Using Industrial Robots to Manipulate the Measured Object in CMM

1. Introduction

In today's global market, the importance of correct, reliable and comparable measurements is the key factor for achieving quality in activities and procedures in every area of industry. Calibration, testing and measurement are necessary elements in the development process or progress in many disciplines of science and industry.

The majority of modern industrial measurements can be categorized as GDT (Geometrical Dimensioning and Tolerancing). Increasingly, measurements obtained by coordinate measuring machines (CMM) are being used.

CMMs are measuring devices with high measuring speed. Positioning and rotation of the measured object are always performed manually in the work area of the coordinate measuring machine. The object, with defined dimensions, shape and measuring surfaces, is measured and controlled from several different sides. The measured object needs to be positioned in certain positions relative to the measuring device, which requires complex and time-consuming actions. Each change of position also requires a certain time, which can cause increased costs in control and production processes. In order to reduce these costs, several options for positioning a measured object inside the working area of a CMM using an industrial robot have been considered.

In a modern industrial environment the majority of robots are automated systems controlled by computers. Industrial robots have one or more robotic arms, control devices with memory, and sometimes use sensors for data acquisition. They usually support the manufacturing process by positioning objects during machining or welding, transportation, various technological operations, automatic assembly, etc. They are also sometimes used for pre-process, process and post-process control. Industrial robots are widely used in processes which require high quality and productivity.

Research along these lines has been conducted by Santolaria and Aguilar [1]. They conducted a survey about the development of kinematic modelling of robotic manipulators and articulated arm coordinate measuring machines (AACMM), taking into consideration the influences of the chosen model on procedure parameters. Their optimization algorithm included the terms linked to the accuracy and repeatability of the procedure presented. The algorithm follows the simple optimization scheme of data obtained by investigation in several spheres of objects placed at various positions within the working area of both systems.

The majority of research mostly concentrates on differences in the influences on measurement uncertainty in Coordinate Measuring Machines. Lawford [2] observed dysfunctional CMS software with unknown measurement uncertainty and compared its influence with measurement results. He also investigated the prescribed testing algorithms and the checking of industrial software, testing by comparison with given algorithms. This yielded solutions to similar problems.

Fang, Sung and Lui [3] observed the influence of measurement uncertainty from CMM calibration and temperature in the working environment. If the measurement uncertainty in CMM calibration is reduced, this will also reduce the measurement uncertainty of the machine itself. They also pointed out the importance of temperature balance in the working environment before the measurement is performed, i.e., the temperature should be controlled in order to fit CMM working specifications. CMM uncertainty can be reduced using highly precise instruments such as a laser interferometer.

One of the first papers about CMM error compensation was presented by Zhang et al. [4]. They described the error compensation on bridge-type CMM, which resulted in an improvement of accuracy by a factor of 10. They also presented the correction of vector of equally distributed points in the measured volume.

Software error compensation has been reported by a number of authors. Ferreira and Liu [5], for example, developed the analytical model for geometric errors of the machining assembly; Duffle and Yang [6], meanwhile, invented a method to generate the kinematic error function from volumetric measurement error using a vectorial approach.

Robot-CMM integration was performed for the first time by the Mitutoyo Company [7]. They developed a software module in order to adjust the actions of CMM and robotic handling machines used for manipulation of measured parts. Mitutoyo has released the source code in the hope that third party software vendors will be able to use it as a basis to develop products. To the authors' knowledge, at the time of writing no achievements have been made in this direction.

Hansen et al. [8] estimated measurement uncertainty of a hybrid system consisting of an Atomic force microscope attached to a coordinate measuring machine, using linear combination of these two components. Although Hansen et al. also combined two devices, their approach nevertheless differs from ours: we use one system to position the measured object, while they used a two-component system to perform the measurement.

Aggogeri et al. [9] used simulation and planned experimentation to assess the measurement uncertainty of CMMs. They identified and analysed five influence factors, and showed that simulation can successfully be used to estimate CMM uncertainty.

Weckenmann et al. [10] investigated how measurement strategy affects the uncertainty of CMM results. They defined the measuring strategy in relation to “operator influence”, which has been neglected in other research. This work showed that measuring strategy influences CMM uncertainty, and that scanning capabilities of modern CMMs, using significantly a larger number of touch points, overcome this influence.

Wilhelm et al. [11] also investigated the influence of measurement strategy, which they defined as the “task specific uncertainty”. They also showed that virtual CMM, using Monte Carlo simulation, can be used to estimate uncertainty. Nevertheless, although these authors mentioned that part fixture influences uncertainty, they did not analyse this thoroughly.

Feng et al. [12] applied the factorial design of experiments (DOE) to examine measurement uncertainty. They also studied the effect of five factors and their interaction, and showed that there is statistically significant interaction between speed and probe ratio. They also showed that uncertainty is minimized when speed is highest, stylus length is shortest, probe ratio is largest, and the number of pitch points is largest.

Piratelli-Filho and Giacomo [13] proposed an approach based on a performance test using a ball bar gauge and a factorial design technique to estimate CMM uncertainty. They investigated the effect of length, position, and orientation in work volume on CMM measurement errors. The analysis of variance results showed a strong interaction between the orientation and measured length.

Unlike the mentioned studies, the goal of the research presented here was to assess whether robots can be used to position the measured object in complex measuring systems, using measurement uncertainty analysis and estimating the factors affecting it.

2. Problem description

Fig. 1 shows a typical assembly used to clamp the measured object in a CMM. Such a system requires multiple measuring and clamping operations when different features of the measured object are being measured, especially if the geometry is complex and some portions are inaccessible by CMM probes. Such products should be measured in multiple steps.

OPEN IN VIEWER

Figure 1.

Typical fixturing assembly for CMM measurements

OPEN IN VIEWER

Figure 2.

Calibration with ceramic reference sphere.

The major time-consuming task in coordinate measurement of complex geometry with CMMs is to choose the position of the measured object and the optimal combination of probe stylii, which often prevent the touch probes from coming into contact with the measured object. The obscured surfaces are difficult to reach without repositioning the measured object. Every repositioning of the measured object introduces new error sources in the measurement. Our idea is to use the industrial robot to manipulate the measured object, i.e., to position it automatically.

In order to test this possibility, the measurements were conducted with three different systems:

It is common to use the measurement uncertainty to quantify the quality of the measurement process. Therefore, we decided to estimate it for these three systems, and to compare them with the uncertainty of the robot and the CMM as stated by the manufacturers.

3. Experiment description

3.3 Measured object

The measured object was selected to have geometrical features typically found in coordinate measurements: planes, cones, and cylinders. The material of the measured object was not of great importance, since temperature deviations were negligible (laboratory conditions), and typical measurement force (200 mN) did not deform the object. The object was made of PVC and the surface was metalized, reducing surface roughness to a minimum. Fig. 3 shows the four features measured.

OPEN IN VIEWER

Figure 3.

Measured object with defined geometrical features (a - top plane of cone, b - conical portion with larger angle, c - conical portion with smaller angle and d - top plane of cylinder)

3.5 Geometrical features

Dimensional measurements were repeated a certain number of times on previously described surfaces defining different workpiece geometries. For each particular surface, the dimensions were measured 25 times, under the same conditions, in order to compensate random errors. The number of measurements (the size of the sample) was determined according to the significance level of the test α = 0.01 and the probability of failing to detect a shift of one standard deviation β = 0.01 for a two-sided test, assuming normal distribution and known standard deviation [14].

The planar features were measured by sets of 250 points distributed circularly, and conical features were measured by measuring two circles, each consisting of 250 points, at distances of 1 mm (cone “b”) and 3 mm (cone “c”) from the edges, in order to avoid filleted edges. The measurement results were used to estimate the measurement uncertainty, as a measure of validity of the results and confirmation of hypotheses 1 and 2.

For hypothesis 3, the procedure was identical, but with increased mass of measured object. The first measuring cycle was performed on a CMM with the measuring object fixed on the CMM's granite table. The second and third cycles were measured by the complex CMM-robot system.

Figs. 4 and 5 show the planes and cones used to define the dimensions to be measured.

OPEN IN VIEWER

Figure 4.

Geometric features measured by CMM.

OPEN IN VIEWER

Figure 5.

Measured features in CMM software.

The measured dimensions were defined as follows:

—

Diameter d₁ is the intersection of plane “a” and cone 1

—

Diameter d₂ is the intersection of cone 1 and cone 2

—

Diameter d₃ is the intersection of plane “d” and cone 2

—

Height H is the distance between planes “a” and “d”.

4. Experiment

In the first measuring cycle, the measured object was positioned and fixed to the CMM's measuring table, and in the second cycle the position of the measured object was defined by the robot's arm position (i.e., auxiliary elements in the robot's arm were holding the measured object) inside the CMM's coordinate space. Between every single measurement in this measuring cycle came a phase of the robot's arm movement from one position to another and back again. Coordinates of the robot's arm in both positions were defined by the robot's off-line programming in such a way that the robot could be positioned manually and that position memorized. After this, the robot's operating speed was defined. Since the robot repeated this operation for each measurement, the robot's repeatability was ±0.04 mm. Figs. 6 and 7 show the first and the second robot arm positions, respectively.

OPEN IN VIEWER

Figure 6.

Robot arm position 1.

OPEN IN VIEWER

Figure 7.

Robot arm position 2.

Since the measured object's mass was constant during the first two measuring cycles, the mass of the measured object was increased by adding mass m = 600 g (Fig. 8). The third measuring cycle was conducted with increased mass and the results were compared with the first cycle.

OPEN IN VIEWER

Figure 8.

Measuring object with additional mass.

5. Measurement results

Measurement results are shown in Table 1.

OPEN IN VIEWER

Table 1.

Results of first measuring cycle - measuring object fixed on CMM's measurement table

Measure-ment No.	Measured values (mm)
	Diameter d1	Diameter d2	Diameter d3	Height H
1.	39.0485	42.5430	42.9402	18.8070
2.	39.0482	42.5432	42.9400	18.8065
3.	39.0484	42.5430	42.9401	18.8067
4.	39.0483	42.5431	42.9401	18.8068
5.	39.0485	42.5433	42.9404	18.8072
6.	39.0484	42.5432	42.9405	18.8070
7.	39.0484	42.5434	42.9407	18.8071
8.	39.0486	42.5436	42.9408	18.8072
9.	39.0486	42.5436	42.9410	18.8073
…	…	…	…	…
23.	39.0481	42.5439	42.9405	18.8071
24.	39.0481	42.5440	42.9404	18.8081
25.	39.0487	42.5437	42.9409	18.8074
Mean value x1m	39.0484	42.5436	42.9407	18.8074
Standard deviation	0.00026	0.00038	0.00043	0.00050
Max	39.0487	42.5444	42.9415	18.8082
Min	39.0478	42.5430	42.9400	18.8065
Absolute range E1	0.0010	0.0014	0.0015	0.0017

5.1 Measurement uncertainty

The declared measurement uncertainty of the CMM used in this experiment is 1.80 μm. In all three cases there is Type A standard measurement uncertainty, which equals standard deviation times coverage factor 2. The standard measurement uncertainties of three measurement cycles are shown in Table 2 and Fig. 9.

OPEN IN VIEWER

Figure 9.

Comparison of measurement uncertainties (μm).

OPEN IN VIEWER

Table 2.

Standard measurement uncertainty in three measuring cycles: 1. measuring only on CMM; 2. measuring on CMM-robot system; 3. measuring on CMM-robot, with additional mass

Measured value	Standard measurement uncertainty (μm)
	Case 1	Case 2	Case 3
	only CMM	CMM-robot	CMM-robot with added mass
Diameter d1	0.53	12.33	3.86
Diameter d2	0.76	12.98	4.69
Diameter d3	0.86	10.92	3.95
Height H	1.00	9.72	4.17

We can conclude that all measurements in the first case expressed lower uncertainty than stated on the CMM's calibration certificate. The measurement results in the second case show that measurement uncertainty of the system CMM-robot is significantly larger than the declared uncertainty of the CMM. In the third cycle, where the CMM-robot system was used to measure the object with increased mass, the measurement uncertainties are larger than the declared uncertainty of the CMM, but still lower than those in the second case. This means that increased mass of the measured object slightly reduced uncertainty. This phenomenon could be explained by the increased inertia of the measured object, which stabilizes the system and leads to more accurate results.

5.2 Statistical analysis

The first step in statistical analysis was to question the normality of distribution of the measurement results. Kurtosis of all results was between −1.40 and 0.54, and the skew ranged between −0.99 and 0.72. For 25 samples, the standard error of the skew is 0.49 and standard error of the kurtosis is 0.98; therefore, both skew and kurtosis are lower than twice the standard error, and we can assume normal distribution of measured data.

The histograms of distributions for measured values of diameter d₁ (Figs. 10 and 11) illustrate the normality of distribution. The distributions of most other measured values have a similar shape.

OPEN IN VIEWER

Figure 10.

Histogram of distribution of diameter d₁ (Case 2: CMM-robot) with fitting normal distribution.

OPEN IN VIEWER

Figure 11.

Histogram of distribution of diameter d₁ (Case 3: CMM-robot + added mass).

OPEN IN VIEWER

Figure 12.

Boxplots of diameter d₁ reveal no outliers.

Kolmogorov-Smirnov test was used to check the difference in the means of results obtained in the first (measurements performed on fixed measured object using only CMM) and in the second measurement cycle (measurement obtained by the system robot-CMM). The results for all four measured geometrical values (diameters d₁ to d₃ and height H) are shown in Table 3.

OPEN IN VIEWER

Table 3.

The results of the statistical Kolmogorov-Smirnov test (significance level p < 0.05) and t-Test: Two-Sample Assuming Unequal Variances (α<0.01, Hypothesized Mean Difference 0)

	d1	d2	d3	H
Mean CMM only	39.0484 mm	42.5436 mm	42.9408 mm	18.8075 mm
Mean CMMrobot	39.0278 mm	42.5194 mm	42.9214 mm	18.8238 mm
Variance CMM only	6.97E-08 mm	1.46E-07 mm	1.86E-07 mm	2.48E-07 mm
Variance CMMrobot	3.80E-05 mm	4.21E-05 mm	2.98E-05 mm	2.36E-05 mm
D CMM only	0.166	0.107	0.091	0.129
p-value CMM only	0.458	0.920	0.980	0.771
D CMM-robot	0.109	0.116	0.201	0.151
p-value CMM-robot	0.912	0.871	0.235	0.585
t Stat	16.7275	18.5872	17.6842	−16.7088
P(T<=t) 1tail	4.94E-15	4.69E-16	1.43E-15	2.24E-15
t Critical 1tail	2.492 (α<0.01), 1.711 (α<0.05)
P(T<=t) 2tail	9.89E-15	9.38E-16	2.87E-15	4.48E-15
t Critical 2tail	2.797 (α<0.01), 2.064 (α<0.05)

As the computed p-value is greater than the significance level 0.05, we cannot reject the null hypothesis HO (the sample follows a Normal distribution).

Levene's test confirmed that variances in the two observed cases are different. Therefore the Welch's t-Test, Two-Sample Assuming Unequal Variances, was performed in order to check the difference in the means of the results obtained in the first and the second measurement cycles.

The results of the t-Test, shown in Table 3, lead to the conclusion that Hypothesis 2 (“Uncertainty of measurement for complex CMM-robot measuring system is within the limits of allowed uncertainty of measurement of CMM”) should be rejected, since P-value for both one-tail and two-tail are significantly lower the than critical value of t-variable for sample size 25, α being either 0.05 or 0.01.

6. Conclusion

The principal idea of this paper was to extend the possibilities for automating the measurement process with coordinate measuring machines. The obstacle most often encountered with CMM measurements are limitations of geometry, requiring more measurement sequences in order to reach difficult places on the measured object. It is possible to perform measurements of such objects, but manual repositioning of the measured object, including redefinition of the local coordinate system, slows down the process. If an industrial robot is used to manipulate the measured object, such a process could be automated. The ultimate goal is to keep the measurement uncertainty within allowable limits The measurements of the dimensions of the measured object were conducted by complex CMM-robot measuring system, with movements performed between each single measurement. These results were compared with the results obtained by measuring the same object fixed in the CMM. The measurement results in these two cases were different; one of the reasons for this could be the slight impacts and vibrations that were obvious during every movement phase between measurements.

Although the obtained measurement results still have great accuracy and precision, they do not meet the criteria of the CMM's prescribed measurement uncertainty.

It can be concluded that it is possible to conduct measurements using complex CMM-robot measuring systems, but the measurement results are dictated by the measurement uncertainty of the least accurate component of the system, which in this case was the industrial robot. Significant differences and deviations in measurement results can be confirmed by comparing obtained measurement results with results measured on an object with a different mass. This confirms the significant influence of variation in the mass of the measured object on the measurement uncertainty of the complex CMM-robot measuring system.

It can be argued that it could still be possible to confirm hypothesis 2, assuming the fulfilment of certain conditions such as: