- Email: [email protected]
Influence of part material and sensor adjustment on the quality of digitised point-clouds using conoscopic holography
Precision Engineering 42 (2015) 42–52
Contents lists available at ScienceDirect
Precision Engineering journal homepage: www.elsevier.com/locate/precision
Influence of part material and sensor adjustment on the quality of digitised point-clouds using conoscopic holography ˜ ∗ , Pedro Fernández, J. Carlos Rico, Sabino Mateos David Blanco, Gonzalo Valino Department of Construction and Manufacturing Engineering, University of Oviedo, Campus de Gijón, 33203 Gijón, Spain
a r t i c l e
i n f o
Article history: Received 2 May 2014 Received in revised form 18 December 2014 Accepted 23 March 2015 Available online 2 April 2015 Keywords: Conoscopic holography Non-contact digitising Material influence Quality
a b s t r a c t Conoscopic holography is an interferometric measurement technique commonly used for non-contact surfaces digitising in quality assessment, in-process inspection and reverse engineering. Among other factors, accuracy of measurements provided by this technology is influenced by the surface optical properties. Parameters such as laser power (P) or frequency of acquisition (F) are commonly used to adjust the sensor until a quality indicator of the signal acquired (Signal-to-Noise-Ratio) is maximised. Nevertheless, measurements taken under this adjusting criterion does not necessarily ensure the most accurate results from a metrological point of view. Taking this into account, the present work proposes two additional indicators to analyse the influence of sensor setting parameters on the quality of digitised point-clouds for different metals and polymers. Digitising tests have been performed on flat specimens of each material by means of a conoscopic holography sensor integrated on a Coordinate Measuring Machine. In order to meet an optimal scanning of each material, the study provides a series of recommendations about adjustment of the sensor as well as the most suitable indicator to be used in each case. © 2015 Elsevier Inc. All rights reserved.
1. Introduction Non-contact measuring methods can efficiently capture dense point-clouds in terms of acquisition speed. For this reason they are commonly used for industrial surface digitising although most of them are generally less precise than contact methods. Their accuracy strongly depends on the interaction between the sensors used, the workpiece geometry and the optical properties of the surface material [1,2]. Optical sensors using different principles may react, in fact, in different ways to the optical behaviour of the surface to be measured. The influence of surface material on the performance of some widely diffused techniques, such as triangulation systems, has been deeply analysed in literature. However, the performance of other technologies has not been fully described yet. This is the case of conoscopic holography (CH). CH is an interferometric technique based on the double refractive property of birefringent crystals. It was first described by Sirat and Psaltis [3] and patented by Optimet Optical Metrology LTD. Malet and Sirat [4] stated that the performance of a conoscopic system can be described by the quartet of precision, depth of field, speed and transverse resolution. Furthermore, many advantages of CH in front of laser triangulation have been reported by Sirat
∗ Corresponding author. Tel.: +34 985 18 2442. ˜ E-mail address: [email protected] (G. Valino). http://dx.doi.org/10.1016/j.precisioneng.2015.03.008 0141-6359/© 2015 Elsevier Inc. All rights reserved.
et al. [5]: better accuracy and repeatability (up to 10 times for a given depth of field), good behaviour for a wide variety of materials (even for translucent ones) and steep slope surfaces up to 85◦ . Other practical characteristic is that a single conoscopic sensor can be combined with different lenses to be adapted to various depths of field (0.6 mm up to 120 mm) with accuracy from less than 1 m up to 60 m, respectively. Finally, being a collinear system allows for accessing to complex geometries such as holes or narrow cavities, by using simple devices for light redirection. These characteristics have led CH to be incorporated in a wide variety of fields, including quality assessment, reverse engineering and in-process inspection. The importance of accuracy becomes an essential target in industrial applications, such as those reviewed by Álvarez et al. [6]. This group has successfully applied CH for multiple industrial on-line applications, including sub-micrometric roughness measurements, on-line measurement of high production rate products, surface defect detection in steel at high temperatures and simultaneous inspection of external and internal shape of hollow cylindrical parts. Potential of CH as a valuable alternative to current wellestablished technologies (laser triangulation, range sensors or photogrammetry) has led researchers to work on analysing the performance of CH sensors under different scanning conditions. The ability of CH for digitising highly sloped surfaces was highlighted by Ko and Park [7] when they compared the capabilities of triangulation, conoscopic holography and interferometry methods
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
for accurate measuring of micro burr geometries formed in micro drilling. They proved that the conoscopic holography method was the most appropriate for measuring small scale burr (20 m height and 0.1 mm width). Similar results were accomplished by Toropov [8] who presented CH as an effective technology for the measurement of burrs above 10 m. Paviotti et al. [9] developed an experimental procedure for analysing the performance of CH sensors when digitising highly sloped surfaces. They proved that standard deviation error for the CH sensor remains stable for slope angles up to 60◦ , but it experiments a sharp increase in the range over 70◦ . Fernández et al. [10] analysed how the quality of a measurement by a CH sensor is affected by depth of field and configuration parameters. They demonstrated that quality of measurements worsens as position of the measured surface is furthest from the stand-off distance within the depth of field. Frequency of data acquisition (F) and laser power (P) are the parameters used for adjusting quality of the signal acquired by CH sensor so that two quality indicators (SNR and Total) meet acceptable values. SNR (Signal-to-Noise-Ratio) shows the relationship between the power value used for the measurement with respect to the whole signal power, including the signal noise. On the other hand, Total is proportional to the area limited by the signal envelope and it increases as signal intensity does. CH sensor performance is affected by surface properties, as it was highlighted by Lathrop et al. [11]. They applied a Conoprobe Mark 3 with a 200-mm objective lens for surface digitising of different types of biological tissues. Experiments were performed for each tissue by adjusting laser power (P) and acquisition frequency (F) to provide a good quality signal, with a SNR above 50%. They found repeatability quite stationary in the whole optical working range of the lens for each tissue, although different values were met for each material (about = 0.01 mm in one case and about = 0.15 mm in the other two). Therefore, it can be concluded that the nature of surface material (colour, roughness, texture) has an influence on the digitising quality and, consequently, different adjustment of tuning parameters (P and F) might be required for different materials. Zhu et al. [12] also used the SNR value as a quality indicator for digitising turbine blades. Low SNR values (below 70%) were filtered and rejected. Registered measurements revealed part defects which were very difficult to detect with current industrial practices. Lonardo and Bruzzone [13] applied CH to measure micro and macro geometries of Selective Laser Sintering (SLS) workpieces for zirconia and silica sands. They analysed the influence of SNR on three different roughness parameters and found that curves for both materials showed a stationary trend for SNR values between 20% and 80%. They also found a dependency of SNR with the angle between the optical direction (incidence direction) and the surface normal direction. It was reported that SNR remains stationary for both materials from 0◦ (normal direction) up to 75◦ . Therefore, there was a difference of 10◦ with regard to the maximum incidence angle of 85◦ reported by the manufacturer. In the same field, Lombardo et al. [14] compared roughness measurement with CH for two identical parts generated by rapid manufacturing techniques, one by stereolithography and the other one by Fused Deposition Modelling (FDM) in a 3D printer. The former showed slightly lower values for parameters Ra and Rz than the latter. From this review of works, it is commonly assumed that SNR is the suitable indicator for quality assessment when using a CH sensor for digitising. Nevertheless, some works reveal differences when other indicators are used for quality assessment. Apart from SNR, Lathrop et al. [11] used as quality indicator the standard deviation calculated from a collection of measurements on a single point. Their work suggests that different materials, digitised under adjusted P and F for a similar SNR value, could provide different values for the standard deviation indicator. In a similar way, Lonardo
43
and Bruzzone [13] showed that individual measurements of different points on a material surface present significant variations in the SNR value. Actually, SNR only reflects the quality of the optical signal, but has not relationship with any metrological parameter regarding the measured object. Therefore, it may be considered whether adjusting working conditions for the highest SNR value shall provide the most accurate measurements or not. Taking these considerations into account, the present work aims at showing that the adjustment of a CH sensor for digitising based exclusively on improving the quality of the acquired signal is not enough to guarantee the highest precision in geometrical measurements. Moreover, the adjustment of the sensor should consider another indicator based on the type of metrological magnitude to be verified. To demonstrate this assertion, numerous digitising tests were performed over flat samples of different materials by means of a Conoprobe Mark III CH sensor. Experiments were carried out under all the feasible configurations of the CH sensor (combinations of frequency F and power level P) whose yield satisfactory values of signal quality according to the recommendations by the sensor manufacturer. Then, considering that all the test specimens were flat, an additional parameter related to flatness was chosen as metrological indicator. Thus, values of flatness measured by the CH sensor under each configuration were compared to reference values measured by a touch probe. The analysis of this metrological indicator is discussed in relationship to the signal quality. Finally, different recommendations for adjustment of the CH sensor are provided, depending on the type of material and the metrological requirements considered. 2. Experimental method 2.1. Equipment The tests described in this study have been performed using the conoscopic sensor Optimet Conoprobe Mark III (see Table 1). This sensor was equipped with a lens of 50 mm focal length, with a depth of field of 8 mm. The visible light source is a Class II laser diode which wavelength is 655 nm. This is a point-type sensor whose readings provide the value of the distance () from the transmitter to the projection of the laser beam on the material surface (spot). To achieve a complete digitised surface, a relative displacement between the sensor and the surface is required, which provides a virtual representation of the surface by means of an ordered array of acquired points. In order to perform accurate sweeps of the surface,
Table 1 Characteristics of the Conoprobe Mark III sensor provided by the manufacturer. Property
Focal Lens (mm)
Value
Dimensions Weight Max. measurement rate, F Power level, Pa Linearity Wavelength Spot size X–Y Working range (WR) Stand-off Precisionb Reproducibility 1c Angle measurementd
50 50 – – – – 50 50 50 50 50 50
79 × 167 × 57 mm 750 g 3000 Hz 0–63 0.1% 655 nm 45–30 m 8 mm 42 mm <6 m <1 m 170◦
a
Maximum power level (63) is equivalent to 1 mW. Twice the maximum error when measuring a step located at 5–6 different positions throughout the WR. c As measured on a flat diffusive metallic surface, average of 5 scans offset in “y” direction. Minimum sampling step ½ of the spot size, average over 200 points in each scan. Measured over a step higher than 50% of the working range. d Measurable points up to ±85◦ from normal incidence angle. b
44
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
the CH sensor has been integrated into a DEA Swift Coordinate Measuring Machine (CMM) as shown in Fig. 1. The Maximum Permissible Error for linear dimensions (MPEE ) and Maximum Permissible Error for probing (MPEP ) of this CMM were certified according to ISO 10360-2:2001 as follows:(1)MPEE = 4 +4 · 10−3 · L [m], being L in mm(2)MPEP = 4 [m] The CMM is operated by means of the measurement and control software PC-DMIS. The coordinates of each digitised point are directly provided by the CMM measuring system, after performing a calibration procedure of the CH sensor as described by Fernández et al. [15] and inspired on the work by Smith and Zheng [16]. 2.2. Test materials
Fig. 1. General view of the integrated CH sensor on the CMM.
Twelve different materials have been evaluated in this work as representative of those used in industrial manufacturing. They have been classified into two groups (see Fig. 2):
Fig. 2. Groups of materials and general view of the test specimens.
D. Blanco et al. / Precision Engineering 42 (2015) 42–52 Table 2 Reference values of flatness and roughness of the specimens tested. Group
Material
Flatness (mm)
Roughness (m)
Averagea
St. deviationa
Rab
Gradec
G1
M01
0.0202
0.000173
6.717
N9
G2
M02 M03 M04 M05 M06
0.0194 0.0242 0.0150 0.0127 0.0221
0.000404 0.000737 0.000153 0.001060 0.000346
6.306 9.423 6.318 6.391 10.324
N9 N9 N9 N9 N9
M07 M08 M09 M10 M11 M12
0.0198 0.0165 0.0188 0.0096 0.0174 0.0230
0.000600 0.000058 0.000643 0.001852 0.000473 0.001069
11.509 10.481 10.908 8.492 11.579 8.835
N9 N9 N9 N9 N9 N9
a b c
Over 5 repetitions. According to ISO4287. According to ISO/R468 and ISO2632-1.2.
• Group G1: comprises five metals such as medium carbon steel (DIN C45k), low alloy steel (DIN CrMo4), stainless steel (AISI 316) and aluminium alloys (CuAl10 and EN-AW 6082). • Group G2: comprises seven types of plastics such as PET (white and green), PA (white and black), PTFE, PU and PP. Each material has been evaluated by digitising tests on flat surfaces of 31.75 mm of diameter. Specimens were manufactured by face milling under identical cutting conditions so that a uniform roughness was achieved in order to avoid its influence on the scanning results. The actual values of roughness were measured by means of a Leica DCM 3D microscopy according to ISO 4287 standard and they are shown in Table 2. All the values of Ra belong to grade N9 according to ISO 2632-1.2. 2.3. Sensor parameters The adjustment of the CH sensor is described by the manufacturer in the operator manual [17]. The main recommendations regarding this topic are summarised in this section. There are two main setting parameters for the Conoprobe Mark III sensor: • Working Frequency (F) represents the data acquisition rate and can be set up to a maximum of 3000 Hz. • Power Level (P) represents the value of the laser beam energy used and can be set up in a range from 0 to 63, where 63 is equivalent to 1 mW of power.
45
The adjustment procedure consists on working with the highest F possible, since measurement error can be minimised by better use of averaging filters. Then, P should be adjusted so that a proper amount of energy reaches the sensor. In general, there is a direct relationship between F and the corresponding required P. This means that working at high frequency will require a high laser power level, whereas working at low frequency will require a lower laser power level. The optimum combination of F and P should be set by means of two indicators of the signal quality: • SNR (Signal-to-Noise Ratio), which represents the percentage of the signal power received by the sensor with respect to the total power used including the signal noise. SNR may range from 0% to 100% and it is commonly assumed that the higher the SNR, the higher the accuracy of measurement. Reliable values of SNR should be above 50%. • Total, which is proportional to the area limited by the signal envelope, increases as signal intensity does. Values of Total between 1200 and 18,000 may yield accurate results but recommended values range from 2000 to 16,000. 2.4. Experimental procedure As an optic technique, CH is affected by factors such as the surface slope [9] and its position within depth of field [10]. In order to avoid these influencing factors, the test specimens were installed in a test bench (see Fig. 3) taking care that the test surfaces were located parallel to the XY plane of the CMM at distance equal to the sensor stand-off. Once the surface was precisely orientated, the test surface of each specimen was digitised by a touch probe on the CMM in order to provide a reference value of flatness (ft ), which will be one of the quality indicators used in this study. With this probe, 15 points distributed within a square mesh of 10 × 10 mm were captured every 2.5 mm along X direction and every 5 mm along Y direction. This procedure was repeated 5 times on each specimen. The average and standard deviation values were included in Table 2. In a similar way, the same points were scanned by the CH sensor taking 500 measurements on each point. The resulting average was considered as the actual value of distance (). The reliability of this value is provided by the standard deviation p . This routine was repeated under each combination of F and P. The range of power P was varied from 1 to 63 at increments of 1, whereas frequency F was varied from 500 to 3000 Hz at increments of 500 Hz. Therefore, 378 combinations of P and F were tested. Since 12 types of materials were considered in the study, an amount of 4536 tests were finally performed.
Fig. 3. General view of the test bench and detailed view of a test specimen being scanned.
46
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
Fig. 4. Distribution of SNR values within the RA for materials in G1.
According to the work carried out by Fernández et al. [15], tests were conducted at least 60 min after switching on the laser in order to reach thermal stabilisation of the sensor and avoid thermal drift.
• Measurement Dispersion ( ): it is calculated by expression in Eq. (1) as the average value of p when considering 15 points within the point-cloud. In this case, the lower , the less dispersion.
15
2.5. Filtering process Data acquired in the tests have been filtered in order to remove from the study those combinations of F and P which yield low quality signals, as well as those in which a valid register is not obtained. Taking into account the manufacturer recommendations aforementioned in Section 2.3, the filtering process consisted in rejecting the combinations of F and P with SNR lower than 50% and parameter Total out of the range 2000 to 16,000. Those combinations of F and P that passed this filtering process provided a reliable signal and defined a Reliability Area (RA). 3. Analysis of results 3.1. Quality indicators Three quality indicators have been used in this work to perform the analysis of results: • Signal-to-Noise-Ratio (SNR): this indicator provides information about the quality of the signal used by the sensor to process distance measurements.
=
p=1 p 15
(1)
• Flatness Deviation (ıf ): it shows the difference between the flatness value measured by CH sensor (fc ) with respect to the reference value measured by a standard touch probe (ft . Therefore, it can be considered as an indicator of the reliability of the CH sensor for a flat surface reconstruction. It is calculated as in Eq. (2), ıf = |fc − ft |
(2)
The lower value of ıf , the more coincidence between the reconstructed surface and the reference one. 3.2. Experimental results In order to facilitate the analysis, results have been organised according to the material groups included in Fig. 2. Distribution of the quality indicators SNR, and ıf are described in this section by means of graphs for different combinations of F and P within the RA. A summarising table is provided for each group of materials, which includes the maximum, minimum, average value and standard deviation of each quality indicator within the RA. The average value shows the most probable value of measurement while the standard deviation gives an idea of the uniformity of each
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
47
Fig. 5. Distribution of values within the RA for materials in G1.
quality indicator. Correlation coefficient (R2 ) has been also included between combinations of the three indicators (SNR – ; –ıf ; SNR – ıf ). 3.2.1. Metals (G1) Figs. 4–6 show the value of the three quality indicators (SNR, and ıf ) obtained within the RA for materials in the group G1: • In general, there exist lots of valid combinations of F and P. • For M03 and M04, the amplitude of the valid ranges of power is almost independent of the value of frequency used. • For M01, M02 and M05 the valid ranges of power become wider as the value of frequency increases. The general warm colour represented within the RA on the SNR graphs shows the high value of this indicator (SNR > 70%) and therefore, the good quality of the signal acquired by the CH sensor for all the combinations of F and P and all the materials within this group (see Fig. 4 and Table 3). Moreover, the uniformity of colour shows the low variability of the indicator in most part of the RA (0.122% ≤ SNR ≤ 2.240%). With regard to , the cold colour of the graphs shows the low value of this indicator (see Fig. 5). Minimum values lower than 0.240 m are met, which are clearly lower than reproducibility of the sensor (1 < 1 m). Besides, variability of the indicator is low ( ≤ 0.308 m) which shows a good measurement quality in almost all the RA.
With respect to ıf , the material M05 presents the lowest values (3.116 m ≤ ıf ≤ 9.352 m) and the lowest variability (ıf =
0.754 m). The rest of materials have got graphs with warmer colours and, therefore, high values of the indicator which are far from the optimum (see Fig. 6). Moreover, these materials also have high variability (ıf > 1.2 m). If correlations among the different quality indicators are analysed, it can be noticed that they are very low (see Table 3), except for M03 (R2 [ − ıf ] = 0.805) and M05 (R2 [SNR − ıf ] = 0.845). Based on the variability observed for the three indicators, it can be said that the election of one or another combination of F and P within the RA does only affect significantly to ıf . Therefore, when a geometrical feature such as flatness has to be measured for this type of materials, a combination of F and P which minimises this indicator should be selected.
3.2.2. Plastics (G2) Before discussing the results for this group, it should be noted that M12 has been excluded from the analysis since signal acquired by the CH sensor does not pass the filters mentioned in Section 2.5. Figs. 7–9 show the value of the three quality indicators (SNR, and ıf ) obtained within the RA for materials in the group G2: • For M06, M07, M08 and M10 the valid ranges of power are very small (from 7 to 18) and independent of the frequency value used.
48
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
Fig. 6. Distribution of ıf values within the RA for materials in G1.
• For M09 and M11 the valid ranges of power become wider as the value of frequency increases. The graphs of SNR for M09 and M11 present a uniform warm colour (see Fig. 7). This proves the high quality of the signal acquired by the CH sensor for these materials (SNR > 78%). Furthermore, low variability is observed since the standard deviation is low ( SNR ≤ 0.140%) (see Table 4). In contrast, the SNR graphs for the
rest of materials include colder colours which are related to signals of lower quality (50.151% ≤ SNR ≤ 76.331%) as well as to high variability ( SNR > 2.2%). In relation to the indicator , it is again observed that materials M09 and M11 show the coldest colours within the RA. The indicator is lower for high values of power and low variability is met ( = 0.279 m for M09 and = 0.223 m for M11). Variability for M06 is also low but this value is not significant since
Table 3 Value of the quality indicators within the RA for materials in the group G1. M01
M02
M03
M04
M05
SNR (%)
Max. Value Min. Value SNR SNR
79.492 77.833 79.065 0.380
80.320 79.602 80.149 0.122
79.472 69.147 77.713 2.240
79.098 72.870 78.218 1.106
80.144 77.113 79.323 0.793
(m)
Max. Value Min. Value
0.628 0.240 0.377 0.074
0.925 0.151 0.307 0.146
1.762 0.217 0.571 0.308
1.006 0.169 0.433 0.185
1.189 0.168 0.304 0.152
ıf (m)
Max. Value Min. Value
29.938 9.655
17.575 12.059
20.503 4.881
26.928 10.496
9.352 3.116
ıf ıf
19.318 5.185
14.903 1.203
13.220 3.881
20.778 4.304
5.740 0.754
R2 [SNR − ]
0.521
0.485
0.549
0.538
0.100
R [ − ıf ]
0.223
0.265
0.805
0.381
0.122
R2 [SNR − ıf ]
0.644
0.343
0.335
0.400
0.845
2
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
49
Fig. 7. Distribution of SNR values within the RA for materials in G2.
there are few values within the RA. On the contrary, value of the indicator is high for the rest of materials, such as the case of M07 where the maximum ( = 9.991 m) is met out of the represented range in Fig. 8. Moreover, variability is also notable, with extreme values of = 1.890 m for M07 and = 2.218 m for M08.
With respect to ıf , materials M06 and M08 present the lowest values (1.106 m ≤ ıf ≤ 4.166 m). and the lowest variability (ıf ≤ 0.349 m). The rest of materials meet high values of the
indicator and high variability (ıf > 1.5 m). For this indicator material M07 scores very high values which are out of the represented range in Fig. 9.
Table 4 Value of the quality indicators within the RA for materials in the group G2. M06
M07
M08
M09
M10
M11
SNR (%)
Max. value Min. value SNR SNR
63.296 51.032 58.510 3.701
60.079 50.906 57.348 2.287
71.507 50.151 67.024 5.368
79.121 78.377 78.838 0.140
76.331 62.149 70.983 4.724
80.445 79.589 80.268 0.093
(m)
Max. value Min. value
3.192 1.236 1.928 0.657
9.991 2.186 3.953 1.890
9.100 0.809 2.590 2.218
2.136 0.416 0.694 0.279
6.802 0.527 2.097 1.724
1.654 0.226 0.427 0.223
ıf (m)
Max. value Min. value
4.166 2.948
41.217 35.094
1.983 1.106
33.151 13.628
20.956 15.115
12.266 3.375
ıf ıf
3.571 0.349
37.895 1.507
1.552 0.254
18.545 4.022
17.465 1.658
8.216 2.050
R2 [SNR − ]
0.786
0.799
0.873
0.263
0.811
0.356
R [ − ıf ]
0.321
0.430
0.428
0.285
0.674
0.358
R2 [SNR − ıf ]
0.082
0.508
0.536
0.615
0.688
0.002
2
50
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
Fig. 8. Distribution of values within the RA for materials in G2.
Since correlations R2 [SNR − ıf ] and R2 [ − ıf ] are very low it is not possible to set a combination of F and P which satisfy simultaneously the three indicators. Nevertheless, the correlation between SNR and is quite high for materials M06, M07, M08 and M10 (R2 [SNR − ] > 0.78) but it is not significant since the extension of the RA is very small. Similarly to the group of metals (G1), it can be stated that the election of one or another combination of F and P within the RA for materials M09 and M11 does only affect significantly to ıf . Therefore, when a geometrical feature such as flatness has to be measured for these two materials, a combination of F and P which minimises this indicator should be selected. For materials M06, M08 and M10 the election of a combination of F and P should be performed in order to improve the SNR and , which have a good correlation, whereas the indicator ıf will not be practically altered. The behaviour of the three quality indicators for material M07 is particularly different, since this is slightly translucent. The SNR is low, is high and ıf is very high. Although the sensor is capable of acquiring points, these have very low quality and it is not feasible to perform measurements of the geometry within the usual tolerance limits. 3.3. General discussion Values of the quality indicators used in the tests (SNR, and ıf ) show differences among one materials and others, even within the same group. However, the test materials can be grouped according to the values of the indicators, as shown in Table 5.
Table 5 Classification of materials according to the value of indicators. Group
Type A
Type B
G01 G02
SNR > 75% and SNR < 1% M01, M02, M03, M04, M05 M09, M11
SNR < 75% and SNR > 1% – M06, M07, M08, M10
G01 G02
< 1 m and < 1 m M01, M02, M03, M04, M05 M09, M11
> 1 m and > 1 m – M06 a , M07, M08, M10
ıf > 10 m and ıf > 1 m
ıf < 10 m and ıf < 1 m
M01, M02, M03, M04 M07, M09, M10, M11a
M05 M06, M08
G01 G02
a Materials are included into this Type although they do not meet strictly both conditions.
Type A materials are those whose indicators SNR and reach the best values whereas the indicator ıf meets the worst results with great variability. In these cases, it would be advisable to perform the adjustment of the conoscopic sensor by choosing a combination of F and P which optimises the metrological indicator ıf . From a practical point of view, the search for this optimal combination requires to carry out previous tests of the material to be scanned, using a similar procedure as it is proposed along this work. Table 6 shows the recommended values of F and P for this type of materials. On the other hand, Type B materials are those whose metrological indicator ıf reaches the best values and slight variability
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
51
Fig. 9. Distribution of ıf values within the RA for materials in G2.
whereas indicators SNR and are far from an optimal behaviour. In these cases, it would be advisable to perform the adjustment of the conoscopic sensor by choosing a combination of F and P which optimises indicators SNR and . Since there exists a certain correlation between these two indicators, the optimal combination should be directly found by tuning F and P until value of the SNR rises to the maximum. Among these types of materials, the behaviour of M05, M06, M07, M10 and M11 does not match strictly to the general rule described. The three indicators of material M05 show a good behaviour which is far from the rest of the test materials. This is because in this case there is a greater correlation between indicators SNR and ıf . For this reason, it is suggested for this material to adjust F and P for maximising the value of the indicator SNR, as it is recommended for the materials of Type B.
With regard to M06, variability of the indicator for this material is low but this result is not actually significant since the extension of the RA is very small. Furthermore, since behaviour of the rest of indicators for this material is similar to that described for Type B materials, sensor adjustment should be carried out in a same way. Regarding materials M07 and M10, it is observed that the three indicators have a bad behaviour. Low quality of acquired points can be related to the materials optical characteristics, since M07 is translucent and M10 is porous. In consequence, it is advisable not to use a CH sensor for digitising these materials, especially if there exists any tight geometric specification on the product. Finally, although M11 presents low values for ıf , it has got high variability and the behaviour of the rest of indicators is similar to that of Type A materials. Then, it is reasonable to treat this case in the same way as materials in this group.
Table 6 Recommended combinations of P and F for adjustment of the CH sensor according to the test material. Group
Type A
G01 P (1–63) F (Hz)
M01 13 500
M02 52 3000
M03 41 500
M04 60 3000
M05 11 500
P (1–63) F (Hz)
M07 – –
M09 60 3000
M10 – –
M11 12 500
M06 10 2000
G02
Type B
M08 15 2500
52
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
4. Conclusions This work demonstrates that the adjustment of a CH sensor for digitising based exclusively on improving the quality of the acquired signal is not enough to guarantee the highest precision in geometrical measurements. Moreover, the adjustment of the sensor should consider another indicator based on the type of metrological magnitude to be verified. With this purpose, numerous digitising tests were performed by means of a Conoprobe Mark III CH sensor over flat samples of 12 different materials grouped into metals (G1) and plastics (G2). Each material was scanned under different combinations of frequency F and power P and filtered based on recommendations by the manufacturer in order to provide a Reliability Area (RA) where the quality of signal was guaranteed. Three quality indicators were calculated for each combination of F and P within the RA: Signal-to-Noise Ratio (SNR), measurement dispersion ( ) and flatness deviation of the point-cloud (ıf ). Several conclusions can be highlighted from the study, which are summarised as follows: • Extension of RA varies greatly for each material, even among materials included in the same group. Consequently, it is not possible to state a general rule for the limits of the RA for different materials. • In general, results show a very dissimilar behaviour of the three quality indicators considered (SNR, , ıf ) so it is not possible to carry out a simultaneous optimisation. • Based on the tests performed, it has been observed that those materials whose indicators SNR and reach the best values have got the worst results for the indicator ıf (Type A). On the contrary, materials with better results for the metrological indicator ıf are those whose indicators SNR and reach the worst values (Type B). • It is advisable for Type A materials to perform the adjustment of the conoscopic sensor by selecting a combination of F and P which optimises the metrological indicator ıf . In the case of materials of Type B, it is suggested to select a combination of F and P which optimises the indicators SNR and . Although reviewed studies consider the maximum value of the SNR as the main digitising quality indicator, the present work reveals that not always there is a link between signal quality and measurement quality. Actually, it was observed that an optimal sensor configuration which maximises signal quality does not necessarily indicate a minimisation of the geometrical error ıf . Therefore, when high precision measurements are required, a geometrical type indicator should be used to ensure minimum errors in materials of Type A. The main drawback consists in the
necessity of performing previous tests before digitising in order to adjust the parameters of the CH sensor for such a material. Acknowledgments The work described is part of a research supported by the Spanish Ministry of Economy and Competitiveness and FEDER (DPI2012-30987) and the Regional Ministry of Economy and Employment of the Principality of Asturias (SV-PA-13-ECOEMP15). References [1] Li F, Longstaff A, Fletcher S, Myers A. Integrated tactile and optical measuring systems in three dimensional metrology. In: Proceedings of the Queen’s Diamond Jubilee, CEARC’12. 2012. p. 1–6. [2] Shiou FJ, Chen MJ. Intermittent process hybrid measurement system on the machining centre. Int J Prod Res 2003;41(18):4403–27. [3] Sirat G, Psaltis D. Conoscopic holography. Opt Lett 1985;10(1):4–6. [4] Malet Y, Sirat GY. Conoscopic holography application: multipurpose rangefinders. J Opt 1998;29(3):183–7. [5] Sirat GY, Paz F, Agronik G, Wilner K. Conoscopic systems and conoscopic holography [Internet]. Israel: Laboratory for CAD & Lifecycle Engineering. Faculty of Mechanical Engineering, Technion-Israel Institute of Technology; 2012 [cited 2012 November 19]. Available from: http://mecadserv1. technion.ac.il/public html/IK05/Sirat 9375.pdf [6] Álvarez I, Enguita JM, Frade M, Marina J, Ojea G. On-line metrology with conoscopic holography: beyond triangulation. Sensors 2009;9(9):7021–37. [7] Ko SL, Park SW. Development of an effective measurement system for burr geometry. Proc Inst Mech Eng Part B: J Eng Manuf 2006;220(4):507–12. [8] Toropov A. An effective visualization and analysis method for edge measurement. In: Gervasi O, Gavrilova M, editors. Computational science and its applications – ICCSA 2007, Lecture notes in computer science – LNCS 4706, Part II. Berlin Heidelberg: Springer-Verlag; 2007. p. 941–50. [9] Paviotti A, Carmignato S, Voltan A, Laurenti N, Cortelazzo GM. Estimating angle-dependent systematic error and measurement uncertainty for a conoscopic holography measurement system. In: Proceeding SPIE 7239, ThreeDimensional Imaging Metrology, 72390Z. 2009. ˜ G, Mateos S. Influence of surface [10] Fernández P, Blanco D, Rico JC, Valino position along the working range of conoscopic holography sensors on dimensional verification of AISI 316 wire EDM machined surfaces. Sensors 2014;14(3):4495–512. [11] Lathrop RA, Hackworth DM, Webster III RJ. Minimally invasive holographic surface scanning for soft-tissue image registration. IEEE Trans Biomed Eng 2010;57(6):1497–506. [12] Zhu L, Barhak J, Srivatsan V, Katz R. Efficient registration for precision inspection of free-form surfaces. Int J Adv Manuf Technol 2007;32(5–6):505–15. [13] Lonardo PM, Bruzzone AA. Measurement and topography characterization of surfaces produced by selective laser sintering. CIRP Ann – Manuf Technol 2000;49(1):427–30. [14] Lombardo E, Martorelli M, Nigrelli V. Non-contact roughness measurement in rapid prototypes by conoscopic holography. In: XII ADM International Conference, B2. 2001. p. 11–8. ˜ G, Hoang H, Suárez CM, Mateos S. Integration of [15] Fernández P, Blanco D, Valino a conoscopic holography sensor on a CMM. In: Marcos M, Salguero J, editors. AIP Conference Proceedings 1431. 2012. p. 225–32. [16] Smith KB, Zheng YF. Point laser triangulation probe calibration for coordinate metrology. J Manuf Sci Eng 2000;122(3):582–6. [17] ConoProbe MKIII OEM Manual. Optimet manual P/N 3J06007 [Internet]. Optimet (Israel); [cited 2013 April 5]. Available from: http://www.optimet.com
Contents lists available at ScienceDirect
Precision Engineering journal homepage: www.elsevier.com/locate/precision
Influence of part material and sensor adjustment on the quality of digitised point-clouds using conoscopic holography ˜ ∗ , Pedro Fernández, J. Carlos Rico, Sabino Mateos David Blanco, Gonzalo Valino Department of Construction and Manufacturing Engineering, University of Oviedo, Campus de Gijón, 33203 Gijón, Spain
a r t i c l e
i n f o
Article history: Received 2 May 2014 Received in revised form 18 December 2014 Accepted 23 March 2015 Available online 2 April 2015 Keywords: Conoscopic holography Non-contact digitising Material influence Quality
a b s t r a c t Conoscopic holography is an interferometric measurement technique commonly used for non-contact surfaces digitising in quality assessment, in-process inspection and reverse engineering. Among other factors, accuracy of measurements provided by this technology is influenced by the surface optical properties. Parameters such as laser power (P) or frequency of acquisition (F) are commonly used to adjust the sensor until a quality indicator of the signal acquired (Signal-to-Noise-Ratio) is maximised. Nevertheless, measurements taken under this adjusting criterion does not necessarily ensure the most accurate results from a metrological point of view. Taking this into account, the present work proposes two additional indicators to analyse the influence of sensor setting parameters on the quality of digitised point-clouds for different metals and polymers. Digitising tests have been performed on flat specimens of each material by means of a conoscopic holography sensor integrated on a Coordinate Measuring Machine. In order to meet an optimal scanning of each material, the study provides a series of recommendations about adjustment of the sensor as well as the most suitable indicator to be used in each case. © 2015 Elsevier Inc. All rights reserved.
1. Introduction Non-contact measuring methods can efficiently capture dense point-clouds in terms of acquisition speed. For this reason they are commonly used for industrial surface digitising although most of them are generally less precise than contact methods. Their accuracy strongly depends on the interaction between the sensors used, the workpiece geometry and the optical properties of the surface material [1,2]. Optical sensors using different principles may react, in fact, in different ways to the optical behaviour of the surface to be measured. The influence of surface material on the performance of some widely diffused techniques, such as triangulation systems, has been deeply analysed in literature. However, the performance of other technologies has not been fully described yet. This is the case of conoscopic holography (CH). CH is an interferometric technique based on the double refractive property of birefringent crystals. It was first described by Sirat and Psaltis [3] and patented by Optimet Optical Metrology LTD. Malet and Sirat [4] stated that the performance of a conoscopic system can be described by the quartet of precision, depth of field, speed and transverse resolution. Furthermore, many advantages of CH in front of laser triangulation have been reported by Sirat
∗ Corresponding author. Tel.: +34 985 18 2442. ˜ E-mail address: [email protected] (G. Valino). http://dx.doi.org/10.1016/j.precisioneng.2015.03.008 0141-6359/© 2015 Elsevier Inc. All rights reserved.
et al. [5]: better accuracy and repeatability (up to 10 times for a given depth of field), good behaviour for a wide variety of materials (even for translucent ones) and steep slope surfaces up to 85◦ . Other practical characteristic is that a single conoscopic sensor can be combined with different lenses to be adapted to various depths of field (0.6 mm up to 120 mm) with accuracy from less than 1 m up to 60 m, respectively. Finally, being a collinear system allows for accessing to complex geometries such as holes or narrow cavities, by using simple devices for light redirection. These characteristics have led CH to be incorporated in a wide variety of fields, including quality assessment, reverse engineering and in-process inspection. The importance of accuracy becomes an essential target in industrial applications, such as those reviewed by Álvarez et al. [6]. This group has successfully applied CH for multiple industrial on-line applications, including sub-micrometric roughness measurements, on-line measurement of high production rate products, surface defect detection in steel at high temperatures and simultaneous inspection of external and internal shape of hollow cylindrical parts. Potential of CH as a valuable alternative to current wellestablished technologies (laser triangulation, range sensors or photogrammetry) has led researchers to work on analysing the performance of CH sensors under different scanning conditions. The ability of CH for digitising highly sloped surfaces was highlighted by Ko and Park [7] when they compared the capabilities of triangulation, conoscopic holography and interferometry methods
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
for accurate measuring of micro burr geometries formed in micro drilling. They proved that the conoscopic holography method was the most appropriate for measuring small scale burr (20 m height and 0.1 mm width). Similar results were accomplished by Toropov [8] who presented CH as an effective technology for the measurement of burrs above 10 m. Paviotti et al. [9] developed an experimental procedure for analysing the performance of CH sensors when digitising highly sloped surfaces. They proved that standard deviation error for the CH sensor remains stable for slope angles up to 60◦ , but it experiments a sharp increase in the range over 70◦ . Fernández et al. [10] analysed how the quality of a measurement by a CH sensor is affected by depth of field and configuration parameters. They demonstrated that quality of measurements worsens as position of the measured surface is furthest from the stand-off distance within the depth of field. Frequency of data acquisition (F) and laser power (P) are the parameters used for adjusting quality of the signal acquired by CH sensor so that two quality indicators (SNR and Total) meet acceptable values. SNR (Signal-to-Noise-Ratio) shows the relationship between the power value used for the measurement with respect to the whole signal power, including the signal noise. On the other hand, Total is proportional to the area limited by the signal envelope and it increases as signal intensity does. CH sensor performance is affected by surface properties, as it was highlighted by Lathrop et al. [11]. They applied a Conoprobe Mark 3 with a 200-mm objective lens for surface digitising of different types of biological tissues. Experiments were performed for each tissue by adjusting laser power (P) and acquisition frequency (F) to provide a good quality signal, with a SNR above 50%. They found repeatability quite stationary in the whole optical working range of the lens for each tissue, although different values were met for each material (about = 0.01 mm in one case and about = 0.15 mm in the other two). Therefore, it can be concluded that the nature of surface material (colour, roughness, texture) has an influence on the digitising quality and, consequently, different adjustment of tuning parameters (P and F) might be required for different materials. Zhu et al. [12] also used the SNR value as a quality indicator for digitising turbine blades. Low SNR values (below 70%) were filtered and rejected. Registered measurements revealed part defects which were very difficult to detect with current industrial practices. Lonardo and Bruzzone [13] applied CH to measure micro and macro geometries of Selective Laser Sintering (SLS) workpieces for zirconia and silica sands. They analysed the influence of SNR on three different roughness parameters and found that curves for both materials showed a stationary trend for SNR values between 20% and 80%. They also found a dependency of SNR with the angle between the optical direction (incidence direction) and the surface normal direction. It was reported that SNR remains stationary for both materials from 0◦ (normal direction) up to 75◦ . Therefore, there was a difference of 10◦ with regard to the maximum incidence angle of 85◦ reported by the manufacturer. In the same field, Lombardo et al. [14] compared roughness measurement with CH for two identical parts generated by rapid manufacturing techniques, one by stereolithography and the other one by Fused Deposition Modelling (FDM) in a 3D printer. The former showed slightly lower values for parameters Ra and Rz than the latter. From this review of works, it is commonly assumed that SNR is the suitable indicator for quality assessment when using a CH sensor for digitising. Nevertheless, some works reveal differences when other indicators are used for quality assessment. Apart from SNR, Lathrop et al. [11] used as quality indicator the standard deviation calculated from a collection of measurements on a single point. Their work suggests that different materials, digitised under adjusted P and F for a similar SNR value, could provide different values for the standard deviation indicator. In a similar way, Lonardo
43
and Bruzzone [13] showed that individual measurements of different points on a material surface present significant variations in the SNR value. Actually, SNR only reflects the quality of the optical signal, but has not relationship with any metrological parameter regarding the measured object. Therefore, it may be considered whether adjusting working conditions for the highest SNR value shall provide the most accurate measurements or not. Taking these considerations into account, the present work aims at showing that the adjustment of a CH sensor for digitising based exclusively on improving the quality of the acquired signal is not enough to guarantee the highest precision in geometrical measurements. Moreover, the adjustment of the sensor should consider another indicator based on the type of metrological magnitude to be verified. To demonstrate this assertion, numerous digitising tests were performed over flat samples of different materials by means of a Conoprobe Mark III CH sensor. Experiments were carried out under all the feasible configurations of the CH sensor (combinations of frequency F and power level P) whose yield satisfactory values of signal quality according to the recommendations by the sensor manufacturer. Then, considering that all the test specimens were flat, an additional parameter related to flatness was chosen as metrological indicator. Thus, values of flatness measured by the CH sensor under each configuration were compared to reference values measured by a touch probe. The analysis of this metrological indicator is discussed in relationship to the signal quality. Finally, different recommendations for adjustment of the CH sensor are provided, depending on the type of material and the metrological requirements considered. 2. Experimental method 2.1. Equipment The tests described in this study have been performed using the conoscopic sensor Optimet Conoprobe Mark III (see Table 1). This sensor was equipped with a lens of 50 mm focal length, with a depth of field of 8 mm. The visible light source is a Class II laser diode which wavelength is 655 nm. This is a point-type sensor whose readings provide the value of the distance () from the transmitter to the projection of the laser beam on the material surface (spot). To achieve a complete digitised surface, a relative displacement between the sensor and the surface is required, which provides a virtual representation of the surface by means of an ordered array of acquired points. In order to perform accurate sweeps of the surface,
Table 1 Characteristics of the Conoprobe Mark III sensor provided by the manufacturer. Property
Focal Lens (mm)
Value
Dimensions Weight Max. measurement rate, F Power level, Pa Linearity Wavelength Spot size X–Y Working range (WR) Stand-off Precisionb Reproducibility 1c Angle measurementd
50 50 – – – – 50 50 50 50 50 50
79 × 167 × 57 mm 750 g 3000 Hz 0–63 0.1% 655 nm 45–30 m 8 mm 42 mm <6 m <1 m 170◦
a
Maximum power level (63) is equivalent to 1 mW. Twice the maximum error when measuring a step located at 5–6 different positions throughout the WR. c As measured on a flat diffusive metallic surface, average of 5 scans offset in “y” direction. Minimum sampling step ½ of the spot size, average over 200 points in each scan. Measured over a step higher than 50% of the working range. d Measurable points up to ±85◦ from normal incidence angle. b
44
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
the CH sensor has been integrated into a DEA Swift Coordinate Measuring Machine (CMM) as shown in Fig. 1. The Maximum Permissible Error for linear dimensions (MPEE ) and Maximum Permissible Error for probing (MPEP ) of this CMM were certified according to ISO 10360-2:2001 as follows:(1)MPEE = 4 +4 · 10−3 · L [m], being L in mm(2)MPEP = 4 [m] The CMM is operated by means of the measurement and control software PC-DMIS. The coordinates of each digitised point are directly provided by the CMM measuring system, after performing a calibration procedure of the CH sensor as described by Fernández et al. [15] and inspired on the work by Smith and Zheng [16]. 2.2. Test materials
Fig. 1. General view of the integrated CH sensor on the CMM.
Twelve different materials have been evaluated in this work as representative of those used in industrial manufacturing. They have been classified into two groups (see Fig. 2):
Fig. 2. Groups of materials and general view of the test specimens.
D. Blanco et al. / Precision Engineering 42 (2015) 42–52 Table 2 Reference values of flatness and roughness of the specimens tested. Group
Material
Flatness (mm)
Roughness (m)
Averagea
St. deviationa
Rab
Gradec
G1
M01
0.0202
0.000173
6.717
N9
G2
M02 M03 M04 M05 M06
0.0194 0.0242 0.0150 0.0127 0.0221
0.000404 0.000737 0.000153 0.001060 0.000346
6.306 9.423 6.318 6.391 10.324
N9 N9 N9 N9 N9
M07 M08 M09 M10 M11 M12
0.0198 0.0165 0.0188 0.0096 0.0174 0.0230
0.000600 0.000058 0.000643 0.001852 0.000473 0.001069
11.509 10.481 10.908 8.492 11.579 8.835
N9 N9 N9 N9 N9 N9
a b c
Over 5 repetitions. According to ISO4287. According to ISO/R468 and ISO2632-1.2.
• Group G1: comprises five metals such as medium carbon steel (DIN C45k), low alloy steel (DIN CrMo4), stainless steel (AISI 316) and aluminium alloys (CuAl10 and EN-AW 6082). • Group G2: comprises seven types of plastics such as PET (white and green), PA (white and black), PTFE, PU and PP. Each material has been evaluated by digitising tests on flat surfaces of 31.75 mm of diameter. Specimens were manufactured by face milling under identical cutting conditions so that a uniform roughness was achieved in order to avoid its influence on the scanning results. The actual values of roughness were measured by means of a Leica DCM 3D microscopy according to ISO 4287 standard and they are shown in Table 2. All the values of Ra belong to grade N9 according to ISO 2632-1.2. 2.3. Sensor parameters The adjustment of the CH sensor is described by the manufacturer in the operator manual [17]. The main recommendations regarding this topic are summarised in this section. There are two main setting parameters for the Conoprobe Mark III sensor: • Working Frequency (F) represents the data acquisition rate and can be set up to a maximum of 3000 Hz. • Power Level (P) represents the value of the laser beam energy used and can be set up in a range from 0 to 63, where 63 is equivalent to 1 mW of power.
45
The adjustment procedure consists on working with the highest F possible, since measurement error can be minimised by better use of averaging filters. Then, P should be adjusted so that a proper amount of energy reaches the sensor. In general, there is a direct relationship between F and the corresponding required P. This means that working at high frequency will require a high laser power level, whereas working at low frequency will require a lower laser power level. The optimum combination of F and P should be set by means of two indicators of the signal quality: • SNR (Signal-to-Noise Ratio), which represents the percentage of the signal power received by the sensor with respect to the total power used including the signal noise. SNR may range from 0% to 100% and it is commonly assumed that the higher the SNR, the higher the accuracy of measurement. Reliable values of SNR should be above 50%. • Total, which is proportional to the area limited by the signal envelope, increases as signal intensity does. Values of Total between 1200 and 18,000 may yield accurate results but recommended values range from 2000 to 16,000. 2.4. Experimental procedure As an optic technique, CH is affected by factors such as the surface slope [9] and its position within depth of field [10]. In order to avoid these influencing factors, the test specimens were installed in a test bench (see Fig. 3) taking care that the test surfaces were located parallel to the XY plane of the CMM at distance equal to the sensor stand-off. Once the surface was precisely orientated, the test surface of each specimen was digitised by a touch probe on the CMM in order to provide a reference value of flatness (ft ), which will be one of the quality indicators used in this study. With this probe, 15 points distributed within a square mesh of 10 × 10 mm were captured every 2.5 mm along X direction and every 5 mm along Y direction. This procedure was repeated 5 times on each specimen. The average and standard deviation values were included in Table 2. In a similar way, the same points were scanned by the CH sensor taking 500 measurements on each point. The resulting average was considered as the actual value of distance (). The reliability of this value is provided by the standard deviation p . This routine was repeated under each combination of F and P. The range of power P was varied from 1 to 63 at increments of 1, whereas frequency F was varied from 500 to 3000 Hz at increments of 500 Hz. Therefore, 378 combinations of P and F were tested. Since 12 types of materials were considered in the study, an amount of 4536 tests were finally performed.
Fig. 3. General view of the test bench and detailed view of a test specimen being scanned.
46
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
Fig. 4. Distribution of SNR values within the RA for materials in G1.
According to the work carried out by Fernández et al. [15], tests were conducted at least 60 min after switching on the laser in order to reach thermal stabilisation of the sensor and avoid thermal drift.
• Measurement Dispersion ( ): it is calculated by expression in Eq. (1) as the average value of p when considering 15 points within the point-cloud. In this case, the lower , the less dispersion.
15
2.5. Filtering process Data acquired in the tests have been filtered in order to remove from the study those combinations of F and P which yield low quality signals, as well as those in which a valid register is not obtained. Taking into account the manufacturer recommendations aforementioned in Section 2.3, the filtering process consisted in rejecting the combinations of F and P with SNR lower than 50% and parameter Total out of the range 2000 to 16,000. Those combinations of F and P that passed this filtering process provided a reliable signal and defined a Reliability Area (RA). 3. Analysis of results 3.1. Quality indicators Three quality indicators have been used in this work to perform the analysis of results: • Signal-to-Noise-Ratio (SNR): this indicator provides information about the quality of the signal used by the sensor to process distance measurements.
=
p=1 p 15
(1)
• Flatness Deviation (ıf ): it shows the difference between the flatness value measured by CH sensor (fc ) with respect to the reference value measured by a standard touch probe (ft . Therefore, it can be considered as an indicator of the reliability of the CH sensor for a flat surface reconstruction. It is calculated as in Eq. (2), ıf = |fc − ft |
(2)
The lower value of ıf , the more coincidence between the reconstructed surface and the reference one. 3.2. Experimental results In order to facilitate the analysis, results have been organised according to the material groups included in Fig. 2. Distribution of the quality indicators SNR, and ıf are described in this section by means of graphs for different combinations of F and P within the RA. A summarising table is provided for each group of materials, which includes the maximum, minimum, average value and standard deviation of each quality indicator within the RA. The average value shows the most probable value of measurement while the standard deviation gives an idea of the uniformity of each
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
47
Fig. 5. Distribution of values within the RA for materials in G1.
quality indicator. Correlation coefficient (R2 ) has been also included between combinations of the three indicators (SNR – ; –ıf ; SNR – ıf ). 3.2.1. Metals (G1) Figs. 4–6 show the value of the three quality indicators (SNR, and ıf ) obtained within the RA for materials in the group G1: • In general, there exist lots of valid combinations of F and P. • For M03 and M04, the amplitude of the valid ranges of power is almost independent of the value of frequency used. • For M01, M02 and M05 the valid ranges of power become wider as the value of frequency increases. The general warm colour represented within the RA on the SNR graphs shows the high value of this indicator (SNR > 70%) and therefore, the good quality of the signal acquired by the CH sensor for all the combinations of F and P and all the materials within this group (see Fig. 4 and Table 3). Moreover, the uniformity of colour shows the low variability of the indicator in most part of the RA (0.122% ≤ SNR ≤ 2.240%). With regard to , the cold colour of the graphs shows the low value of this indicator (see Fig. 5). Minimum values lower than 0.240 m are met, which are clearly lower than reproducibility of the sensor (1 < 1 m). Besides, variability of the indicator is low ( ≤ 0.308 m) which shows a good measurement quality in almost all the RA.
With respect to ıf , the material M05 presents the lowest values (3.116 m ≤ ıf ≤ 9.352 m) and the lowest variability (ıf =
0.754 m). The rest of materials have got graphs with warmer colours and, therefore, high values of the indicator which are far from the optimum (see Fig. 6). Moreover, these materials also have high variability (ıf > 1.2 m). If correlations among the different quality indicators are analysed, it can be noticed that they are very low (see Table 3), except for M03 (R2 [ − ıf ] = 0.805) and M05 (R2 [SNR − ıf ] = 0.845). Based on the variability observed for the three indicators, it can be said that the election of one or another combination of F and P within the RA does only affect significantly to ıf . Therefore, when a geometrical feature such as flatness has to be measured for this type of materials, a combination of F and P which minimises this indicator should be selected.
3.2.2. Plastics (G2) Before discussing the results for this group, it should be noted that M12 has been excluded from the analysis since signal acquired by the CH sensor does not pass the filters mentioned in Section 2.5. Figs. 7–9 show the value of the three quality indicators (SNR, and ıf ) obtained within the RA for materials in the group G2: • For M06, M07, M08 and M10 the valid ranges of power are very small (from 7 to 18) and independent of the frequency value used.
48
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
Fig. 6. Distribution of ıf values within the RA for materials in G1.
• For M09 and M11 the valid ranges of power become wider as the value of frequency increases. The graphs of SNR for M09 and M11 present a uniform warm colour (see Fig. 7). This proves the high quality of the signal acquired by the CH sensor for these materials (SNR > 78%). Furthermore, low variability is observed since the standard deviation is low ( SNR ≤ 0.140%) (see Table 4). In contrast, the SNR graphs for the
rest of materials include colder colours which are related to signals of lower quality (50.151% ≤ SNR ≤ 76.331%) as well as to high variability ( SNR > 2.2%). In relation to the indicator , it is again observed that materials M09 and M11 show the coldest colours within the RA. The indicator is lower for high values of power and low variability is met ( = 0.279 m for M09 and = 0.223 m for M11). Variability for M06 is also low but this value is not significant since
Table 3 Value of the quality indicators within the RA for materials in the group G1. M01
M02
M03
M04
M05
SNR (%)
Max. Value Min. Value SNR SNR
79.492 77.833 79.065 0.380
80.320 79.602 80.149 0.122
79.472 69.147 77.713 2.240
79.098 72.870 78.218 1.106
80.144 77.113 79.323 0.793
(m)
Max. Value Min. Value
0.628 0.240 0.377 0.074
0.925 0.151 0.307 0.146
1.762 0.217 0.571 0.308
1.006 0.169 0.433 0.185
1.189 0.168 0.304 0.152
ıf (m)
Max. Value Min. Value
29.938 9.655
17.575 12.059
20.503 4.881
26.928 10.496
9.352 3.116
ıf ıf
19.318 5.185
14.903 1.203
13.220 3.881
20.778 4.304
5.740 0.754
R2 [SNR − ]
0.521
0.485
0.549
0.538
0.100
R [ − ıf ]
0.223
0.265
0.805
0.381
0.122
R2 [SNR − ıf ]
0.644
0.343
0.335
0.400
0.845
2
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
49
Fig. 7. Distribution of SNR values within the RA for materials in G2.
there are few values within the RA. On the contrary, value of the indicator is high for the rest of materials, such as the case of M07 where the maximum ( = 9.991 m) is met out of the represented range in Fig. 8. Moreover, variability is also notable, with extreme values of = 1.890 m for M07 and = 2.218 m for M08.
With respect to ıf , materials M06 and M08 present the lowest values (1.106 m ≤ ıf ≤ 4.166 m). and the lowest variability (ıf ≤ 0.349 m). The rest of materials meet high values of the
indicator and high variability (ıf > 1.5 m). For this indicator material M07 scores very high values which are out of the represented range in Fig. 9.
Table 4 Value of the quality indicators within the RA for materials in the group G2. M06
M07
M08
M09
M10
M11
SNR (%)
Max. value Min. value SNR SNR
63.296 51.032 58.510 3.701
60.079 50.906 57.348 2.287
71.507 50.151 67.024 5.368
79.121 78.377 78.838 0.140
76.331 62.149 70.983 4.724
80.445 79.589 80.268 0.093
(m)
Max. value Min. value
3.192 1.236 1.928 0.657
9.991 2.186 3.953 1.890
9.100 0.809 2.590 2.218
2.136 0.416 0.694 0.279
6.802 0.527 2.097 1.724
1.654 0.226 0.427 0.223
ıf (m)
Max. value Min. value
4.166 2.948
41.217 35.094
1.983 1.106
33.151 13.628
20.956 15.115
12.266 3.375
ıf ıf
3.571 0.349
37.895 1.507
1.552 0.254
18.545 4.022
17.465 1.658
8.216 2.050
R2 [SNR − ]
0.786
0.799
0.873
0.263
0.811
0.356
R [ − ıf ]
0.321
0.430
0.428
0.285
0.674
0.358
R2 [SNR − ıf ]
0.082
0.508
0.536
0.615
0.688
0.002
2
50
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
Fig. 8. Distribution of values within the RA for materials in G2.
Since correlations R2 [SNR − ıf ] and R2 [ − ıf ] are very low it is not possible to set a combination of F and P which satisfy simultaneously the three indicators. Nevertheless, the correlation between SNR and is quite high for materials M06, M07, M08 and M10 (R2 [SNR − ] > 0.78) but it is not significant since the extension of the RA is very small. Similarly to the group of metals (G1), it can be stated that the election of one or another combination of F and P within the RA for materials M09 and M11 does only affect significantly to ıf . Therefore, when a geometrical feature such as flatness has to be measured for these two materials, a combination of F and P which minimises this indicator should be selected. For materials M06, M08 and M10 the election of a combination of F and P should be performed in order to improve the SNR and , which have a good correlation, whereas the indicator ıf will not be practically altered. The behaviour of the three quality indicators for material M07 is particularly different, since this is slightly translucent. The SNR is low, is high and ıf is very high. Although the sensor is capable of acquiring points, these have very low quality and it is not feasible to perform measurements of the geometry within the usual tolerance limits. 3.3. General discussion Values of the quality indicators used in the tests (SNR, and ıf ) show differences among one materials and others, even within the same group. However, the test materials can be grouped according to the values of the indicators, as shown in Table 5.
Table 5 Classification of materials according to the value of indicators. Group
Type A
Type B
G01 G02
SNR > 75% and SNR < 1% M01, M02, M03, M04, M05 M09, M11
SNR < 75% and SNR > 1% – M06, M07, M08, M10
G01 G02
< 1 m and < 1 m M01, M02, M03, M04, M05 M09, M11
> 1 m and > 1 m – M06 a , M07, M08, M10
ıf > 10 m and ıf > 1 m
ıf < 10 m and ıf < 1 m
M01, M02, M03, M04 M07, M09, M10, M11a
M05 M06, M08
G01 G02
a Materials are included into this Type although they do not meet strictly both conditions.
Type A materials are those whose indicators SNR and reach the best values whereas the indicator ıf meets the worst results with great variability. In these cases, it would be advisable to perform the adjustment of the conoscopic sensor by choosing a combination of F and P which optimises the metrological indicator ıf . From a practical point of view, the search for this optimal combination requires to carry out previous tests of the material to be scanned, using a similar procedure as it is proposed along this work. Table 6 shows the recommended values of F and P for this type of materials. On the other hand, Type B materials are those whose metrological indicator ıf reaches the best values and slight variability
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
51
Fig. 9. Distribution of ıf values within the RA for materials in G2.
whereas indicators SNR and are far from an optimal behaviour. In these cases, it would be advisable to perform the adjustment of the conoscopic sensor by choosing a combination of F and P which optimises indicators SNR and . Since there exists a certain correlation between these two indicators, the optimal combination should be directly found by tuning F and P until value of the SNR rises to the maximum. Among these types of materials, the behaviour of M05, M06, M07, M10 and M11 does not match strictly to the general rule described. The three indicators of material M05 show a good behaviour which is far from the rest of the test materials. This is because in this case there is a greater correlation between indicators SNR and ıf . For this reason, it is suggested for this material to adjust F and P for maximising the value of the indicator SNR, as it is recommended for the materials of Type B.
With regard to M06, variability of the indicator for this material is low but this result is not actually significant since the extension of the RA is very small. Furthermore, since behaviour of the rest of indicators for this material is similar to that described for Type B materials, sensor adjustment should be carried out in a same way. Regarding materials M07 and M10, it is observed that the three indicators have a bad behaviour. Low quality of acquired points can be related to the materials optical characteristics, since M07 is translucent and M10 is porous. In consequence, it is advisable not to use a CH sensor for digitising these materials, especially if there exists any tight geometric specification on the product. Finally, although M11 presents low values for ıf , it has got high variability and the behaviour of the rest of indicators is similar to that of Type A materials. Then, it is reasonable to treat this case in the same way as materials in this group.
Table 6 Recommended combinations of P and F for adjustment of the CH sensor according to the test material. Group
Type A
G01 P (1–63) F (Hz)
M01 13 500
M02 52 3000
M03 41 500
M04 60 3000
M05 11 500
P (1–63) F (Hz)
M07 – –
M09 60 3000
M10 – –
M11 12 500
M06 10 2000
G02
Type B
M08 15 2500
52
D. Blanco et al. / Precision Engineering 42 (2015) 42–52
4. Conclusions This work demonstrates that the adjustment of a CH sensor for digitising based exclusively on improving the quality of the acquired signal is not enough to guarantee the highest precision in geometrical measurements. Moreover, the adjustment of the sensor should consider another indicator based on the type of metrological magnitude to be verified. With this purpose, numerous digitising tests were performed by means of a Conoprobe Mark III CH sensor over flat samples of 12 different materials grouped into metals (G1) and plastics (G2). Each material was scanned under different combinations of frequency F and power P and filtered based on recommendations by the manufacturer in order to provide a Reliability Area (RA) where the quality of signal was guaranteed. Three quality indicators were calculated for each combination of F and P within the RA: Signal-to-Noise Ratio (SNR), measurement dispersion ( ) and flatness deviation of the point-cloud (ıf ). Several conclusions can be highlighted from the study, which are summarised as follows: • Extension of RA varies greatly for each material, even among materials included in the same group. Consequently, it is not possible to state a general rule for the limits of the RA for different materials. • In general, results show a very dissimilar behaviour of the three quality indicators considered (SNR, , ıf ) so it is not possible to carry out a simultaneous optimisation. • Based on the tests performed, it has been observed that those materials whose indicators SNR and reach the best values have got the worst results for the indicator ıf (Type A). On the contrary, materials with better results for the metrological indicator ıf are those whose indicators SNR and reach the worst values (Type B). • It is advisable for Type A materials to perform the adjustment of the conoscopic sensor by selecting a combination of F and P which optimises the metrological indicator ıf . In the case of materials of Type B, it is suggested to select a combination of F and P which optimises the indicators SNR and . Although reviewed studies consider the maximum value of the SNR as the main digitising quality indicator, the present work reveals that not always there is a link between signal quality and measurement quality. Actually, it was observed that an optimal sensor configuration which maximises signal quality does not necessarily indicate a minimisation of the geometrical error ıf . Therefore, when high precision measurements are required, a geometrical type indicator should be used to ensure minimum errors in materials of Type A. The main drawback consists in the
necessity of performing previous tests before digitising in order to adjust the parameters of the CH sensor for such a material. Acknowledgments The work described is part of a research supported by the Spanish Ministry of Economy and Competitiveness and FEDER (DPI2012-30987) and the Regional Ministry of Economy and Employment of the Principality of Asturias (SV-PA-13-ECOEMP15). References [1] Li F, Longstaff A, Fletcher S, Myers A. Integrated tactile and optical measuring systems in three dimensional metrology. In: Proceedings of the Queen’s Diamond Jubilee, CEARC’12. 2012. p. 1–6. [2] Shiou FJ, Chen MJ. Intermittent process hybrid measurement system on the machining centre. Int J Prod Res 2003;41(18):4403–27. [3] Sirat G, Psaltis D. Conoscopic holography. Opt Lett 1985;10(1):4–6. [4] Malet Y, Sirat GY. Conoscopic holography application: multipurpose rangefinders. J Opt 1998;29(3):183–7. [5] Sirat GY, Paz F, Agronik G, Wilner K. Conoscopic systems and conoscopic holography [Internet]. Israel: Laboratory for CAD & Lifecycle Engineering. Faculty of Mechanical Engineering, Technion-Israel Institute of Technology; 2012 [cited 2012 November 19]. Available from: http://mecadserv1. technion.ac.il/public html/IK05/Sirat 9375.pdf [6] Álvarez I, Enguita JM, Frade M, Marina J, Ojea G. On-line metrology with conoscopic holography: beyond triangulation. Sensors 2009;9(9):7021–37. [7] Ko SL, Park SW. Development of an effective measurement system for burr geometry. Proc Inst Mech Eng Part B: J Eng Manuf 2006;220(4):507–12. [8] Toropov A. An effective visualization and analysis method for edge measurement. In: Gervasi O, Gavrilova M, editors. Computational science and its applications – ICCSA 2007, Lecture notes in computer science – LNCS 4706, Part II. Berlin Heidelberg: Springer-Verlag; 2007. p. 941–50. [9] Paviotti A, Carmignato S, Voltan A, Laurenti N, Cortelazzo GM. Estimating angle-dependent systematic error and measurement uncertainty for a conoscopic holography measurement system. In: Proceeding SPIE 7239, ThreeDimensional Imaging Metrology, 72390Z. 2009. ˜ G, Mateos S. Influence of surface [10] Fernández P, Blanco D, Rico JC, Valino position along the working range of conoscopic holography sensors on dimensional verification of AISI 316 wire EDM machined surfaces. Sensors 2014;14(3):4495–512. [11] Lathrop RA, Hackworth DM, Webster III RJ. Minimally invasive holographic surface scanning for soft-tissue image registration. IEEE Trans Biomed Eng 2010;57(6):1497–506. [12] Zhu L, Barhak J, Srivatsan V, Katz R. Efficient registration for precision inspection of free-form surfaces. Int J Adv Manuf Technol 2007;32(5–6):505–15. [13] Lonardo PM, Bruzzone AA. Measurement and topography characterization of surfaces produced by selective laser sintering. CIRP Ann – Manuf Technol 2000;49(1):427–30. [14] Lombardo E, Martorelli M, Nigrelli V. Non-contact roughness measurement in rapid prototypes by conoscopic holography. In: XII ADM International Conference, B2. 2001. p. 11–8. ˜ G, Hoang H, Suárez CM, Mateos S. Integration of [15] Fernández P, Blanco D, Valino a conoscopic holography sensor on a CMM. In: Marcos M, Salguero J, editors. AIP Conference Proceedings 1431. 2012. p. 225–32. [16] Smith KB, Zheng YF. Point laser triangulation probe calibration for coordinate metrology. J Manuf Sci Eng 2000;122(3):582–6. [17] ConoProbe MKIII OEM Manual. Optimet manual P/N 3J06007 [Internet]. Optimet (Israel); [cited 2013 April 5]. Available from: http://www.optimet.com