impulse line blockage in differential pressure sensors, and a novel coriolis mass flow meter prototype with SEVA capabilities. 2. THE SELF-VALIDATING (SEVA) ...
th
4 Practical Developments in Gas Flow Metering Workshop
SENSOR VALIDATION IN GAS FLOW MEASUREMENT Dr. Manus Henry, Oxford University
1
INTRODUCTION
Over the last twelve years, Oxford University and the Foxboro Company have developed the concept of the Self-Validating (SEVA) sensor (Henry and Clarke, 1993). Papers, talks and industrial trials have resulted in a consensus among industrial users and vendors that the SEVA concept would be valuable in industrial practice. Accordingly over the last year there have been a number of initiatives leading to a British Standard for on-line measurement quality assessment based upon SEVA. It is hoped that this will subsequently be made available at the European (Cenelec) and World (IEC) forums. At the same time, Invensys (the parent company of Foxboro and other automation companies) is developing commercial products based upon SEVA. In addition, research at Oxford is continuing to develop new theory and demonstrations of SEVA. This paper provides an overview of the SEVA concept, describes the recent standardization activities, and gives two examples of validation applied to gas flow instrumentation: detecting impulse line blockage in differential pressure sensors, and a novel coriolis mass flow meter prototype with SEVA capabilities.
2 THE SELF-VALIDATING (SEVA) SENSOR The Sensor Validation Research Group at Oxford began in 1988 to examine the impact of digital technology on instruments, and to local fault detection in particular. It developed a theoretical model of how a ‘self-validating’ or SEVA instrument should behave [1]. This assumes the availability of internal computing power for self-diagnostics, and of digital communications to convey measurement and diagnostic data. A generic set of metrics are proposed for describing measurement quality (see figure below). For each measurement, three parameters are generated: •
The Validated Measurement Value (VMV). This is the conventional measurement, but if a fault occurs, the VMV is a corrected best estimate of the true measurand value.
•
The Validated Uncertainty (VU). This is the metrological uncertainty, or probably error of the VMV. For example, if the VMV is 4.31 l/s, and the VU is 0.05 l/s, then the sensor is claiming that the true measurement value lies between 4.26 l/s and 4.36 l/s with 95% confidence.
•
The Measurement Value Status (MV Status). Given the requirement to provide a measurement, even with a serious fault, the MV Status indicates how the current measurement has been calculated. It takes one of a small set of values, of which the most important are:
VMV Raw Data
SEVA Sensor
VU MV STATUS
}
PARAMETERS FROM SEVA SENSOR
one per measurement
DEVICE STATUS RAW MEASUREMENT DETAILED DIAGNOSTICS
1
CLEAR: the measurement has been calculated normally. BLURRED: live raw data is still being obtained, but the VMV has been corrected for some fault condition. DAZZLED: a temporary state ; it is known that current raw data is uncorrelated with the true process variable (e.g. the input is saturated), but it is not (yet) known whether the condition is permanent. The VMV is projected from past history, and the VU increases with time to reflect the reduced confidence in this projected VMV. BLIND: like DAZZLED except that there is evidence to suggest that the loss of raw data is permanent.
The VMV, VU and MV Status are generated for each measurement output from the sensor. For example, many industrial sensors measure process temperature as well as (say) flow or pressure. The validity of each is distinct, and each will be affected by a fault in a different way. However, for maintenance purposes, a single Device Status parameter is also provided, which indicates the level of maintenance action currently requested by the sensor (None, Low, High, Critical), alongside any device-specific detailed diagnostics. The most important indicator of measurement quality is the on-line uncertainty of each measurement, the VU. It is calculated based upon all error sources affecting the on-line measurement, including: • • • • •
The transduction itself - the mapping from the true process measurand to the observed transducer signal; The components used to manufacture the instrument; The characterization procedure at the end of the production line, and/or in-situ calibration procedures; The instantaneous operating point, and process noise; The effect of any faults, whether instrument- or process-induced, after compensation has been applied.
Thus the VU provides useful information about measurement quality whether or not a fault has occurred. By contrast, diagnostics are only provided in the (hopefully) rare occurrence of a fault, and describe only the nature of the fault and not its impact on measurement quality. The adjacent figure illustrates how the SEVA interface behaves when a fault does occur. Data has been generated from a prototype dissolved oxygen sensor based upon an electrolytic cell [2]. After an hour a fault begins in which fouling occurs on the membrane separating the electrolytic cell from the process. This is a common fault in, for example, waste water treatment plants. The fouling results in a reduction in the permeability of the membrane to oxygen molecules, and the raw dissolved oxygen measurement slowly drops over many hours. However, this self-validating sensor has two electrodes. One is used continuously for oxygen measurement, while the other is pulsed occasionally (typically every half hour) to detect fouling. Modelling of the time response of the test electrode can be used to detect fouling, and to calculate the drop in membrane permeability, and hence to compensate the measurement. An uncertainty analysis is also provided for the normal measurement and the correction for fouling. Thus, for the SEVA device, when the fault occurs the VMV is corrected, and the VU (as shown by the shaded interval surrounding the VMV) is increased to indicate the reduced quality of the corrected measurement. At the same time the MV status is switched to BLURRED, indicating that the measurement is being corrected for a fault. An independent measurement of the true dissolved oxygen level is also DOx SEVA sensor
2
shown, to demonstrate that the corrected measurement provides a reasonable estimate of the true process value. Consider now the added value provided by the SEVA metrics to the user of the measurement, for example a water treatment plant operator or an automated control system, with responsibility for maintaining a certain level of oxygen, say 7.5ppm (e.g. to avoid environmental damage). In the absence of any fault detection, all that is seen is the raw measurement, which drops rapidly. With no other indication of the true oxygen level, then countermeasures may be taken, such as switching on air pumps, which are in reality unnecessary. Alternatively, if a sensor fault is suspected, then emergency (and hence expensive) maintenance action will be required to check and repair the sensor. Possibly both actions will be necessary until the fault state of the sensor is resolved. Suppose now the sensor has diagnostics, and thus is able to indicate that a fault has occurred, for example by a device-specific error code. In this case it is known that the sensor is the cause of the apparent drop in oxygen. However, the impact of this fault on the measurement cannot be assessed by the user, and so until the sensor is repaired there is no way of knowing the true dissolved oxygen level. Hence again, to continue process operation, immediate maintenance action will be required. Consider now the response of the SEVA sensor. The operator is informed that a fault has occurred (the MV Status changes to BLURRED), but a corrected measurement is provided. Its increased uncertainty indicates the reduced confidence in the corrected measurement. Various responses are possible, depending upon whether the corrected measurement with its increased uncertainty is considered acceptable, as determined by local operating criteria. If not, then maintenance action must be carried out immediately. If however, the corrected measurement is deemed acceptable, then plant operation can continue and maintenance can occur when convenient, at a lower cost. Further steps might include action to increase the reported dissolved oxygen level to say 8.5ppm or 9ppm to ensure that the true measurand does indeed stay above 7.5ppm, despite the increase in VU. To summarize, SEVA maximizes the availability of the measurement by providing on-line correction for faults. It further provides an estimate of measurement quality in a standard, generic form, thus enabling operational and maintenance decisions to be taken based on application-specific criteria, without detailed knowledge of the sensor fault modes.
3
STANDARDIZATION AND COMMERCIALIZATION
Recent years have seen much activity in digital communications for industrial automation, usually termed Fieldbus. A range of fieldbus specifications are now available as IEC and Cenelec standards [3], for example: •
IEC 61158-n covers 8 different and incompatible specifications. These include the original IEC specification and its subsets as used by Foundation Fieldbus, plus ControlNet, Profibus, P-NET, SwiftNet, Interbus and WorldFIP.
•
Cenelec EN 50170-n includes P-NET, Profibus, WorldFIP, Foundation Fieldbus and is expected to add ControlNet early in 2000.
•
Cenelec EN 50254-n includes Interbus and simple subsets of Profibus and WorldFIP.
•
Cenelec EN 50325-n includes DeviceNet and is expected to include SDS early in 2000.
These various fieldbusses have different properties, strengths and weaknesses, making each more suited to some applications than others. Various guides are available to suggest which fieldbus might be used in which application, for example [4]. However, measurement quality is not adequately addressed in most of the fieldbus standards. In 1999 a survey was carried out by SIRA [5], funded by the DTI. It explored current and future requirements for fieldbusses, intelligent measurements and diagnostics among users and
3
suppliers of flow measurement in the UK automation industry. It discovered a consensus on the need for a common standard for describing sensor diagnostics and measurement quality, and that the SEVA concepts were a suitable starting point. The survey recommended that a standard be developed by the British Standards Institute (BSI). This request was taken up by the BSI's AMT/7 (Fieldbus) committee which drafted a standard on metrics for describing measurement quality, and the standard is now available as BS-7986. The draft standard was tabled at a recent IEC Fieldbus committee meeting, and this may lead to a new working group being set up to develop a global standard for on-line measurement quality. Note that the SEVA concepts themselves are independent of any particular fieldbus protocol, and can thus be used with any fieldbus. However, it will be necessary to agree for each fieldbus type the particular convention used to transmit SEVA data. Although aspects of the work have been ripe for commercial exploitation for some time, there have yet to be any major product releases based on SEVA. Two key reasons can be identified: • •
The slow uptake of fieldbus, particularly in the USA; The desire of users for SEVA to be an open standard, and not tied to one specific vendor.
As discussed above, there have been recent improvements in both areas, and commercialization is now being undertaken. Currently Oxford is working only with Invensys companies, but it is understood that several instrumentation and system vendors also wish to develop products conforming to the SEVA standard. Naturally it is not possible at this stage to give details of the commercialization plans. However, it can be disclosed that several instruments, including pressure, flow and positioner products, will be released with SEVA interfaces from several Invensys companies. In addition, a range of control system hardware and software products will be developed to provide support for SEVA instrumentation. This is of course essential if full use of SEVA functionality is to be possible, for example in control and maintenance activities. 4
EXAMPLE 1: IMPULSE LINE BLOCKAGE DETECTION IN D.P. SENSORS
A common form of gas flow measurement is the combination of an orifice plate and differential pressure sensor. As studies have shown, not least by NEL, the quality of the flow measurement is highly dependent upon the condition of the orifice plate itself, which is prone to degradation over time, particularly in harsh process conditions. Thus although it is certainly possible to provide an uncertainty analysis of the flow measurement performance for nominal or even typical conditions, it is inherently difficult to validate the mechanical and geometrical properties of an orifice plate over time, particularly using automated, off-the-shelf means that would be sufficiently cost-effective to become and integral part of a self-validating instrument. It is therefore suggested that if the SEVA concept becomes widely adopted within industrial practice, this is likely to accelerate the decline of orifice plate technology as being inherently difficult to validate, in favour of other technologies, such as vortex flowmeters, which are more amenable to integrated cost-effective validation. It is perfectly possible however to validate the operation of the differential pressure instrument at the centre of the flow measurement, and to detect important fault modes that may occur within the device, or in the interface between the instrument and the process. For example, the research group at Oxford has developed and demonstrated a technique to detect impulse line blockage, which is a common failure mode in a range of operating environments. Trials were carried out at a coal-fired power station, which uses large mills to grind coal to dust before entering the boilers. The measurement of air flow through the mills is achieved using differential pressure sensors. These are prone to impulse line blockage by coal dust. The practice at the plant was to use air jets to clear the impulse lines if the operator saw unusual measurement behaviour and suspected a blockage; however, no systematic analysis of the measurement errors induced by such faults had been carried out. Furthermore, the unblocking procedure could lead to delays in taking appropriate action if the cause of measurement aberration was not in fact blocked impulse lines.
4
The Oxford team made use of an augmented D.P. sensor in which absolute pressure sensors monitor the low and high pressure sides, as shown in the adjacent diagram. Signal processing based upon the absolute pressure sensor signals are able to detect the presence of blockages.
D.P. sensor with additional instrumentation
SEVA sensors in parallel with conventional plant instrumentation across coal mill
At the power plant, a number of these devices were installed across a coal mill. As shown in the adjacent diagram, each was place in parallel with an existing conventional D.P. cell used for plant monitoring and control. The ‘control’ sensor was used for plant control, while the ‘indication’ sensor was used as a secondary check on the control sensor value. Blockages tended to occur in the long impulse line runs from the mill to the sensors, and so it was reasonable to expect that the prototype devices would be liable to the same blockages as the corresponding plant instruments, which shared almost the entire impulse line. For the duration of the trials, plant personnel were detailed to blow down the control D.P. impulse lines regularly to ensure that they were fault-free. By contrast the indication D.P. impulse lines were left untouched in the hope that blockages would occur (which would of course have no impact on plant operation). In this way it would be possible to gauge not only the presence of faults, but also the measurement impact of such faults by comparing the faulty (indication) and nonfaulty (control) measurements. The adjacent diagram shows the percentage variation between the control and indication measurements in the absence of impulse line blockages. As can be seen, the differences were largely kept within about 1%. The fault detection technique works by comparing the frequency response of the absolute and differential signals.
Difference in measurement in the absence of impulse line blockage
5
After several weeks of trials, experimental data emerged suggesting that impulse line blockages were taking place, and these coincided with large differences between the indication and control measurements of up to 40%, as shown in the following diagram:
Measurement error and impulse line blockage alarm indicator
5
EXAMPLE 2: SEVA DIGITAL CORIOLIS MASS FLOW METER
Coriolis mass flow metering was originally developed for fluid flow applications, but as the precision and sensitivity of the instrument has improved, along with flow-zero stability, it is finding increasing application for gas flow, as reported at this conference and previous ones in this series. Much of the Oxford work on coriolis has been focussed upon pure liquid or twophase (gas-liquid) flow applications, although both the SEVA concept and its coriolis implementation are applicable to gas flow applications. A first generation SEVA prototype based on the coriolis mass flow meter was developed at Oxford in the early 90s. This consisted of the vibrating coriolis flowtube (in line with the process piping), and the conventional commercial electronic transmitter attached to a PC [6]. The prototype demonstrated an ability to detect and correct for several fault modes, as well as the generation of SEVA metrics. However, the identification of the major limitations and fault modes of the instrument, and their impact on measurement quality, provided motivation for improvements in the basic design. This led on to the idea of an all-digital transmitter, which has been developed at Oxford. The primary intention was to replace all the analogue circuitry (other than essential front-end op-amps) with a small number of digital components, and to carry out virtually all functionality within software. An all-digital design is attractive not only for research purposes, but also for commercial implementation. Digital (and specifically audio) technology is available for mainstream consumer markets such as mobile phones in large volumes at low cost. The commercial appeal is clear: a product with a low component count and high software content, based on rapidly evolving, competitive, mainstream technology. Examples of the performance improvements in the new digital transmitter include the following (see [6] for more details): •
Repeatability. The digital transmitter shows good repeatability. For example, two meters (with 25mm and 50mm flowtubes) were placed in series in a flow rig, and their mass totals compared. Back-to-back trials, where both meters receive the same start and stop total signals, are effective for assessing repeatability, as some error sources (e.g. the accuracy and repeatability of the flow rig) are eliminated. The difference between meter totals over 20 consecutive batches of 300kg at 1kg/s exhibited a standard deviation of 0.005%.
6
•
Flowtube oscillation control. A new non-linear oscillation control algorithm has been developed [7] which provides excellent stability of flowtube operation, and good response to changes in the setpoint for the amplitude of oscillation. Using this algorithm, it is possible to maintain flowtube oscillation down to 0.3% of the conventional amplitude, drawing only 0.1mA per drive coil, demonstrating the potential for low power operation.
Two aspects of improved performance are now described in more detail, these being batching from empty and dealing with two-phase flow. 4.1
Batching from empty
There are many processes, particularly in the food and petrochemical industries, where the high accuracy and direct mass-flow measurement provided by Coriolis technology would be beneficial in the metering of batches of material. Often however, it is not practical (or highly inefficient) to ensure that the flowmeter remains full of fluid from start to end of the batch. For example, in filling or emptying a tanker, air entrainment is difficult to avoid. In food processing, hygiene regulations may require the cleaning out of pipework between batches. However, large errors are induced in conventional Coriolis meters when ‘wet and empty’. Hydraulic shock and a high drive gain requirement may be caused by the onset of flow in an empty flowtube, leading to large measurement errors and stalling. Most Coriolis manufacturers are thus unable to recommend the use of their products in batch applications unless the flowtube can be kept full.
BATCHING FROM EMPTY EXPERIMENT M a g n et ic F lo w m e te r
C or io lis F lo w tu b e
D ra in Va lv e
W ei gh in g Ta n k
BATCHING FROM EMPTY RESULTS
At the start of the batch the totalisers in the magnetic and Coriolis flowmeters are reset and flow commences. At the end of the batch the shutoff valve is closed and the totals are frozen (hence in this case the Coriolis meter is full at the end of the batch). Three totals are recorded, from the magnetic and Coriolis flowmeters, and from the weighscale. These totals are not expected to agree, as there is a finite time delay before the Coriolis meter and then finally the weighing tank observe fluid flow. Thus, it would be expected that the magnetic flowmeter would record the highest total flow, then the Coriolis meter, with the weighing tank observing the lowest total.
100
50
Kg Offset From Weighscale
The digital transmitter has been designed to be more robust to the conditions experienced when batching from empty. Experimental data is presented which demonstrates the resulting improvements in accuracy and repeatability. The adjacent figure shows the experimental rig. Fluid is pumped through magnetic and Coriolis flowmeters into a weighing tank. Valves are used to ensure that the magnetic flowmeter is always full, while the Coriolis flowtube begins each batch empty.
0
-50
-100
-150
-200
-250 Magnetic Flowmeter
The graph shows the results obtained from a series of trial runs, each of about 550kg. The data shown is the offset observed between the weighscale and
Analogue Transmitter
Digital Transmitter
7
either the magnetic flowmeter or the Coriolis meter. As explained above, positive offsets are expected from both instruments. The (always full) magnetic flowmeter delivers a consistently positive offset, with a repeatability (defined here as the maximum difference in reported value for identical experiments) of 4.0kg. The analogue transmitter generates large negative offsets, with a mean of −164.2kg and a repeatability of 87.7kg. This poor performance is attributable to its inability to deal with the onset of flow and the variable time taken to restart the flowtube. By contrast the digital transmitter shows a positive offset averaging 25.6kg and a repeatability of 0.6kg. As discussed in [6] it is difficult to determine the exact mass flow through the Coriolis flowtube, given that the meter starts empty, but it is clear that these results represent a substantial improvement on the analogue transmitter performance, and suggests that batching to/from empty is viable using coriolis mass flow metering. 4.2
Two-phase flow
Another important condition which presents difficulties for analogue transmitters is sporadic or continuous two-phase (gas/liquid) flow. The underlying mechanisms are very similar to the case of batching from empty: the dynamics of gas-liquid flow causes high damping. To maintain oscillation a high drive gain is required, but typically at low levels of gas fraction the maximum drive gain of the transmitter is reached, and the flowtube stalls. The digital transmitter has been designed to maintain oscillation in the presence of two-phase flow. In experimental trials, it has not been possible to stall a flowtube of any size with any level of gas phase when controlled by the digital transmitter. By contrast, the analogue transmitter stalls with about 2% gas phase. More modern commercial designs are more robust to twophase flow, being able to withstand up to 40% gas phase. However, even where vibration is maintained, large mass flow errors are generated. A recently completed research project at Oxford has explored the possibility of detecting and correcting for two-phase flow [8]. A neuralnet model has been developed to predict the mass flow error from several internal parameters. The figure below shows the uncorrected mass flow error against flow rate and gas phase (as indicated by the drop in observed mixture density) for a particular flowtube design. Here, the 25mm flowtube was in horizontal alignment, and was fed with water/air mixtures essentially restricted to continuous bubbly flow. The results below were obtained over 134 experiments. RAW MASS FLOW ERRORS FROM TWO-PHASE FLOW
8
CORRECTED MASS FLOW ERRORS FROM TWO-PHASE FLOW
The graph above shows results obtained in further experiments over the same parameter range, during which the correction algorithm is applied. In all cases the mass flow errors are kept within 2%. REAL-TIME RESPONSE TO TWO-PHASE FLOW
The adjacent graph illustrates how the SEVA device responds to the onset of two-phase flow in real time. With single phase flow (up to t=8s), the mass-flow measurement is CLEAR and it has a small uncertainty of about 0.2% of reading. Once twophase flow begins, the correction algorithm is applied, the measurement is set to BLURRED, and the uncertainty increases substantially, reflecting the reduced accuracy of the corrected measurement. The lower line shows the uncorrected mass flow rate while the dashed line shows the (single phase -liquid) flow rate as measured independently by another coriolis meter prior to the air injection point.
In a real application, the user would have the option of continuing operation with the reduced quality of the corrected mass-flow rate, switching to an alternative measurement, or shutting down the process. Where aeration occurs sporadically, the batch total uncertainty will be a function of both the duration and severity of the two-phase flow periods. It is hoped that as the
9
understanding of the effects of two-phase flow improve, better correction techniques can be developed with correspondingly smaller uncertainty bounds. As discussed in [9], the digital transmitter is the first example of a ‘second-generation’ SEVA instrument, in which the knowledge gained through validating the current commercial transmitter has led to a much more robust design. This demonstrates that SEVA can be viewed as a philosophy promoting continuous measurement quality assessment and improvement, having impact on all aspects of the sensor lifecycle (e.g. design, manufacturing, characterisation, installation, calibration), and not merely on-line operation.
6
REFERENCES
[1] Henry, M. P. and D.W. Clarke (1993), The Self-Validating Sensor: Rationale, Definitions and Examples. Control Engineering Practice, 1 (4), pp.585-610. [2] Clarke, D.W. and Fraher, P.M. (1997). A self-validating Dissolved Oxygen Sensor. Control Engineering Practice, 4(9), 1313-1320. [3] Wood, G.G. (2000). UK activities in measurement validation and data quality. IEE Computing and Control Journal, Special Issue on Intelligent and Self-Validating Instrumentation, October 2000. [4] BSI (2000). Guide to the Evaluation of Fieldbus Protocols, British Standards Institute, 389 Chiswick High Rd London W4 4AL . [5] SIRA (1998). The impact of Fieldbus and sensor fusion on flow measurement. DTI NMS Programme for Flow Measurement, Project OR9A Technical Awareness Studies. [6] Henry, M. P, Clarke, D. W., Archer, N., Bowles, J., Leahy, M. J., Liu, R. P., Vignos, J., Zhou, F.B. (2000). A self-validating digital coriolis mass-flow meter: an overview. Control Engineering Practice, 8(5), pp. 487-506. [7] Clarke, D.W. (1998). Non-linear control of the oscillation amplitude of a Coriolis mass-flow meter. European Journal of Control. 4 (3), pp196-207. [8] Liu, R.P., M.J. Fuent, M.P. Henry, M. D. Duta. (2001). A neural network to correct mass flow errors caused by two-phase flow in a digital coriolis mass flowmeter. Flow Measurement and Instrumentation, 12 (1), pp 53-65. [9] Henry, M.P. (2001). On-line compensation in a digital coriolis mass flow meter. Flow Measurement and Instrumentation, 12 (2), pp147-161. Acknowledgements The author would like to acknowledge the hard work, support and enthusiasm of colleagues at Oxford and Invensys, along with the financial support of Invensys and the EPSRC.
10
