GB2553018A

GB2553018A - Radiation intensity measurement apparatus for nucleonic level or density gauge

Info

Publication number: GB2553018A
Application number: GB1707932.8A
Authority: GB
Inventors: Metcalfe Benjamin; John Roe Stephen; Bryan Storey Paul
Original assignee: Johnson Matthey PLC
Current assignee: Johnson Matthey PLC
Priority date: 2016-05-20
Filing date: 2017-05-17
Publication date: 2018-02-21
Anticipated expiration: 2037-05-17
Also published as: GB201707932D0; GB201608887D0; WO2017199027A1; GB2553018B

Abstract

A radiation measurement apparatus for use in a nucleonic level or density gauge comprises a radiation detector and a control unit receiving radiation measurements from the radiation detector. The control unit is calculates an exponential moving average over time with a first time constant and a linear moving average over time with a smaller second time constant. The control unit selects either the exponential moving average or the linear moving average to use in determining a level or density based on the criterion that if difference of the moving averages is more than a threshold the linear moving average is used. The threshold may be 2 to 4 times the standard deviation, with the further criterion of the difference lasting for more than a predetermined length of time.

Description

(54) Title of the Invention: Radiation intensity measurement apparatus for nucleonic level or density gauge Abstract Title: Radiation Intensity Measurement for Nucleonic Level or Density Gauge (57) A radiation measurement apparatus for use in a nucleonic level or density gauge comprises a radiation detector and a control unit receiving radiation measurements from the radiation detector. The control unit is calculates an exponential moving average over time with a first time constant and a linear moving average over time with a smaller second time constant. The control unit selects either the exponential moving average or the linear moving average to use in determining a level or density based on the criterion that if difference of the moving averages is more than a threshold the linear moving average is used. The threshold may be 2 to 4 times the standard deviation, with the further criterion of the difference lasting for more than a predetermined length of time.

At least one drawing originally filed was informal and the print reproduced here is taken from a later filed formal copy.

08 17

9000 /10

ο

120

08 17

2/ 10

(Not embodying the invention)

FIG. 3

3/10

08 17

4/10

08 17

200 300

Time (seconds)

5/10

08 17

6/10

08 17

..........Input — Output — UpperSigma — LowerSigma

200 300

Time (seconds)

7/10

08 17

200 300

Time (seconds)

8/10

08 17

WO 200 300 400 500

Time (seconds)

(PRIOR ART)

9/10

08 17 “Ο c:

ο ο

φ co u_

Φ

CX co εζ:

Ζ5

Ο

Ο φ -«—» 2 CD c:

ο

Ο

100

200 300 400 500

Time (seconds)

10/10

08 17

-1RADIATION INTENSITY MEASUREMENT APPARATUS FOR NUCLEONIC LEVEL OR DENSITY GAUGE

Field ofthe Invention

The present invention relates to radiation measurement apparatus, in particular radiation intensity measurement apparatus for use in nucleonic level or density gauges. The invention also relates to nucleonic level or density gauges and to methods of measuring fluid level or density.

Background

It is well known to use radiation to measure the level or density of fluids. For example arrays of radiation sources may be arranged opposite arrays of radiation detectors so that the radiation passes from the sources to the detectors through the fluid to be measured. Since the attenuation of the radiation during its passage from the sources to the detectors is determined by the density of the fluid between the sources and the detectors, the measured intensity arriving at the detectors can be used to calculate the density ofthe fluid. The density thus calculated can either be used directly or can be used to determine fluid levels by comparing the densities determined at different heights. An apparatus that uses such an approach is disclosed for example in US6633625. Many other density and level gauge arrangements are also known in the art.

Regardless ofthe arrangement and number of sources and detectors, all such nucleonic instruments must cope with the statistical variations that are inherently present in radiation-based measurements. Those variations arise from the random nature ofthe radioactive decay ofthe sources, and the random nature ofthe interactions ofthe radiation with the fluid through which it passes. Nucleonic instruments are typically used on industrial plant for monitoring and control purposes and the statistical variations can result in noisy measurements, which are not useful for such purposes. As a result it is usual to operate some form of smoothing ofthe recorded intensities so as to produce a more stable measurement. When a process is at steady state, the intensities can simply be averaged overtime to produce a stable value. An exponential moving average is typically used to provide a computationally efficient average. The time constant ofthe average (i.e. a value representative ofthe time over which the average is taken) can be selected to be sufficiently long that a required precision is reached for the measurement. However, in a dynamic process the situation is more complex. A long time constant will produce measurements with a high degree of precision during times of relative stability ofthe process, but will be slow to react to changes in the process. That may be unacceptable since changes in the process will often require control responses without undue delay. For example, a typical requirement might be that the measurement system should react quickly enough to a step change in the process that the measured value has achieved 90% ofthe change in the process value within 10s ofthe change occurring (this would be referred to as the measurement system having a Tgo of 10s). An

-2alternative would be to have a short time constant. In those circumstances, the response to change will be rapid, but the measurement will be susceptible to the statistical variations and thus will not give a sufficiently precise value at times of process stability.

In order to provide a reasonable dynamic response along with a reasonable precision, it is known to use two exponential moving averages with different time constants. At times of stability the average with the longer time constant is used, while during dynamic situations the average with the shorter time constant is used. The selection of which average to use is based upon the difference between the two averages. A relatively small deviation between the two is assumed to be the result of the susceptibility of the shorter average to statistical variations that do not reflect any underlying change in the process conditions (e.g. a change in the fluid level), while a larger deviation is taken as indicating that the process conditions have changed and that the shorter average has reacted to that change while the longer average has not yet reacted. Thus in the first case the longer average is used and in the second the shorter average is used.

The prior art systems can produce a reasonable balance between response and precision in many processes. However, they may not reliably produce a target Tgo response time, particularly in very noisy systems such as heavy oil processes where the count rate is low.

Another important criterion may be the number of false excursions of the measured value. A false excursion occurs when the apparatus selects the shorter time constant moving average (thus assuming that the process variable has changed significantly) when in fact the process variable has not changed significantly and the shorter time constant moving average differs from the longer time constant moving average due to the sensitivity of the shorter time constant moving average to the statistical noise in the measurement. Prior art systems may be unable to improve the Tgo response without also producing an unacceptably large number of false excursions.

Preferred embodiments of the present invention seek to overcome one or more of the above disadvantages of the prior art. In particular, preferred embodiments of the present invention seek to provide an improved system for radiation intensity measurement in nucleonic level or density gauges.

Summary of Invention

According to a first aspect of the invention, there is provided a radiation measurement apparatus for use in a nucleonic level or density gauge, the measurement apparatus comprising a radiation detector and a control unit configured to receive radiation measurements from the radiation detector, wherein the control unit is configured to use the radiation measurements to calculate an exponential moving average overtime with a first time constant and a linear moving average over time with a second time constant, the second time constant being smaller than the first time

-3constant, and to select either the exponential moving average or the linear moving average to use in determining a level or density based on the criterion that if the linear moving average differs from the exponential moving average by more than a pre-determined threshold the linear moving average is used and, if not, the exponential moving average is used.

The invention thus differs from the prior art systems in that the shorter average is a linear moving average filter. We have found that the combination of a short time constant linear moving average with a long time constant exponential moving average and decision logic to select one or the other based on the difference between them results in a particularly advantageous system for nucleonic level and density gauges. The linear moving average has a quicker rise time compared to an exponential moving average with the equivalent smoothing capacity and thus using a linear moving average during dynamic changes will advantageously deliver the measured value to its new final value in a quicker time. Meanwhile, maintaining the exponential moving average with the longer time constant at times of relative stability results in a precise measurement at those times and maintains the advantage of computational efficiency. In particular, the combination of the different types of average results in an improvement to the important Tgo response of the apparatus, without compromising measurement stability. While the response of a short time constant exponential moving average may be similar to a comparable linear moving average for less-stringent criteria such as T50 (the time taken for the measured value to respond to 50% of a change in the process value), the linear moving average provides superior performance on the T90 criterion. For modern control processes, the T90 response may be the more important criterion and the radiation measurement apparatus of the present invention may therefore be superior to prior art measurement apparatus.

Preferably the measurement apparatus is a radiation intensity measurement apparatus. Preferably the control unit is configured to receive radiation intensity measurements from the radiation detector. Preferably the smoothing occurs before the intensity measurements are used to calculate the level or density. In that way the measurement apparatus outputs a smoothed radiation intensity measurement, which can then be used in any desired calculation. Thus the control unit is preferably configured to calculate an exponential moving average intensity overtime with a first time constant and a linear moving average intensity overtime with the second time constant, the second time constant being smaller than the first time constant, and to select either the exponential moving average intensity or the linear moving average intensity to use in a level or density calculation based on the criterion that if the linear moving average intensity differs from the exponential moving average intensity by more than a pre-determined threshold the linear moving average intensity is used and, if not, the exponential moving average intensity is used.

Preferably the criterion is that if the linear moving average differs from the exponential moving average by more than the pre-determined threshold for more than a pre-determined length of time

-4the linear moving average is used and, if not, the exponential moving average is used. By requiring the linear moving average to differ from the exponential moving average by more than the predetermined threshold for more than a predetermined time the possibility of false excursions is reduced. Such a system may be particularly advantageous in the control of heavy oil systems such as desalters, where the signal to noise ratio may be low. In such systems, the possibility of false excursions due to the noise is increased. The possibility of false excursions could be reduced by increasing the time constant of the linear moving average. However, there is a limit to how long the time constant of the linear moving average can be if a specified Tgo response is to be achieved.

The apparatus of the invention provides advantages over the prior art in that respect because the linear moving average may have a longer time constant and still achieve a desired Tgo than the prior art exponential moving average filter. However, further improvements can be made by requiring the linear moving average to differ from the exponential moving average by more than the pre-determined threshold for more than a predetermined time. The predetermined time should be less than the target Tgo and therefore may be less than the second time constant. For example the predetermined time may be not more than half of the second time constant. The predetermined time may be at least 0.05 times the second time constant or at least 0.1 times the second time constant. Preferably, the predetermined time is expressed as a number of consecutive measurements. Preferably the predetermined time is from 2 to 10 consecutive measurements, more preferably from 3 to 5 consecutive measurements, most preferably from 3 to 4 consecutive measurements. In some embodiments, the predetermined time may be at least two consecutive measurements, preferably at least three consecutive measurements, and not more than half of the second time constant. The predetermined time is small enough that a genuine change in the underlying process state (for example, a fall in the fluid level) will be reliably detected within the target Tgo time, but long enough that an excursion due to noise is unlikely to result in a false measurement due to the statistical likelihood that an excursion due to noise will be reversed by another statistical fluctuation in the opposite direction.

Preferably the predetermined threshold is based on the standard deviation of the linear moving average. For example the pre-determined threshold may be a value between 2 and 4 times the standard deviation of the linear moving average. In that way the absolute value of the threshold is increased when the uncertainty in the linear moving average value is increased.

Preferably the first time constant is not less than 20 seconds. More preferably the first time constant is not less than 60 seconds or not less than 100 seconds. The first time constant is preferably chosen on the basis of a required precision for the measurement. It is an advantage of the present invention that the improved dynamic response of the linear moving average allows the first time constant to be increased, thus improving the precision of the measurement during periods of steady operation. Preferably the second time constant is not more than 15 seconds. More

-5preferably the second time constant is not more than 10 seconds. The second time constant is preferably chosen based on the desired Tgo response of the apparatus.

According to a second aspect of the invention there is provided a nucleonic level or density gauge comprising a radiation measurement apparatus according to the first aspect. The nucleonic level or density gauge may be a gauge for use in a heavy oil process. Preferably the level or density gauge is for use in a desalter. Preferably the level or density gauge comprises a source of radiation and a radiation measurement apparatus according to the first aspect. The gauge may comprise a plurality of radiation detectors. The control unit may be the same for all of the radiation detectors. There may be a control unit for each of the radiation detectors. The gauge may comprise a plurality of sources of radiation arranged in an array and a plurality of detectors held in a second array, the sources of radiation being associated with collimators so that a beam of radiation extends from each source to a corresponding detector. The source array and the detector array may be contained in dip pipes for insertion into a vessel. The source array and the detector array may be held in a fixed relationship to one another. The gauge may comprise more radiation detectors than sources. In that case, the source or sources may be collimated so that radiation from a source passes to more than one detector.

According to a third aspect of the invention there is provided a method of measuring fluid level or density comprising measuring a level of radiation passing through the fluid, using the measured radiation level to calculate an exponential moving average overtime with a first time constant and a linear moving average overtime with a second time constant, the second time constant being smaller than the first time constant, and using either the exponential moving average or the linear moving average to determine the level or density wherein if the linear moving average differs from the exponential moving average by more than a pre-determined threshold the linear moving average is used and, if not, the exponential moving average is used.

Preferably the intensity of radiation passing through the fluid is smoothed and the smoothed intensity used to calculate the level or density. The method thus preferably comprises measuring radiation intensity of radiation passing through the fluid, calculating an exponential moving average intensity overtime with a first time constant and a linear moving average intensity overtime with a second time constant, the second time constant being smaller than the first time constant, and using either the exponential moving average intensity or the linear moving average intensity to calculate the level or density wherein if the linear moving average intensity differs from the exponential moving average intensity by more than a pre-determined threshold the linear moving average intensity is used and, if not, the exponential moving average intensity is used.

Preferably the radiation intensity is measured as a number of counts received over a time window, with a series of radiation intensity measurements being generated one after the other over a period

-6of time. The time between measurements, which also corresponds to the time over which counts are recorded for an individual measurement, may be referred to as the sampling period, s. Preferably the exponential moving average, If, at any given time is calculated according to the formula:

Where l_P is the previous exponential moving average, l_c is the current intensity measurement and F is a filter factor which is related to the time constant T_c by the relationship F = y + 0.5. The linear moving average, If, may be calculated according to the formula:

γη-l j , _ ^i=0 ‘c-l if — n

In other words, the linear moving average is the average of the last n values. The time constant of the linear moving average is thus given by T_c = ns. Preferably the linear moving average is computed recursively, for example using the formula:

Ip — Ip — Ic-n + lc

Thus the current linear moving average is the previous linear moving average, l_P, minus the oldest measurement in that previous linear moving average, l_c-_n, plus the current measurement, l_c. It may be that the apparatus comprises a programmable controller having a moving average function available as a standard function. In that case the built-in function may be used.

Preferably the fluid level is the level of an oil-water interface in a desalter. Desalters may be characterised by thick walls and heavy oils, which combine to reduce the signal received by radiation detectors. The method of the invention may be particularly suited to capturing dynamic changes in desalters whilst also providing reliable measurement of steady state levels.

It will be appreciated that features described in relation to one aspect of the invention may be equally applicable in another aspect of the invention. For example, features described in relation to the apparatus of the invention, may be equally applicable to the method of the invention, and vice versa. Some features may not be applicable to, and may be excluded from, particular aspects of the invention.

Description of the Drawings

Embodiments of the present invention will now be described, byway of example, and not in any limitative sense, with reference to the accompanying drawings, of which:

Figure 1 is a histogram of the smoothed count rate response times to 49,999 step inputs containing random noise to a prior art system;

Figure 2 is a histogram of the smoothed count rate response times to 49,999 step inputs containing random noise to a system embodying the present invention;

-7Figure 3 is a histogram ofthe smoothed count rate response times to 49,999 step inputs containing random noise to a system not embodying the present invention;

Figure 4 is a histogram ofthe smoothed count rate response times to 49,999 step inputs containing random noise to a system embodying the present invention;

Figure 5 is a histogram ofthe smoothed count rate response times to 49,999 step inputs containing random noise to a system embodying the present invention;

Figure 6 is a plot of count rate against time for a series of step changes in input count rate containing random noise to a prior art system;

Figure 7 is a plot of count rate against time for a series of step changes in input count rate containing random noise to a system embodying the present invention;

Figure 8 is a plot of count rate against time for a series of step changes in input count rate containing random noise to a prior art system;

Figure 9 is a plot of count rate against time for a series of step changes in input count rate containing random noise to a system embodying the present invention;

Figure 10 is a plot of count rate against time for a series of step changes in input count rate containing random noise to a prior art system;

Figure 11 is a plot of count rate against time for a series of step changes in input count rate containing random noise to a system embodying the present invention; and Figure 12 is a plot of count rate against time for a series of step changes in input count rate containing random noise to a system embodying the present invention.

Detailed Description

For comparison to the present invention, 49,999 step inputs containing random noise were passed through a prior art control system to observe the response. The control system calculated two exponential moving averages. The first had a time constant of 20s and the second had a time constant of 4s. That configuration was intended to deliver a tgo response of less than 10s. The control system used the first exponential moving average (the “long” average) as the output unless the second exponential moving average (the “short” average) differed from the first exponential moving average by more than 2 standard deviations based on the short average. The time taken for the output to react to the change in the step input was recorded based on the time taken for the output to achieve 90% ofthe step change (the tgo response) and the time taken for the output to achieve 95% ofthe step change (the tgs response). In figure 1, the response time is on the x-axis and the frequency of achieving the response time is on the y-axis. The tgo (1) and tgs (2) responses are plotted. The average tgo was 12.22s, with a minimum of 4s and a maximum of 45s. The tgo response was below 10s in 25,544 out of 49,999, or 51 % of cases. The average tgs was 17.52s, with a minimum of 4s and a maximum of 78s. The tgs response was below 10s in 14,829 out of 49,999, or 30% of cases.

-8The same test was run using a control system according to the present invention, which calculated an exponential moving average having a time constant of 20s and a linear moving average having a time constant of 7s. The linear moving average having a time seconds of 7s gives equivalent smoothing performance to the 4s exponential filter. The control system used the exponential moving average as the output unless the linear moving average differed from the exponential moving average by more than 2 standard deviations based on the linear moving average. In figure 2, the response time is on the x-axis and the frequency of achieving the response time is on the yaxis. The tgo (3) and tgs (4) responses are plotted. The average tgo was 8.35s, with a minimum of 5s and a maximum of 41s. The tgo response was below 10s in 41,234 out of 49,999, or 82% of cases. The average tgs was 11.80s, with a minimum of 5s and a maximum of 75s. The tgs response was below 10s in 34,571 out of 49,999, or 69% of cases.

Comparing the two outcomes, it can be seen that the control system according to the present invention had a superior performance to the prior art system. In the control systems compared, the response may be judged against a target of tgo < 10s. Such a target may, for example, be important for the stable control of levels in a vessel. It can be seen that the system of the invention achieved this target in 82% of cases compared to 51 % of cases for the prior art system.

An advantage of the present invention is that it may permit the use of more advanced logic by the control system in choosing between the short and long averages. For example, the control system of the invention may use the long, exponential moving average unless the short linear average has differed from it by at least a predetermined amount for at least a predetermined length of time. For example, when the intensity data is received as a series of readings, the logic may require the control system to use the exponential moving average unless the linear moving average has differed from the exponential moving average by more than a set number of standard deviations over a number of consecutive readings. That may permit optimisations of the control system performance that were not possible with prior art systems. To demonstrate this, a system not embodying the invention uses two exponential moving averages, as in the prior art systems, one having a time constant of 20s and one having a time constant of 4s. The control system uses the first exponential moving average unless the second differs from it by more than 3 standard deviations for at least 4 measurements. The tgo (5) and tgs (6) response are plotted in figure 3 in the same way is in figures 1 and 2. The average tgo was 14.53s, with a minimum of 5s and a maximum of 48s. The tgo response was below 10s in 13,888 out of 49,999, or 28% of cases. The average tgs was 27.25s, with a minimum of 5s and a maximum of 75s. The tgs response was below 10s in 1583 out of 49,999, or 3% of cases. Thus it can be seen that modifying the prior art to include a so-called ‘switch delay’, whereby the short average is only used if it has differed from the long average for a pre-determined length of time, results in worse performance in terms of average response and achieving the tgo < 10s target. By contrast, a control system according to the invention is shown in figure 4. As with the system of figure 2, the exponential moving average has a time constant of 20s

-9and the linear moving average has a time constant of 7s. However, in figure 4 the control system uses the exponential moving average unless the linear moving average has differed from the exponential moving average by 3 standard deviations for 4 measurements. The tgo (7) and tgs (8) response are plotted in figure 4 in the same way is in figures 1,2 and 3. The average tgo was 6.64s, with a minimum of 5s and a maximum of 29s. The tgo response was below 10s in 49668 out of 49,999, or 99% of cases. The average tgs was 7.57s, with a minimum of 5s and a maximum of 61s. The tgs response was below 10s in 45,934 out of 49,999, or 92% of cases. In contrast to the prior art systems therefore, including a switch delay in the control system of the present invention results in a significant improvement in the response.

A further advantage of the control systems of the present invention is illustrated in figure 5. In this case the control system is as in figure 4, but the time constant of the exponential moving average filter has been increased to 120s. The tgo (9) and tgs (10) response are plotted in figure 5 in the same way is in figures 1,2, 3 and 4. The average tgo was 12.21s, with a minimum of 5s and a maximum of 123s. The tgo response was below 10s in 39,206 out of 49,999, or 78% of cases. The average tgs was 30.16s, with a minimum of 5s and a maximum of 199s. The tgs response was below 10s in 31,515 out of 49,999, or 63% of cases. The advantage of a long time constant on the exponential moving average is that it provides a more stable reading during steady state conditions. Comparing with figures 1 and 2, the control system of the present invention enables the length of the long exponential moving average to be increased in this way without significantly impacting on the tgo < 10s target and while still providing improved performance with respect to that target as compared to the prior art systems.

In figure 6 a prior art control system is shown responding to a series of step changes in an input 12. The output 14 is selected from a first exponential moving average having a time constant of 20s and a second exponential moving average having a time constant of 4s. This is the same control system as in figure 1. In figure 6 the boundaries of two standard deviations above the second exponential moving average (the upper boundary 11) and two standard deviations below the second exponential moving average (the lower boundary 13) are also shown. It can be seen that the control system acts by returning an output 14 corresponding to the first exponential moving average, until that output crosses either of the boundaries 11 and 13. That typically occurs during the step changes in the input 12. When that occurs the output switches to the shorter exponential moving average, resulting in a correction of the output 14 back towards the centre of the ‘carriageway’ formed between the upper 11 and lower 13 boundaries. Such a correction can be seen for example at 14’.

Referring now to figure 7, a control system according to the invention is used. In this case an input 16 again includes random noise and a series of step changes. The control system now selects between an exponential moving average having a time constant of 20s and a linear moving

-10average having a time constant of 4s. The switch between the two is made when the linear moving average differs by more than two standard deviations from the exponential moving average. This is shown in figure 7 by the upper boundary 15 and the lower boundary 17. The output 18 lies within the boundaries 15 and 17 until a step change occurs. However, by comparison with figure 6, the carriageway between the upper 15 and lower 17 boundaries tracks the step change more closely across the full range of the step change. Thus where the carriageway has a curved transition out of the step change in the prior art systems, in the system of the invention, the transition is a sharp change. That has the advantage that the output 18, which was able to pass around the edges of the curved transition in some instances under the prior art system, is forced to respond more completely to the step change in order to stay within the carriageway of the present invention. As a result, the tgo response of the present invention is improved.

In figures 8 and 9 the prior art control system of figure 6 and the inventive control system of figure 7 are used again, but now with the time constant of the second exponential moving average of the prior art control system and the linear moving average of the inventive control system being increased to 10s. In figure 8 it is clear that the upper boundary 19 and the lower boundary 21 allow significant curving of the carriageway in which the output 22 is confined in the prior art system.

Thus when the input 20 undergoes a step change, while the initial reaction is quick, the time for the output to correctly arrive at the new level is poor. This reduces the percentage of cases that achieve the tgo<1Os target. By contrast, the system of the present invention has sharp changes in both the start and the finish of the step changes in the upper 23 and lower 25 boundaries and thus the output 26 is constrained to follow the step change in the input 24 more closely. That results in a much-improved tgo response.

The control system is used in nucleonic level or density gauges. In such instruments changes in the intensity of the radiation are related to changes in level or density of the material in a vessel in which the gauge is operating. Typically the level or density will be an important variable for process control and thus it is important to track changes in that variable. The challenge in many such instruments is that there may be significant noise in the input signal. That is not least because nucleonic instruments detect radiation emitted by a radioactive source. That emission occurs as random events and thus the count rate is inherently noisy because the number of emission events in a given time window is itself uncertain. Because the aim of such instruments is to provide control on the order of seconds, it is not possible to simply collect data over a long period of time in order to even out the random nature of the radioactive emission from the source. Further noise is also present from the detectors, electronics and the process variables themselves such as temperature and pressure. Moreover, the measured intensity is determined by the level of attenuation of the material through which the radiation has passed. However, that attenuation is the result of further random processes such as scattering and thus contributes a further source of noise to the measurements. Thus nucleonic measurements are typically trying to detect significant underlying

-11changes in noisy measurements. The present invention has particular utility in such circumstances because it combines the advantages of averaging the noise out of the measurements when it is appropriate to do so, with reliable tracking of genuine changes.

The challenges are particularly significant when tracking heavy oil levels in processes such as desalters. Such processes may experience rapid changes due, for example to the arrival of slugs in the apparatus and are also characterised by particularly low count rates due to the high density of the heavy oils. Thus there are rapid changes which need to be robustly tracked, but the noise is high due to the lack of counts to smooth out the random nature of the source emissions and scattering events. The system of the present invention has been shown to be particularly advantageous in such processes. Looking at figures 10 and 11, in a prior art system (long exponential average with 60s time constant, short exponential average with 10s time constant, switch if short differs from long by more than two standard deviations) the carriageway between the upper 27 and lower 29 boundaries follows a curving path and barely settles to the new level in the input 28 before the next step occurs. Thus the output 30 is slow to fully react to each of the step changes. By contrast a system according to the invention (exponential moving average with 60s time constant, linear moving average with 10s time constant, switch if linear differs from exponential by more than 2 standard deviations) produces an output 34 which tracks more reliably the step changes in the input 32, without compromising the stability of the measurements in the steady phases between the steps. The carriageway between the upper boundary 31 and the lower boundary 33 (i.e. the +/- two standard deviations region, the boundaries 31 and 33 of which trigger the switch to the linear moving average) tracks the step changes in the input 32 more sharply than the prior art boundaries 27 and 29 thus improving the response time. In such systems it would be typical to target a teo response of 10s. That is achieved with the present invention but not with the prior art.

The present invention also advantageously reduces the number of so-called ‘false excursions’. False excursions occur when the control system erroneously treats a random variation resulting from the noise in the measurement as a change resulting from a change in the process (for example, the level or density) being measured. It will be appreciated that a system could be designed to respond very quickly to changes in the process, for example by carrying out no smoothing, but that such a system would give very poor stability in periods in which the process is unchanged. The advantage of the control system of the invention is that it combines satisfactory response to changes in the underlying process with good measurement stability at all other times.

In figure 12, the system of figure 11 is shown controlling a process over a longer length of time.

The upper boundary 35, lower boundary 37, input 36 and output 38 are shown. Occasionally the output demonstrates a false excursion 39 where the output incorrectly switches to the linear moving average to track a change that is actually simply an artefact of the random noise. The presence of such false excursions 39 can be reduced or even eliminated by altering the logic of the

-12decision by which the control systems selects that average to output. For example, the control system may be configured to only switch to the linear moving average if the linear moving average has differed from the exponential moving average by the predetermined threshold (in this example, two standard deviations) for a predetermined length of time. For example, the false excursions 39 could be eliminated by only switching to the linear moving average if the linear moving average is outside the carriageway between the upper boundary 35 and the lower boundary 37 for three consecutive measurements or 4 consecutive measurements. The advantage of the system of the present invention is that, because the carriageway tracks more closely the changes in the process, logic that requires the measurement to leave the carriageway for more than one measurement before switching to the linear moving average can be implemented without increasing the response time beyond acceptable levels.

It will be appreciated by persons skilled in the art that the above embodiments have been described by way of example only, and not in any limitative sense, and that various alterations and modifications are possible without departure from the scope of the invention as defined by the appended claims.

Claims

1. A radiation measurement apparatus for use in a nucleonic level or density gauge, the measurement apparatus comprising a radiation detector and a control unit configured to receive measurements from the radiation detector, wherein the control unit is configured to use the measurements to calculate an exponential moving average overtime with a first time constant and a linear moving average overtime with a second time constant, the second time constant being smaller than the first time constant, and to select either the exponential moving average or the linear moving average to use in determining a level or density based on the criterion that if the linear moving average differs from the exponential moving average by more than a pre-determined threshold the linear moving average is used and, if not, the exponential moving average is used.

2. A radiation measurement apparatus according to claim 1, wherein the criterion is that if the linear moving average differs from the exponential moving average by more than the predetermined threshold for more than a pre-determined length of time the linear moving average is used and, if not, the exponential moving average is used.

3. A radiation measurement apparatus according to any preceding claim wherein the predetermined threshold is based on the standard deviation of the linear moving average.

4. A radiation measurement apparatus according to claim 3, wherein the pre-determined threshold is a value between 2 and 4 times the standard deviation of the linear moving average.

5. A radiation measurement apparatus according to any preceding claim wherein the first time constant is not less than 20 seconds.

6. A radiation measurement apparatus according to any preceding claim wherein the second time constant is not more than 15 seconds.

7. A nucleonic level or density gauge comprising a radiation measurement apparatus according to any preceding claim.

8. A method of measuring fluid level or density comprising measuring radiation intensity of radiation passing through the fluid, using the measured radiation intensity to calculate an exponential moving average over time with a first time constant and a linear moving average overtime with a second time constant, the second time constant being smaller than the first time constant, and using either the exponential moving average or the linear moving average to determine the level or density based on the criterion that if the linear moving average differs from the exponential moving average by more than a pre-determined threshold the linear moving average is used and, if not, the exponential moving average is used.

9. A radiation measurement apparatus substantially as herein described with reference to the accompanying figures.

10. A method substantially as herein described with reference to the accompanying figures.

Intellectual

Property

Office

GB 1707932.8

1 - 10