6.4. Role of sampling in experimental work
6.4.1. Background
Although the sampling of pulp slurry is a common operation in any pulp or paper mill, many analyses are still performed manually in a laboratory, either because on-line measurements are not available or because they are not accurate enough and require frequent calibration. Errors are always liable to occur in both sampling and the analysis of samples. Sampling is generally the most prone to errors, while laboratory analyses are more accurate, thanks to standardised analytical methods. Surprisingly little attention has been paid to techniques for sampling pulp slurry, so that sampling devices have often been validated and placed in position without adequate know-how. In addition, the reproducibility of sampling and the representativeness of the samples have mostly remained unknown. In spite of this, optimisation and control of the process, the calibration of measurements and procedures for experimental testing have been based on such samples.
Sampling may be manual, semi-automatic or automatic. Manual sampling valves are still the most common in the pulp and paper industry, because they are inexpensive and simple to use and maintain. Sampling systems can be considered semi-automatic when their operation is automatic with the exception of manual sample collection and handling sequences. Many such systems have been introduced in the pulp and paper industry lately. Fully automatic sampling systems controlled by a process computer, which are expensive and may require frequent maintenance, are used in the case of on-line pulp analysers.
Few attempts have been made to assess the reliability of sampling in connection with an experimental arrangement. The idea of examining the sampling of pulp slurry arose from the ambiguous results obtained in the screening experiments when manual ball valves were used. These were consequently replaced with semi-automatic piston valves, making it possible to compare the two sampling methods. The following sections will consider the reliability of experimental work in general and particularly of the experiments performed here
6.4.2. Sampling methods and phenomena affecting sampling
In general, as many as possible of the following properties are required for an adequate sampling system:
The sample should be a reliable representative of the process.
The result should be insensitive to process variables such as consistency, flow velocity and pressure.
The system should be simple and sampling easy, quick and safe.
The volume of sample should be controllable.
The result should be independent of the user.
It should be possible to flush the system with water.
The system should not become plugged.
The investment and maintenance costs of the sampling system should not be too high.
The devices used for sampling pulp slurries are usually ball valves or piston valves of some kind. Three types of flow geometry for sampling, denoted as S1, S2 and S3, are illustrated in Fig. 38. S1 represents a common mounting on the wall of the pipe, S2 involves a mounting where the sampling device extends into the process stream, and S3 represents a mounting for isokinetic sampling which is based on parallel and equal flux velocities in both the sampling pipe and the process pipe.
Reisto (1990), studying flow geometries for sampling bleached kraft pulp (from Scot pine), found statistically significant, systematic dilution effects on the sample concentrations. The sample flow geometry as such, the process pressure and the flow velocity in the process pipe had an influence on the relative dilution of the sample. The flow geometry S1, for instance, always produced markedly lower consistency values than S2, the difference being within the range 5–20%, with an average of 11%. Any increase in main flow velocity or sampling flow velocity, which depended on the process pressure, increased the dilution of the sample in both flow geometries S1 and S2. The average relative difference between these two geometries seemed to be independent of the consistency level in the experiments.
Karaila (1998) later provided a qualitative explanation for the physical background to Reisto’s findings. His theory was based on the law of the conservation of momentum. During a change in flow velocity and direction, the particles must dissipate momentum in the direction of the main flow and gain new momentum in the direction of the sample flow. The only phenomenon changing the momentum of each particle is the hydrodynamic force, which is only generated if there are velocity differences between the particles and the fluid, which means that excess water must pass by the particles and into the sample pipe. For a sampling system in which the main flow and the sample flow are perpendicular to each other (as in S1 and S2), the sample is thus always diluted to some extent. If isokinetic sampling is used, there is no change in the flow velocity or flow direction and therefore the sample does not have to be diluted.
The dilution effect due to the change in particle momentum does not explain the experimental findings, however, which showed that the S1 flow geometry always produces a lower sample concentration than S2. One possible explanation is that the velocity gradient near the solid layer may cause fibres to become oriented parallel to the direction of flow, causing variable anisotropy in the suspension (Zirnak et al. 1994), which is manifested in changing apparent spatial viscosity in the pulp. In the case of pipe flow, the apparent radial viscosity of the pulp is greater at the wall than near the process pipe axis, where the fibres are more randomly oriented. As a consequence, samples extracted at the wall have a lower concentration than samples extracted at the pipe axis. This may explain why it is advantageous to mount the sampling probe within the process stream, as in the S2 type of flow geometry (Fig. 38). An additional possible explanation may be that the local consistency is higher near the pipe axis than near the pipe wall, as measured by Sanders & Meyer (1971).
The theory contains a number of simplifications, such as morphological and spatial (three-dimensional) homogeneity of the fibre suspension, straightforward flow behaviour and the absence of interaction between fibres or flocs. In practice the situation is more complicated. Fibre suspensions have physical properties that cause special forms of flow behaviour, and these can also be affected by the morphology of the fibres, their concentration and flow conditions.
The flow geometry of the process pipe together with the flow conditions may cause mixing or segregation of fibres due to plug disruption after a major disturbance in the flow (Moller & Norman 1975). The former increases the local spatial homogeneity of fibre suspensions and the latter reduces it. The typical factors affecting flow disturbances include pipe fittings, such as bends, tees, constrictions and expansions, and process equipment, such as pumps, valves and in-stream measuring instruments. Julkunen & Kukko-Liedes (1982), for instance, have shown the existence of spatial consistency variations in pipe flow after valves.
Wood fibre suspensions are morphologically heterogeneous, which affects both anisotropy and local consistency variations. The degree of heterogeneity depends on the wood species being processed, the pulping process (mechanical or chemical) and the further processing of the fibres (e.g. beating). Suspensions consist of loose fibrils and cell wall fragments, broken and whole fibres and fibre bundles, and the behaviour of such suspensions depends on the proportions and physical properties of these fractions.
Additionally, a suspension may contain non-wood material, such as fillers or entrained air, which also change the rheological properties of the pulp suspension and may cause spatial heterogeneity because material segregates or settles in the pipe. As heterogeneity increases, the reliability of sampling inevitably decreases.
The flow characteristics of a suspension at low fibre consistencies are very close to those of water. If the fibre concentration is high enough, the fibres will form a network structure with elastic characteristics and measurable strength. This network will not be continuous, however, but rather will be formed of flocs loosely connected to each other. As the fibre concentration increases, the network becomes more uniform and stronger. Consequently some water seeps through the fibre plug during sampling, because the network structure tends to keep the fibres bound in the plug.
In conclusion, the explanation for the failure of pulp slurry sampling may well be a combination of four factors:
a dilution effect due to the flow geometry of the sampling system and process variables,
anisotropy of the fibre suspension, which depends on the fibre properties and velocity gradients,
local consistency changes caused by the flow conditions and fittings of the process pipe, and
network strength, which results from the consistency and fibre properties.
6.4.3. Analytical error
Analysis error is a combination of several error components, the magnitudes of which are often unknown, as are their distributions, although a normal distribution is generally a good enough assumption. The components of the error may be systematic or stochastic. A systematic error may be due to a fault in the construction or location of the sampling device, while a stochastic error may be caused by a fluctuation in the process and/or by the human factor. A fluctuation in the process may also cause a systematic error, however, if both the fluctuation and the sampling frequency are regular.
Errors in different phases of the analytical procedure will prove cumulative in the final result. Provided that the error components are independent, the uncertainty attached to the result can be determined by the sum of the variances of the components, as shown in Eq. 30. Only three main components are presented in the equation, but they can be divided into several sub-components if necessary.
If one wants to increase the precision of an analytical procedure, the variances of these three components in Eq. 30 can in principle be minimised. Since there are standardised methods which define how to measure the consistency of pulp slurry in the laboratory, e.g. TAPPI T-240 and SCAN-M1:64, the accuracy of analyses depends mainly on the proficiency of the laboratory technician, and the scope for reducing the variance in this component is thus limited. It is only possible to reduce the level of random or regular variation in a process by improving the process itself or its control, and this is not always possible, or at least not practicable. Thus the only means left is to improve the sampling procedure. This can be done by taking a statistically sufficient number of samples, replacing the sampling device with a better one, or adjusting the location of the device.
6.4.4. Manual vs. semi-automatic sampling
The determination of quantitative error in the sampling of pulp slurries is always problematic because the exact values for consistency and other pulp properties are unknown. Use of the mass balance error is one not a perfect way of doing this, but it is probably the best for conducting comparable tests on different sampling systems. One disadvantage is the difficulty in identifying sampling errors and error sources if the consistencies, flow velocities, pipe diameters and fibre properties change in the course of the process. Another disadvantage is that systematic effects caused by the flow geometry of the sampling device may not become apparent as a whole, because the error may be parallel and of the same magnitude in each flow. In addition, possible errors in flow measurement and errors due to process fluctuation cannot easily be distinguished from those due to the sampling device. Despite such minor disadvantages, the mass balance error was used here as the research method.
Analysis of the experimental data confirmed that semi-automatic piston valves were superior to manual ball valves. This was obviously attributable to both the bias and standard deviation in mass balance. The bias in mass balance with the manual valves was probably due to the flow geometry together with changing process conditions, which caused a considerable systematic dilution effect on the pulp samples. When piston valves were used, no significant bias occurred. Their construction and mounting within the process streams hence seemed to be feasible and sampling was independent of the moderate changes in process conditions. The difference in the standard deviations of the mass balance error between the manual and semi-automatic valves was presumably mainly due to the human factor, because the other error sources — such as the laboratory analyses, the flow measurements and process fluctuations — can be considered to be equal in both test series. The sampling flow rate could be controlled only approximately using the manual valves, while the use of semi-automatic valves nearly always resulted in a constant sampling flow rate that depended almost exclusively on the pressure in the process pipe, which did not change much in these experiments. Understandably, the flow conditions in the sampling valves were more stable and the deviation was smaller. In addition, the mounting of the manual valves at the side of the stream might render them more sensitive to changes in process conditions than the mounting of semi-automatic valves within the stream.
The results indicated that the sampling of pulp slurry with semi-automatic valves produced systematically reliable results and not even moderate changes in the process conditions could detract from the reliability of sampling. The larger the number of samples taken, the more obvious is the difference between the manual and semi-automatic sampling methods, because the latter is insensitive to the human factor. The sampling results obtained here corresponded to the earlier observations of Reisto (1990), although a different research method was used and the results were therefore only qualitatively comparable.
The example calculation of the number of repetitions shows that, considerably less samples are required to reach the same precision in determining the mass reject rate with a proper sampling system than if an inadequate one is used. Systematic error in the sampling devices was not taken into account in these calculations, but in the case of the manual ball valves both the level of bias and the variation are dependent on the user of the valves, so that the difference between the manual and semi-automatic sampling systems may be a pronounced one in practice. The calculated numbers of samples are only valid for this particular case and with the assumptions that were made here. The pulp type and consistency may affect the deviation in the consistency analysis, for example, and thereby the number of samples needed. In addition, the standard deviation in the consistency analysis in many laboratories may be 3 to 4 times greater than that used in these calculations, which means that more samples will have to be taken to achieve the same level of certainty as calculated above.
Although it was not feasible to investigate directly what sampling error in consistency affected the other pulp properties, i.e. freeness, fibre length distribution etc., it can be assumed that if there was a deviation in consistency there would also be a deviation in fibre properties. The deviation in pulp properties probably occurred in proportion to that in consistency. It is possible, of course, that the sampling devices themselves may act as screens, as proposed in a recent paper (Ämmälä et al. 2000a). It was not possible within the scope of this work to ascertain completely what is the effect of sampling on the deduced fibre properties, an issue that would need further investigation with a proper experimental arrangement.
In conclusion, most of the variation in the screening results could be attributed to the sampling error, especially in the first part of the experimental work, where manual valves were employed. Replacing the manual sampling valves with semi-automatic ones led to considerably better reliability in the experimental work, and this together with the large number of screening experiments performed here suggests that the results can be considered reliable at a high level of confidence. The probability of misinterpreting the screening results could therefore be considered to be minimal.