CHAPTER 4 EXPERIMENTAL STUDIES 4.1 INTRODUCTION

CHAPTER 4 EXPERIMENTAL STUDIES 4.1 INTRODUCTION The experimental set up and procedures are described in the following subsections. Two sets of experiments were done. The first study involves determination of the best feature-classifier combination for a set of supervised class of data. The vibration signals during the turning of mild steel rod were taken for the four types of pre-determined classes during the experiment. The second set of experiment consisted of acquisition of vibration signals along with the measurement of the corresponding surface roughness during the turning of EN8 material, using carbide and coated carbide tips. For each of the combination of three spindle speeds, feed and depths of cut, surface roughness values were measured for three stages of flank wear. A detailed narration of the experiments carried out is given below in two sub-sections. 4.2 EXPERIMENTAL SETUP FOR THE STUDY ON TOOL WEAR Fig. 4.1 Experimental setup for tool wear classification study The experimental setup, shown in Fig. 4.1 for obtaining the vibration signals during a turning operation consisted of a CNC turning center (ACE Micromatic - Classic 20T), a piezoelectric accelerometer, a signal acquisition and signal conditioning unit and a computer to record the signals. A mild steel shaft of 20mm diameter was held in a pneumatic chuck and a single point carbide tipped cutting tool (turning tool) was fixed 48

on a tool post. A tool nose radius of mm, feed rate of 0.1mm/s, depth of cut of mm and spindle speed of 600 rpm were the parameters set on the CNC machine. The piezoelectric accelerometer (Dytran model) was directly mounted on the tool holder using an adhesive. The accelerometer was then connected to a signal-conditioning unit (DACTRAN FFT analyzer), where the signal goes through the charge amplifier and an Analog-to-Digital Converter (ADC). The vibration signals in digital form were input to the computer through an USB port. RT-Pro-series software was used for recording the signals directly to the computer s secondary memory. The signal was then read and processed to extract different statistical features. 4.2.1 SPECIFICATIONS OF THE ACCELEROMETER USED General purpose piezoelectric accelerometer (Miniature type) with Integrated Electronic Piezoelectric (IEPE) transducer with integrated signal-conditioning unit powered by a constant current supply was used. The specification of the accelerometer is given below: Make : Dytran Instruments Inc. USA Model Number : 3035B1 Weight : 2.5 grams Description : 500 g range Frequency : 10 k Resonance Frequency : 45 khz Sensitivity : 10 mv/g 4.2.2 EXPERIMENTAL PROCEDURE The following section deals the experimental procedure, simulation of conditions and signal acquisition. The scheme of modelling is presented in Fig. 4.2. 4.2.2.1 Acquisition of baseline signal An unused new carbide tool tip (TNMG160408) was mounted on a tool holder that was fixed on the tool post. An accelerometer was fixed on the tool holder using adhesive as shown in the Fig. 4.1. The parameters for signal acquisition like sampling frequency (24 khz), sampling length (8192), type of signal (amplitude only in text format) etc., were 49

set. Since highest frequency was found to be 12 khz the sampling frequency was set to 24 khz (based on Nyquist theorem). The mild steel rod of 20mm diameter was held using a self-centering chuck and the top layer was rough turned to remove the oxidised outer surface and to smoothen out the rod. The data acquisition system was switched on and first few signals were rejected intentionally to avoid initial random variation that is typical in any measurement. One hundred and eighty signals were acquired after the stabilization of the process. Fig.4.2 Scheme of the modeling 4.2.2.2 Condition Simulation The following conditions were considered for the present study. 1 The tool tip is good condition (Good) 2 The tool tip in less blunt (0.3mm) condition (tool blunt low). 3 The tool tip in more pronounced blunt (0.6mm) condition (tool blunt high). 4 The condition, where the tool tip was loosened by a one- twelfth a revolution (tool tip loose). In all the above conditions the tool was assumed to be otherwise perfect than the above said defects. The first condition is the perfectly Good and unused tip. The second and the third condition, namely, tool blunt low and tool blunt high depict the flank wear during normal tool wear at two stages (low and high). The wear (bluntness) considered 50

was for gradual usage and since only flank wear on the nose and the resulting recession of the cutting edge directly affects the work-piece dimension and work piece quality [136], these conditions were considered. The requirement for the fourth condition may be explained as follows. Since, the tip is held by a screw, it may get loosened during cutting process [137], or could be an operator error, hence it was also included as one of the condition of study. These conditions are required during the development of an online tool condition monitoring and alert system. This fourth condition is named as Tool tip loose (TTL) and was simulated by loosening the fully tight tool screw by one twelfth of a revolution. Vibration signature was used to tap the information about condition of the tool. It is to be highlighted here that many researchers relied on signals from multiple sensors or only on acoustic emission. The multiple sensors incurs higher experimental cost while the acoustic emission signals prove to be more effective in detecting sub-surface cracks and tool tip breakages [18, 47, 77]; however, for tool wear studies and surface roughness predictions, vibration signals seem to be a better candidate in terms of simplicity and low experimental costs [80, 94, 119]. Artificial tool wear was created in the tool tip as per the the following procedure. A reference line was marked parallel to the tangent of the nose radius in a fresh tool tip. The distance between the reference line and the highest point on the nose radius was measured and recorded. The tip was then taken to a tool and cutter grinder and the nose was ground by a small distance. The distance from the edge was measured from the reference line and the bluntness was calculated by taking the difference. The same procedure was carried out for a more pronounced bluntness and quantified. 4.2.2.3 Acquisition of Signal The vibration signal from the piezoelectric pickup mounted on the tool holder was recorded after allowing the turning process to stabilize for sometime (about 1 min.). The sampling frequency was 24 khz and the sample length was 8192 for all conditions. The sampling length was chosen to be 8192 which is equal to 213, around 10000 readings. 4.2.2.4 Statistical feature division of test and train data: The entire signals were divided into two halves as train data and test data. The train data was used to train the c4.5 algorithm (and other algorithms) and the test data was used to 51

test the respective trained algorithm. The corresponding classification accuracy was obtained and discussed in chapter 5. The statistical features extracted from the vibration signals acquired for all the above four conditions and used to train the C4.5 (and other) algorithms is given in the Appendix7 from Table A7.1 to A7.4. Similarly, the test data containing the statistical features that were extracted from the vibration signals acquired for all the four conditions is given in Appendix 7 from Table A7.5 to Table A7.8. Fig. 4.3 shows the time-domain signals taken from the tool holder for different conditions. They show time domain plots of vibration acceleration of a good tool tip (new tool tip without any fault), tool tip with flank wear 0.2 mm, tool tip with flank wear 0.4 mm. Fig. 4.3 Plot of time domain signal 4.2.2.5 Histogram feature extraction: The entire signal was scanned for the maximum value of the amplitude and the minimum value of the amplitude. This can also obtained by finding out the maximum value of maximum statistical feature and the minimum of the minimum statistical feature that has already been extracted from the signal. The difference between the two values gives the entire range of the amplitude. This range is divided into twenty equal parts to get the bin values. By using a macro written in MS-Excel we can find the frequency of amplitude values of the signal that falls within each bin. This is carried out 52

for all the 360 signals to obtain the histogram features. The histogram features are then divided into two halves as train data and test data and provided in the Appendix 7 as Tables from Table A7.9 to Table A7.12 and Table A7.13 totablea7.16 respectively. However, the bin values used for extracting the histogram features is given hereunder as Table 4.1. Table 4.1 Bin values used to extract histogram features Histogram Feature Lower value Upper value bin width h1-0.06598-0.05928 h2-0.05928-0.05258 h3-0.05258-0.04588 h4-0.04588-0.03918 h5-0.03918-0.03248 h6-0.03248-0.02578 h7-0.02578-0.01908 h8-0.01908-0.01238 h9-0.01238-0.00568 h10-0.00568 0.00102 h11 0.00102 0.00772 h12 0.00772 0.01442 h13 0.01442 0.02112 h14 0.02112 0.02782 h15 0.02782 0.03452 h16 0.03452 0.04122 h17 0.04122 0.04792 h18 0.04792 0.05462 h19 0.05462 0.06132 h20 0.06132 0.06789 0.0066 4.3 EXPERIMENTAL SETUP FOR SURFACE QUALITY PREDICTION The experimental setup shown in Fig. 4.4 for recording the vibration signals and the corresponding surface roughness during a turning operation consists of a CNC turning center, a piezoelectric accelerometer, a signal acquisition unit, a FFT analyzer, a computer to record signals and a surface measurement unit. The CNC machine - DX200 is a product of Jyothi CNC, Gujarat, India, and the one used for the experiment has the following specifications: 53

Maximum Turning Length : 500 mm Maximum Turning Diameter : 350 mm Swing Over Bed : 500 mm Speed Range : 50-4000 rpm Spindle Motor Power (Continuous rating/30 min. rating) : 9 kw/ 12 kw Rapid Feed (X and Z axis) : 24 M/min Controller System : SINUMERIK - 802D Accelerometer Fig. 4.4 Vibration signals being acquired for surface roughness prediction 4.3.1 FFT ANALYSER AND ACCELEROMETER SETUP The piezoelectric accelerometer (Dytran model) was directly mounted on the tool holder using an adhesive. The accelerometer was then connected to the signalconditioning unit (refer Fig. 4.5 DACTRON make FFT analyzer), where the signal goes through the charge amplifier and an Analog-to-Digital Converter (ADC). The vibration signals in digital form were input to the computer through an USB port. RT-Pro-series software was used for recording the signals directly to the computer s secondary memory. The signal was then read and processed to extract different statistical features. 54

Fig. 4.5 DACTRON make FFT analyzer 4.3.2 TOOL TIP PREPARATION The tools used were carbide inserts TNMG160408. There were three different stages of wear on the flank that were used. Apart from using brand new carbide tipped tool, tool tips of the same grade and make but with a flank wear of 0.2 mm and 0.4 mm were used. The tool tips with flank wear were measured using an optical measuring instrument which had a computerized digital image on the screen with crosswire and a digital readout as shown in Fig. 4.6. A large number of tips that were used in regular production and discarded were collected. They were carefully measured and classified into 0.2 mm and 0.4mm worn out tips by giving a red mark on the 0.2mm tip and a green mark on the 0.4mm tip. Around eighty numbers of tips were collected and were ensured that they were all from the same manufacturer and had the same grade. It was observed that carbide tips do not have a normal wear as in the case of HSS, since these tools are basically powders bonded, compacted and cured. At higher magnification they show that there is always a chip off of various sizes and shapes. It was ensured that enough number of carbide tips was made available so that for each experimental reading the tips were changed. 4.3.3 EXPERIMENTAL PROCEDURE The procedure to acquire signals consisted of mounting an unused new carbide tool tip (TNMG160408) in a tool holder and was fixed on the tool post. The accelerometer was fixed on the tool holder using adhesive mounting technique as shown in the Fig. 4.3. 55

The signal acquisition parameters like sampling frequency (24 khz), sampling length (8192), type of signal (amplitude only in text format) etc., were set. The highest frequency was found to be 12 khz and as per Nyquist sampling theorem the sampling frequency was chosen to be twice that of the highest measured frequency as 24 khz. The 20 mm EN8 (AISI 1040) steel rod was clamped using a soft-jawed self-centering chuck and a rough turning was carried out to remove the top layer that had undergone oxidation and smoothen out the surface of the rod. Fig. 4.6 Flank wear measurement of tool tip. The data acquisition system was switched on and first few signals were ignored purposefully to avoid initial random variation that is typical in any measurement. Once the process was stabilized then one hundred and fifty signals were acquired. 4.3.4 ACQUISITION OF SIGNAL The vibration signal from the piezoelectric pickup mounted on the tool holder was recorded after allowing the turning process to stabilize for sometime (about 1 min). The sampling frequency was 24 khz and the sampling length was chosen as 8192 which is equal to 213 (around 10000 readings). 56

4.3.5 EXPERIMENTAL DESIGN It was decided to use full factorial design. Factorial designs allow for the simultaneous study of the effects that several factors may have on a process. When performing an experiment, varying the levels of the factors simultaneously rather than one at a time is efficient in terms of time and cost, and also allows for the study of interactions between the factors. Interactions are the driving force in many processes. Without the use of factorial experiments, important interactions may remain undetected. 81 experiments (3 x 3 x 3 x 3) were conducted by varying the cutting parameters along with three different flank wear (viz. new tool, flank wear 0.2 mm and flank wear 0.4 mm) of the tool. The spindle speed was set at 500 rpm, 700 rpm and 900 rpm and for each speed; the feed was set at 0.05 mm/sec, 0.7 mm/sec and 0.09 mm/sec. For each speed and feed combination, the depth of cut was kept at mm, mm and mm. Table 4.2 Experimental design matrix for 500 rpm. Speed (rpm) 500 Feed DOC (mm/s) (mm) 0.05 0.07 0.09 Fw0 Fw0.2 Fw0.4 (g) V_fw0_ V_fw0_ V_fw0_ V_fw0_ V_fw0_ V_fw0_ V_fw0_ V_fw0_ V_fw0_. (g) V_fw0.2_ V_fw0.2_ V_fw0.2_ V_fw0.2_ V_fw0.2_ V_fw0.2_ V_fw0.2_ V_fw0.2_ V_fw0.2_ (g) V_fw0.4_ V_fw0.4_ V_fw0.4_ V_fw0.4_ V_fw0.4_ V_fw0.4_ V_fw0.4_ V_fw0.4_ V_fw0.4_ In the Table 4.2, showing the experimental design, V_fw0_ represents a collection of 150 signals at 500 rpm, feed at 0.05 mm/s and a depth of cut (DOC) of mm with a new tool. Similarly, V_fw0.2_ represents a collection of 150 signals at 500 rpm, 0.05mm feed and a depth of cut of mm with a tool tip whose flank wear is 0.2 mm. The vibration signatures were taken for zero flank wear, 0.2 mm flank wear and 0.4 mm flank wear. This experiment was repeated for three levels of depth of cut (DOC) viz., mm, mm and mm keeping the feed and speed constant. Further, the readings 57

were acquired for three levels of feed (0.05 mm/s, 0.07 mm/s and 0.09 mm/s) for 500 rpm. The above set of experiments was repeated for 700 rpm and 900 rpm. Thus, there are a total of 81 experiments. In each experiment, 150 vibration signatures were taken and the corresponding surface roughness Ra value was noted as shown in Table 4.3. The entire set of experiments (81 readings) was repeated for tool tips of the same specification but with titanium nitride coating. Similarly, the surface roughness data was measured for coated carbide tipped tool, which is given in Appendix A9. Table 4.3 Surface roughness data - measured Speed rpm Feed mm/s 0.05 500 0.07 0.09 0.05 700 0.07 0.09 0.05 900 0.07 0.09 DOC mm Surface Roughness (measured) Fw0 Fw0.2 Fw0.4 (microns) (microns) (microns) 2.8 2.95 3.49 2.63 2.66 2.86 2.15 2.55 2.65 3.04 3.21 3.56 2.43 2.52 2.69 2.19 2.27 2.64 3.22 3.45 3.92 2.71 2.92 3.22 2.04 2.16 2.51 3.1 3.12 3.21 2.26 2.28 2.38 1.36 1.45 1.81 3.22 2.81 3 2.65 2.61 2.64 2.3 2.43 2.25 2.4 2.59 2.64 2.31 1.8 1.65 2.3 1.32 1.45 2.29 3.11 3.42 1.61 1.5 1.42 1.74 1.44 1.47 1.97 3.64 1.87 2 1.64 1.32 7 1.13 1.07 2.6 2.18 1.69 8 1.68 1.54 1.11 1.78 1.12 58

4.3.6 SURFACE ROUGHNESS MEASUREMENT USING PROFILOMETER The most common measuring method of measuring the surface roughness is by the use of the stylus. It uses the same principle as phonograph. In this experiment Mitutoyo Surftest ST 301 model shown in Fig. 4.7 and Fig. 4.8 was used. This instrument uses a tracer or pickup incorporating a diamond stylus and a transducer. The stylus is dragged across the specimen. The working principle of the profilometer is shown in Fig. 4.9. The skid will follow the general shape of the object, filtering out any overall slant or large curve. Running the stylus tip across the work piece surface generates electrical signals corresponding to surface roughness. The electrical signals are amplified, converted from analog to digital, processed according to an algorithm, and displayed. The measurement has a fairly good resolution and a large range that satisfies the measurements of most manufactured surfaces. Fig. 4.7 Stylus of Mitutoyo surftest ST-301 59

Fig. 4.8 Mitutoyo Surf test ST 301 After taking the vibration signals, the surface roughness parameter R a values were taken at three randomly chosen places in machined surfaces. The average of three Ra values is used to represent the surface roughness of the machined surface. This measurement is given by the instrument. Surface roughness was measured offline and recorded for the 81 conditions of the experiment. Similarly, the same was measured and recorded for the shafts that were turned using the coated carbide tipped tool. 4.9 Working principle of a profilometer 4.4 CONCLUDING REMARKS The experimental setup and the various measuring devices, the accelerometer, FFT analyser and the profilometer for measuring the surface roughness have been discussed in the above chapter. The vibration analysis of signals obtained for the study using pattern recognition methods is discussed in the Chapter 5. The Chapter 6 deals with the analysis using wavelet features. Prediction of surface roughness using multiple linear regression, Support Vector Regression and RBF (Radial Basis Function Network) analysis are described in Chapter 7 and 8. 60