In this report, the potential of tool wear monitoring using acceleration sensors in gear hobbing was investigated using the acceleration curves of individual generating positions.

Gear hobbing is one of the most used soft machining processes in gear manufacturing due to its high productivity. The optimization goal of a manufacturing process is to reduce production costs while maintaining the required gear quality. Reducing the process time by adjusting the process parameters can result in increasing tool wear and possibly a reduction in gear quality. Therefore, knowledge of tool wear and gear quality during the process is necessary to optimize the manufacturing process. Tool Condition Monitoring (TCM) provides a methodical approach to detect critical tool wear during the manufacturing process. TCM is defined as the use of sensor technology to directly or indirectly monitor and predict the tool condition [1]. Different types of sensors are combined in TCM. For example, sensors are used to measure acoustic emission, airborne sound and acceleration. The main causes of these oscillations are deformation and friction during chip formation, as well as tool and chip breakage and tool vibration [2,3]. By analyzing the signal data, taking into account the process characteristics, it is possible to obtain information about the tool wear.

Widely used options of TCM are the machine-internal spindle power monitoring and acceleration measurements. These options are often used because they are either already installed in the machines or can be integrated into the machine with low effort. However, simply recording the data is not effective without a process-specific evaluation methodology. To date, process monitoring during gear hobbing has only been investigated in a few research projects. Systematic studies on the influence of the varying chip geometries created during gear hobbing and a comparison of different evaluation methods are not yet available.

1 State of the art

A large number of methods have been developed for TCM, which have been classified into the two categories of direct and indirect methods [3-6]. In direct methods, also known as offline measurement methods, the value of a wear parameter is determined directly, for example using optical measurements [6]. The disadvantage is the limited practical applicability due to usually limited accessibility in the process, the insufficient lighting and the use of cooling lubricant in the production process [5,6].

Table 1: Test series.

In comparison to direct methods, indirect methods have been the subject of more research [2]. Indirect methods, also known as online measurement methods, determine values that correlate with wear [4-6]. A universal method for all machining processes that can also be used in the industrial field has not been established yet [4, 7-9]. The individual complexity of machining processes implies a process-specific application of TCM, which means that there is a need for further research in this area [6,9]. By analyzing different characteristic values from different sensor types in the process, conclusions were drawn about various process influences [6,9]. The use of only one sensor made it difficult to detect signal errors caused by external influences [3]. A combination of sensors was described as a sensible approach for future research [2,10].

Table 2: Workpiece geometry.

For gear hobbing, HENDRICKS ET AL. investigated the potential of various sensors, such as acceleration, acoustic emission, and airborne sound sensors, regarding their suitability for the detection of tool breakage during gear hobbing. In the study, acceleration sensors in the range up to f = 200 kHz were considered suitable. Furthermore, it was shown the shift position significantly influences both the overall acceleration in the process and the order distribution. [11]

In addition, HENDRICKS ET AL. identified the sensor positions on the counter holder of the workpiece and counter bearing of the tool spindle as particularly suitable for detecting tool chipping and breakage. For the analysis of the signal data, the signal data was divided into segments to examine specific times in the process in more detail. [12]

Table 3: Tool geometry.

WU analyzed the possibility of using tool condition monitoring to ensure the required workpiece quality during gear hobbing. WU used a neural network to determine the total profile deviation Fα from acceleration data recorded over the process time. The training and subsequent prediction were also carried out using data packages split over the process. The quality of the results for neural networks depends on the training data set. As many parameters have an influence on gear quality in the hobbing process, it must be ensured that all influencing variables are included in the data for the training data set. The prediction of deviations that were not included in the training data set is otherwise only possible to a limited extent. [13]

Figure 3-1: Trial setup.

2 Objective and Approach

The review of the extant literature reveals the potential for process optimization in manufacturing through the implementation of TCM. Due to the time-varying contact and cutting conditions in gear hobbing, it is not possible to transfer a TCM method from other milling or turning processes to gear hobbing without further modification. The variability of the chip sizes and forms depending on the generating positions (GP) during gear hobbing requires a separate analysis of individual signal sections to detect wear.

However, systematic studies of the influence on the signal data of the individual cuts during the hobbing process have not yet been carried out.

The objective of this research is to investigate the potential for monitoring tool wear based on the acceleration data of individual GP. The approach is divided into three parts: First, the experimental methodology is defined. The sensors used and their positioning are presented, as well as the used tools, materials, and process parameters. In the next step, the methodology for data preparation is explained. For this purpose, the respective hobbing process is simulated to establish a correlation between the chip characteristics of the respective cuts and the acceleration data. The recorded acceleration data is then separated into the individual cuts and assigned to the simulated cuts. Finally, the acceleration data is evaluated to determine the influence of the GP on the acceleration data and to identify the GP and signal characteristics best suited for wear detection for different types of wear.

3 Experimental Methodology

The test methodology is explained below. First the design of experiments is presented. The tools and workpieces used are then defined and characterized in detail. Finally, the trial setup is presented.

Figure 4-1: Data separation in the fly-cutting process.

3.1 Design of Experiments

In this report, signal data from three different test series (A, B, C) were analyzed. The normal module of the gears in the three-test series was kept constant at mn = 2.557 mm. The test series differed in the workpiece material and tool substrate, as well as the number of teeth z0 and the number of gashes ni0 of the tool and the number of teeth z2 of the workpiece. The variation caused different wear phenomena on the tool, which are discussed in more detail in Section 6, were provoked. The varied parameters are summarized in Table 1. In all test series, the gear was machined in one climb cut without cooling lubricant. The resulting maximum chip thickness was calculated according to HOFFMEISTER [14]. The sampling frequency of the acceleration sensors was fs = 12.8 kHz in all test series.

The cutting speed was set to vc = 300 m/min to ensure comparability of all test series. As the influence of tool wear was to be investigated, the tests were carried out in the fly-cutting process. The fly-cutting trial is an established analogy trial for wear investigations in gear hobbing, in which the workpiece is machined with a single tooth (fly-cutter) from a hob and therefore good for wear investigations. The fly-cutting trials were carried out up to a previously defined wear criterion of a maximum wear mark width of VBmax = 150 µm or a maximum crater depth of KTmax = 150 µm. This was determined because it was observed that wear increases exponentially above this wear mark widths and crater depth respectively.

Figure 5-1: Undeformed chip geometries of individual generating positions.

3.2 Tool and Workpiece Characterization

Table 2 shows the geometry data of the gear. With a module mn2 = 2.557 mm, the analyzed test gears of the three-test series A, B, and C represented the dimension of an automotive gear. The workpiece geometry was kept constant for test series A and B. For test series C, the number of teeth z2 and the diameter of the gear was increased. The profile shift factor was adjusted from xE2 = 0.0557 to xE2 = 0.0597 in order to manufacture a gap geometry for test C that was comparable to the geometry from test series A and B.

The tools used were one single- and two double-started hobs made of different cutting materials with AlCrN coating. The hob with the number of starts z0 = 1 had ni0 = 16 gashes and the hob with the number of starts z0 = 2 had ni0 = 17 gashes. For test points A and C, a tool made of carbide K30 was used, while for test point B a tool made of powder metallurgically produced high-speed steel S390 was used. Further specifications of the tools are listed in Table 3. For the fly-cutting trials, the fly-cutters were separated from the solid tool using wire erosion.

Figure 5-2: Acceleration curves of individual generating positions.

The material of the workpiece was varied between the test series. Higher-strength materials were selected for the series of tests with the carbide tools. The hardness was measured on the workpiece surface using hardness tests according to BRINELL and converted into the corresponding tensile strength Rm according to DIN EN ISO 18265:2014-02 [15,16]. The material C45 had a hardness of 185 HBW 2.5/62.5, which corresponds to a tensile strength of Rm = 625 MPa. For the quenched and tempered steels 42CrMo4 and 56NiCrMoV7, hardnesses of 395 HBW and 320 HBW 2.5/62.5 were measured, corresponding to tensile strengths of Rm = 1,250 MPa and Rm = 1,035 MPa.

3.3 Trial and Measurement Setup

The trials were carried out on an LC180 gear hobbing machine from LIEBHERR. The trial setup had to be adapted for the two different workpiece geometries (A/B and C, see Section 4.2), see Figure 3-1. A larger workpiece clamping and a longer tool shaft were used for the gear with the larger outer diameter (C). In addition, a different counter holder and a different acceleration sensor were used. Therefore, the test series are not directly comparable. Both acceleration sensors were able to detect vibrations in the frequency range of f = 0.2Hz – 6.4kHz.

Figure 5-3: Characteristic value curves across the generating positions in the individual loops.

The position on the counter holder of the tool was identified as the most suitable position for wear detection in addition to the position on the counter bearing [11]. The present study focuses exclusively on the acceleration at the counter holder in the x-direction, with the intention of providing a practical illustration of the procedure.

4 Data Processing

The signal data was recorded during the fly-cutting trials over the individual loops U. The data was subsequently separated so it could undergo further evaluation and generation of characteristic values. The procedure is explained below using gear A as an example.

Figure 5-4: Machined volume for all generating positions for the A.

The left of Figure 4-1 shows one loop U in the fly-cutting trial. The tool is continuously shifted from the theoretical first generating position GPF to the last generating position GPL. The GPF and GPL are defined by the engagement range lE of the hob and the gear. The generating position GP = 0 is in the center axis of the workpiece. The manufacturing simulation SPARTAPRO calculates the GP with a fixed incremental distance to each other. For test gear A, generating positions from GPF = -64 to GPL = 28 were calculated. However, the undeformed chip geometry in two adjacent GP differs insignificantly, see Figure 4-1 top right.

In the fly-cutting trial, for each of these increments during the shifting process, a specific number of cuts were performed, corresponding to the number of teeth z2. For gear A, this corresponds to z2 = 39 number of teeth. The measured acceleration data of the 39 cuts of the respective GP are shown below the simulated undeformed chip geometries in Figure 4-1. Analogous to the undeformed chip geometry, the acceleration data for the different generation positions do not differ significantly either qualitatively or quantitatively. Consequently, a separate simulation of the exact undeformed chip geometries or the adaptation of the fly-cutting trials was not carried out.

Figure 6-1: Wear behavior of test A.

The duration of one tool revolution was initially selected as the segment size for the subsequent analysis of the acceleration data. The result of Equation 4-1 was a curve with an = 643 data points or a time of tn = 50 ms. As such a small number of data points leads to inaccuracies for the subsequent analysis of the acceleration curves, particularly in the frequency range, the segment size was increased to the duration of tool rotations equal to the number of teeth of the gear z2. During this time, a cut is made in each gap of the gear. The number of data points equals an,new = an · z2 = 643·39 = 25,077.

Equation 1

where:
an is the data points per tool rotation.
da0 is the outer diameter of the tool;.
vc is the cutting spee.
fs is the sampling rate.

Characteristic values were then calculated from the separated time series data. The Python Time Series Feature Extraction Library (TSFEL) created by BARANDAS ET AL. was used for this [17]. The library enables analysis of more than 60 different characteristic values from the time, statistics, and frequency domain. After the preparation of the data, the undeformed chip geometry, the acceleration curves consisting of z2 sections, and the characteristic values derived from them were thus available for each GP of the respective tests. The following section will present an analysis of the data in order to identify potential correlations both among the characteristic values and with respect to wear over the time of the test.

5 Influence of the Generating Positions on the Process Signals

Figure 5-1 shows the simulated undeformed chip geometries of individual GP and loops for test A. The GP in which the maximum chip thickness hcu,max and the maximum chip length lcu,max occur were selected as examples. The upper part of Figure 5-1 shows the undeformed chip geometry of the GP = -60 for the third and seventh loop U. In this GP, the maximum chip thickness hcu,max occurs in the process during the seventh loop. There are only insignificant differences, both qualitatively and quantitatively, in the form of the undeformed chip geometry and the amount of the chip characteristics between this and the same GP in the third loop. In the 11th loop, the tool is not cutting anymore in this GP.

Figure 6-3: Analysis of the acceleration data for test series A.

The lower section of Figure 5-1 shows the chip geometries of GP = -19 for the loops U = 3, U = 7, and U = 11. In the seventh loop, the chip geometry with the maximum chip length lcu,max occurs in this GP. In the third and 11th loop, the undeformed chip geometry is cut off at the end and beginning of the cut. At this point, the tool exits the workpiece at the lower workpiece edge or enters the workpiece at the upper workpiece edge.

Figure 5-2 shows the acceleration curves associated with the respective GP. The curves shown were recorded during the machining of the first gear with an unworn tool. For the GP = -60, no significant difference was found in the qualitative progression between the third and seventh loop. The maximum value of the acceleration was increased by approximately Δ = 15 % in the seventh loop compared to the third loop. However, the chip characteristics of the two cuts only differed by approximately Δ = 2 % in the maximum chip thickness hcu,max, chip length lcu,max, and machined volume Vcu. The maximum wear mark width after manufacturing the first gear was approximately VB = 20 µm along the entire cutting edge. Accordingly, an influence of wear could already be detectable between the third and seventh loop.

Figure 6-4: Wear behavior and analysis of the acceleration data for test series B.

Tool wear increases relatively sharply at the beginning of the tool life until a constant wear level is reached [18]. This is referred to as initial wear. An alternative hypothesis for the observed increase in the maximum value of the acceleration is the changed position of the cutting point in relation to the sensor. With each loop, the tool is moved by the amount of the axial feed fa in z-direction. Hence, between the loops U = 3 and U = 7, there is a difference in distance from the sensor to the cutting point of Δs = 4∙fa = 18.8 mm. This also changes the transfer path to the sensor, which influences the measured values.

Due to the very different chip geometries of the GP = -19 in the three loops, there were also greater differences in the acceleration curves of the GP = -19 compared to the curves of the GP = -60. The maximum acceleration values increased both from loop U = 3 to U = 11 and from U = 11 to U = 7. Thus, a correlation between the maximum acceleration value and both the chip length lcu and the machined volume Vcu, was found. However, no correlation occurred between the maximum acceleration value and the maximum chip thickness, which was identical in U = 7 and U = 11. The maximum acceleration value in U = 7 is slightly higher in GP = -19 with amax = 0.386 than in GP = -60 with amax = 0.382. The peak-to-peak value of these GP showed a difference of Δ = 24%. The maximum acceleration value is therefore not only dependent on the maximum value of the chip thickness. Other possible dependencies are the machined volume Vcu or the amount of the unrolled tool profile in contact.

Figure 6-5: Wear behavior of test C.

In order to further analyze the relationships between the chip geometries and the associated acceleration curves, the characteristic values derived from the acceleration curves for each GP were considered. Figure 5-3 shows four characteristic values as examples: The amounts of the first, 10th, and 15th tool rotation orders as well as the root mean square value of the acceleration. The first tool rotation order corresponded to a frequency of f1 = 19.89 Hz. The 10th and 15th tool rotation orders are 10 and 15 times the first order, respectively.

The curves of the characteristic values of the first and 10th order in the individual loops exhibited an approximately triangular curve and showed a single maximum. This maximum was detected at the beginning of each loop in the first GP. The acceleration values of the first order declined significantly immediately after reaching the maximum, while a drop with a lower gradient in the acceleration values was detected in the 10th order. The acceleration curves of the 15th tool rotation order and the root mean square value showed a second local maximum in the middle GP of the individual loops in addition to the first maximum at the start of the loops in the first GP.

Figure 6-6: Analysis of the acceleration data for test C.

The maxima were detected for all characteristic values of the signal data in the seventh and eighth loop in the range of GP = -40. In these GP, the greatest volume Vcu is machined during the process. For this reason, Figure 5-4, analogous to Figure 5-3, shows the machined volume Vcu calculated using SPARTAPRO for all GP in each loop U. Different areas of the cutting edge were defined in order to determine correlations between the characteristic values from the signal data and the cutting-edge sections in contact. The cutting edge was divided into the areas of the leading flank (LF), the trailing flank (TF), and the tip (T), see Figure 5-4 bottom left. The leading and trailing flanks were further divided into the areas A, the curved area of the flank near to the tip of the tool, and B, for the rest of the flank. In general, all curves have a similar triangular shape with a single maximum. While for the cutting-edge areas TF-A, LF-A, and T, the maximum is in the first GP at the beginning of each loop; in the TF-B area, most of the volume is machined in the middle GP. From the GP = 0, no more volume is machined in the cutting-edge area TF-A. This evaluation can be used for the subsequent analysis of the acceleration signals regarding tool wear. In this way, it is possible to determine the area of the cutting edge that cuts the highest machined volume Vcu, so that it can be assumed that the influence of wear on the acceleration data is particularly high in this area.

In addition to the machined volume Vcu, other factors also affect the acceleration signals. One factor is the chip form. A distinction is made between single-flank chips, see Figure 5-1, U = 3, GP = -19, and multi-flank chips, see Figure 5-1, U = 7 GP = -19. Multiple flank chips can influence the acceleration curves several times in a single cut because different cutting-edge areas enter the cutting process one after the other. These multiple impacts occur primarily in the higher GP in the higher loops and could be better detected with the frequency values of the higher orders. For this reason, orders from the 15th order onwards should be taken into consideration for the detection of wear in these GP.

6 Influence of Wear on the Process Signals

In the following section, the influence of different types of wear on the acceleration signals of individual GP is analyzed. The maximum wear mark widths VBmax were measured on the five cutting edge sections, described in the previous chapter, after each manufactured gear. The maximum crater depth KTmax on the rake face was also measured. The wear was assumed to increase linearly over the production time of a gear for comparison with the acceleration curves of the individual GP over the loops. The BRAVAIS-PEARSON correlation coefficient r was then calculated for each GP according to Equation 6-1 between the amount of wear and the characteristic values of the acceleration data calculated by the TSFEL. This allowed us to determine the most suitable GP and characteristic values for wear detection.

Equation 6-1

where:
r is the correlation coefficient.
x is the characteristic value.
x   is the mean value of characteristic values.
y is the value of wear mark width.
y   is the mean value of the wear mark width values.

The results of test series A are considered first, in which the highly tempered steel 42CrMo4 with a tensile strength of Rm = 1,250 MPa was machined with a carbide fly-cutter. Figure 6-1 shows the wear curve for the five cutting edge sections and the crater wear curve. After N = 8 machined gears, the maximum wear mark width on all cutting-edge areas was still below VBmax = 100 µm. A small cutting-edge breakout in the TF-B area of the fly-cutter can be seen on the microscope images after N = 8 gears in Figure 6-1. During the machining of the ninth gear, a tool breakage occurred at this point, so the process was stopped after U = 9 out of U = 12 loops.

For each loop U of the ninth gear, the acceleration curves of the individual GP were compared with those in the same loop U of the eighth gear in order to detect the time of tool breakage in the process. In loop U = 4, the first significant differences were detected. Therefore Figure 6-2 shows the order spectra of the loop U = 4 for the first and last gear. To determine the influence of wear, the order spectrum of the first gear was subtracted from the order spectrum of the last gear. This eliminates the frequencies excited by machining without wear and allows us to identify the orders that are more strongly excited by tool wear. For the considered loop U = 4, these sensitive orders were in the range from the 20th to the 43rd tool rotation order and from the GP = -32 to GP = -10. The orders around the 30th and 42nd tool rotation order were particularly dominant. Therefore, from GP = -32, the tool wear increased significantly. In these GP, multi-flank chips were present, where several cutting-edge areas engage in sequence.

In addition, the mean value of the acceleration curves in the time domain was calculated from the z2 = 39 cuts in each GP. Figure 6-3 shows this curve for the fourth loop of the two last machined gears. Up to GP = -33, the acceleration curves of the two gears were congruent. After that, a slight difference of 5% between the maximum values of the data could initially be identified, which then increased from GP = -24 to more than 20% difference. From GP = -12, the data were congruent again. From this GP onwards, the cutting-edge area TF-B where the tool broke was no longer involved in the machining process.

For a more detailed analysis, the development of two characteristic values for the GP = -33 to GP = -29 in loop U = 4 of each gear was compared to analyze how early the wear can be detected. The peak-to-peak characteristic value, which had a local maximum in these GP, see Figure 6-3, and the 30th order, which lies in the wear-sensitive range identified in Figure 6-2, were selected for this purpose. The peak-to-peak value is shown in the top right of Figure 6-3. For the GP = -33 and GP = -32, peak-to-peak values of a = 1.17 ± 0.02 g were determined across all gears. The calculation of the correlation factors resulted in values of rptp_VBT = 0.13-0.22, which show there is no correlation with the wear values. For the GP = – 31 and GP = -30, the increased amount of the maximum wear mark width VBmax on the tip of the fly-cutter after gear N = 5 was detected with a correlation coefficient of rptp_VBT = 0.55 and rptp_VBT = 0.75 using the peak-to-peak characteristic values. However, the exponential increase in the maximum wear mark width VBmax in the ninth gear could not be detected in these GP. In contrast, while the peak-to-peak value curve for GP = -29 is almost identical to the previous GP up to the eighth gear, an exponential increase was detected for the ninth gear. The correlation coefficient for this GP was rptp_VBT = 0.92.

Furthermore, the amount of acceleration of the 30th tool rotation order is shown. For this characteristic value, the exponential increase in wear could already be detected in GP = -32 with a correlation coefficient of r30.order_VBT = 0.76. The correlation r30.order_VBT = 0.72-0.90 between the maximum wear mark width VBmax and the characteristic value was also detected in the subsequent GP. The undeformed chip geometry in the considered GP is shown as an example for GP = -31 at the bottom left of Figure 6-3. In this GP a multi-flank chip geometry is formed, in which the trailing flank enters the workpiece first and the tip of the tool enters later. This chip form therefore involves two tool engagements with the workpiece. igher orders therefore not only detect these multiple tool engagements, as shown in Section 6, but can also be used to detect wear occurring in these GP.

The wear curves and exemplary characteristic value investigations from test series B are shown in Figure 6-4. In this test, C45 was machined with an S390 tool. The dominant type of wear in this test was crater wear. This increased exponentially over the last three gears until the maximum permitted wear criterion of KTmax,perm = 150 µm was reached. As the crater wear extended to the clearance surface, the wear mark width at location TF-A also increased exponentially toward the end of the test. The wear mark widths of the other cutting edge areas increased linearly over the test period up to wear mark widths of VBmax,T = 40 µm at the tip and VBmax,TF-B = 90 µm on the trailing flank.

In order to monitor the tool wear, characteristic values were calculated for each GP analogous to the procedure for the test series A and compared for correlation with the maximum wear mark widths VBmax of the cutting-edge areas. The greatest correlation with a correlation coefficient of r30.order_KT = 0.98 between the crater wear depth KTmax and the acceleration values was determined for the amount of the 30th tool rotation order in the GP = -10 in loop U = 20, see Figure 6-4 top right. For the cutting-edge areas TF-B and LF-A, the peak-to-peak value in the GP = 2 in U = 20 was determined as the most suitable characteristic value with a correlation coefficient of rptp_VBTF-B = 0.92 for wear detection. The peak-to-peak value increased approximately linearly with the number of gears and the respective maximum wear mark widths VBmax. However, this value was subject to large variations due to the very low machined volume Vcu in this generation position GP = 2 and the associated large influence of any errors in the test setup.

Figure 6-5 shows the wear behavior of test C. In this test, a carbide tool was used for machining in the same way as in test A. However, the workpiece had a significantly larger outer diameter of da0 = 219 mm and therefore also a higher number of teeth z2 = 76. In contrast to test series A, the greatest wear in this test did not occur at the tip T or in area A of the trailing flank TF-A, but in area B of the trailing flank TF-B. Figure 6-5 on the right shows microscope images of the trailing flank after machining the third, fourth, and fifth workpiece. After the third workpiece, the flank was worn evenly across the width up to the tip. After the fourth gear, chipping of the cutting edge occurred on the flank. At the end of the trials, large parts of the cutting edge in the TF-B area had broken off, while the remaining areas of the tool cutting edge were still below the wear criterion with wear mark widths of VB = 50 µm.

In the top left of Figure 6-6, the acceleration curve of all GP of loop U = 4 for the last and penultimate machined gear is depicted, analogous to the procedure for test series A. The fourth loop was selected on the basis that substantial differences were already apparent at this point. The most significant differences in the acceleration data were detected around GP = -68. Therefore, the GP = -69 to GP = -67 were analyzed in more detail. The diagrams on the right of Figure 6-6 show the peak-to-peak characteristic values and the amount of the 29th tool revolution order over the manufactured gears in loop U = 4. The development of the peak-to-peak value was almost identical for the three generating positions GP = -69 to GP = -67. The largest increase in the peak-to-peak value was detected between gear four and five.

The exponential increase in wear on the trailing flank TF toward the end of the test, which can be seen in the diagram in Figure 6-5, was not represented by the peak-to-peak value. However, in these GP, the majority of the machined volume Vcu was machined with the tip of the tool, which is why the worn area of the cutting edge only had a small influence on the resulting accelerations. It can also be assumed the cutting edge is continuously worn and has therefore not yet reached the measured amount of wear after U = 13 in loop U = 4. The value of the 29th order could not detect the wear behavior for the generating positions GP = -69 to GP = -67. As shown in the previous investigations, this is due to the chip form.

Although the chip form is also multi-flanked, as shown at the bottom left in Figure 6-6, the tool enters the workpiece with the tip continuously followed by the trailing and leading flank. Consequently, there is no second impact of the tool during the cut and therefore the higher orders were less strongly excited.

Conversely, the characteristic values in the GP = -31 and GP = -30 exhibited the opposite trend. Here, two chips were formed in the GP. The value of the 29th order increases linearly from the second to the fifth gear. The peak-to-peak values form an S-shaped curve. This could not be correlated with any wear progression of the cutting-edge areas.

The following conclusions were derived from the investigations. For any type of wear at any point on the cutting edge, the utilization of simulated undeformed chip geometries facilitates the determination of characteristic values for different GP occurring in the process. These values correlate with the respective local wear. To illustrate this, regression Equation 6-2 and Equation 6-3 were established from these characteristic values to describe the wear at two different points on the tool in test C as examples.

Equation 6-2

where:
VBmax,TF-B is the maximum wear mark width at the point TF-B.
X is the amount of the 27th tool rotation order in the GP = -36 in U = 12.

Equation 6-3

where:
VBmax,LF-A is the maximum wear mark width at the point LF-A.
Y is the peak-to-peak value in the GP = -13 in U = 12.

The coefficient of determination was calculated to assess the quality of the equations. The equations demonstrated a commendable capability to represent the wear behavior, with R2TF-B = 95% and R2LF-A = 92%. These equations can be established for each GP in each loop of the process. For further tests, the derived regression equations enable the wear condition to be determined over the entire process time.

7 Summary and Outlook

In this report, the potential of tool wear monitoring using acceleration sensors in gear hobbing was investigated using the acceleration curves of individual GP. For this purpose, acceleration data recorded by acceleration sensors attached to the counterholder of the workpiece were analyzed in fly-cutting trials for different tool and workpiece materials. These acceleration data of the individual GP were first examined with regard to their correlation to the chip characteristics by a manufacturing simulation. The machined volume Vcu and the shape of the undeformed chip geometry for individual GP were identified as the main influencing variables on the resulting accelerations. A complex chip shape, in which different areas of the cutting edge enter the workpiece in clearly separated sequences, results in higher acceleration values. This is particularly evident in the values of the higher orders of the tool rotation order. For comparable undeformed chip geometries, the accelerations depend on the volume Vcu and maximum chip thickness hcu,max of the chip formed.

Correlation coefficients were calculated between the wear amounts of the individual cutting-edge sections and the acceleration parameters for each GP occurring in the process. This allowed us to determine the most suitable GP and characteristic values for wear detection. In order to detect wear at different areas of the cutting edge, the chip shape must also be taken into account. In GP where there were multiple engagements of the tool, the characteristic values of higher orders were particularly suitable for wear detection. In contrast, lower orders or peak-to-peak values were suitable for GP without multiple tool engagements. If the area of the cutting edge with the highest amount of wear was engaged in the GP, only the wear progression of the greatest amount of wear could be detected, even if other areas of the cutting edge had a significantly higher values of machined volume Vcu. In order to detect wear on all cutting-edge areas, it is therefore necessary to consider GP in which other parts of the cutting edge are not engaged as much as possible.

For a more in-depth analysis, larger test volumes on the same gear geometry are required. With more data, the observed trends can be verified, and the developed method can be validated. In addition, further correlations can be analyzed regarding other influences such as cutting speed. If reliable findings on the correlation between wear and characteristic values formed in defined GP are available, these can also be used for wear detection in the gear-hobbing process, where the individual cuts or the individual GP occur in superimposed form. 

Acknowledgement

The project is funded by the German Research Foundation (DFG) under the project number: 521375603.

References

  1. Zhang C., Yao X., Zhang J., and Jin H., 2016, “Tool Condition Monitoring and Remaining Useful Life Prognostic Based on a Wireless Sensor in Dry Milling Operations,” Sensors (Basel, Switzerland), 16(6).
  2. Li X., 2002, “A brief review: acoustic emission method for tool wear monitoring during turning,” International Journal of Machine Tools and Manufacture, 42(2), pp. 157–165.
  3. Chen X., and Li B., 2007, “Acoustic emission method for tool condition monitoring based on wavelet analysis,” International Journal of Advanced Manufacturing Technology, 33(9-10), pp. 968–976.
  4. Ai C. S., Sun Y. J., He G. W., Ze X. B., Li W., and Mao K., 2012, “The milling tool wear monitoring using the acoustic spectrum,” International Journal of Advanced Manufacturing Technology, 61(5-8), pp. 457–463.
  5. Kishawy H. A., Hegab H., Umer U., and Mohany A., 2018, “Application of acoustic emissions in machining processes: analysis and critical review,” International Journal of Advanced Manufacturing Technology, 98(5-8), pp. 1391–1407.
  6. Siddhpura A., and Paurobally R., 2013, “A review of flank wear prediction methods for tool condition monitoring in a turning process,” International Journal of Advanced Manufacturing Technology, 65(1-4), pp. 371–393.
  7. Kosaraju S., Anne V. G., and Popuri B. B., 2013, “Online tool condition monitoring in turning titanium (grade 5) using acoustic emission: modeling,” International Journal of Advanced Manufacturing Technology, 67(5-8), pp. 1947–1954.
  8. Rubio E. M., and Teti R., 2009,“Cutting parameters analysis for the development of a milling process monitoring system based on audible energy sound,” J Intell Manuf, 20(1), pp. 43–54.
  9. Pechenin V. A., Khaimovich A. I., Kondratiev A. I., and Bolotov M. A., 2017, “Method of Controlling Cutting Tool Wear Based on Signal Analysis of Acoustic Emission for Milling,” Procedia Eng., 176, pp. 246–252.
  10. Bhuiyan M., Choudhury I. A., and Dahari M., 2014,“Monitoring the tool wear, surface roughness and chip formation occurrences using multiple sensors in turning,” Journal of Manufacturing Systems, 33(4), pp. 476–487.
  11. Hendricks S., Troß N., Brimmers J., and Bergs T., 2022, “Potential and challenges of tool condition monitoring in gear hobbing,” Forsch Ingenieurwes.
  12. Hendricks S., 2022, Potential of Process Monitoring using Vibration Sensors in Hobbing, Aachen conference on gear production: Gear finishing technology and quality inspection, 9th – 10th November 2022, 1st ed., Bergs T., and Brecher C., eds., Apprimus Verlag, Aachen, pp. 172–177.
  13. Wu D., Yan P., Pei J., Su Y., Zhou H., Yi R., and Hu G., 2022, “Quality estimation method for gear hobbing based on attention and adversarial transfer learning,” Measurement, 188, p. 110383.
  14. Hoffmeister B., 1970, “Über den Verschleiß am Wälzfräser (English translation: About the wear on the hobbing tool (not available in English)),” Ph.D. thesis, Werkzeugmaschinenlabor (WZL) der RWTH Aachen, RWTH Aachen University, Aachen.
  15. DIN EN ISO 18265:2014-02, “Metallic materials – Conversion of hardness values.”
  16. DIN EN ISO 6506-1:2015-02, “Metallic materials – Brinell hardness test.”
  17. Barandas M., Folgado D., Fernandes L., Santos S., Abreu M., Bota P., Liu H., Schultz T., and Gamboa H., 2020,“TSFEL: Time Series Feature Extraction Library,” SoftwareX, 11, p. 100456.
  18. Klocke F., 2011. Manufacturing Processes I: Cutting, Springer, Berlin Heidelberg.