage was divided by cross-validation into 80% for training and 20% for validation. It was developed with a feed-forward and backpropagation neural scheme on the total experimental data.

Table 1. Variation of factors in experimental design
Full size table

3 Results

Figure 2 shows the effect of each dragging parameter on the cutting-edge radius. All parameters show equal incidence on the internal edge Ri as on the external edge Re. All parameters are increasing except for the abrasive mixture percentage RM, where the effect is intermittent. The combined effect of grit size GZ and drag depth ID causes an increase in radius (Fig. 3). This is because face contact (drag force) increases due to the higher pressure that the cutting edge is subjected to when it is under more abrasive (Fig. 1). Therefore, the rapidity and severity of the cutting-edge rounding process is directly related to DT, GZ and ID.

Fig. 2.
figure 2

Effect of parameters on the cutting-edge radius ER

Full size image
Fig. 3.
figure 3

Combined effect of grain size and drag depth on ER cutting edge radius, a) Re and b) Ri.

Full size image

The roughness Ra achieved on the surface of the cutting-edge radius is uniform. But the Rz differs in its values between the outer and inner cutting edges (Fig. 4). Rz is more sensitive to the detection of machined surface imperfections. Therefore, it is used for the control and monitoring of surface irregularities [15, 16]. It was found that the higher inclusion of alumina in the mixture makes the abrasive mixture more aggressive (RM = 50%). However, the Rz values obtained by Drag Finishing in this study are lower or equal to those obtained in tool manufacturing by grinding operation [15, 16].

Fig. 4.
figure 4

Surface roughness performance as a function of drag depth (IR) and abrasive mixture ratio (MR)..

Full size image

Table 2 indicates the sources of variation of the R&R analysis of the measurement system in obtaining the cutting-edge radius. The result of the analysis shows that the variation of the measurement system is equal to 19.35% of the process variation. Moreover, the contribution of repeatability is higher than reproducibility, which indicates that the system is in the marginal zone. Consequently, the acceptance is possible if the limitations of the process, the importance of the application and the cost are known [17, 18].

Table 2. Contribution of variables in R&R analysis.
Full size table
Fig. 5.
figure 5

ANN prediction results of cut-edge radius growth (ER).

Full size image

In reference to the neural network prediction, for the algorithm validation, only 363 epochs were needed to find the best training. Meaning that the errors were no longer reduced but stabilized. A coefficient of determination R2 of 0.943 with a standard deviation of 0.0104 was obtained. Therefore, the prediction was continued (Fig. 5). Finally, the average prediction error was 8.81% compared to the actual edge rounding values.

4 Conclusions

The parameters involved in obtaining a cutting-edge radius were, in order of importance: immersion depth, drag time, abrasive mixture percentage and abrasive size.

The difference found between Re and Ri indicates that the location and orientation of the cutting edge when dragging in the abrasive directly influences the result. In addition, grain mixing rate and grain size influence the surface roughness.

The dragging process allows a reduction of the roughness in terms of Ra and Rz (average: Ra = 0.12 μm, Rz = 0.5 μm), and the edge radius reproduction is acceptable compared to traditional processes such as brushing and blasting.

The prediction accuracy of the preparation radius with ANN was 94%, which proves the efficiency of the algorithm.