|
Intelligent
Condition Monitoring
A
technique called Signature Point Referencing™ can
substantially reduce cost of managing non-destructive
testing with minimal up-front cost.
The
objective of software tools for management of pipe
deterioration is to predict the point at which unacceptable
risk of failure will be reached. It is vital that this be
done early enough to avoid the subsequent loss by taking
appropriate evasive action.
The
monitoring of pipelines and vessels usually requires
exhaustive measurement of a massive number of points. These
usually involve substantial manual effort and at times
reduction in throughput of production fluid, pressures
and/or temperatures. This presents a challenge in deciding
which lines to measure, what points to monitor on those
lines and when.
Computerised
Maintenance Management Systems (CMMS) should not be burdened
with the massive volume of data required in NDT, but should
be provided with meaningful data
indicating the condition of the plant and possibly some
action that should be taken.
A small number of specialised software products have evolved
to manage specific industry, high volume, and recurring
inspection data as occurs in pipeline monitoring. These
products have mostly evolved out of the oil refining
industry despite needs in other areas and suffer some
shortcomings for general use.
Signature
Point Referencing™ provides an intelligent mechanism to
dramatically reduce the effort involved in monitoring those
points. It uses a patented algorithm that intelligently
selects points that are due for measurement, or more
importantly, points that have been calculated as being near
an alarm condition from their relationship to other points
that have been measured. The benefits of this system have
been shown to produce remarkable savings when compared to
conventional methods.
This
new approach to reducing the cost of monitoring is based on
sound engineering principles and has been trialled in
Australiafor
over 18 months.
Clearly,
the reason maintenance people monitor pipeline deterioration
is to determine when an unacceptable level of risk of
failure will be reached.
This
simple requirement is often the source for significant cost
and produces a management challenge. The problem is not so
serious when there is a small number of plant items to
monitor, or a small number of points on them to track. It
becomes an extremely difficult and costly problem when, for
instance, you have hundreds of kilometres of pipelines with
many points to measure on each spool.
The problem is compounded when risk associated with failure
is severe due to high pressures, temperatures or dangerous
process fluids.
Reliability
engineers assess risks associated with failure of a
particular item and decide to track its condition when that
risk surpasses an acceptable limit. Intervals between
measurements are struck, alarm values derived and the
measurement cycle begins.
Often
spreadsheets are used in an attempt to unscramble the vast
amounts of data collected. With thousands of monitored
points and when some history has been collected, these can
easily become unwieldy. It soon becomes more cost effective
to use software designed to manage this kind of data.
Such
software must be able to handle storage of a very large
number of measurements on a very large number of points in a
single database. Serious database management systems (dbms)
such as Oracle and Microsoft SQL Server are required to
provide robust security, history and comparison
capabilities.
Use of smaller dbms such as Microsoft Access and Paradox are
only useful for small datasets, are not as reliable as their
big brothers and are slow in processing the data.
Data
that describes the condition of plant and equipment can be
broken down into two categories: recurring and
non-recurring. Thickness monitoring is recurring as it is
collected in a cyclic manner in the field. It provides an
alert to maintenance decision-makers as to the need for
action.
Recurrent
condition monitoring requires continuous selection of
points, or routes of points, for measurement. New
measurements are entered into some form of long-term storage
and exceptions are then reported.
Extraction
of points from paper and spreadsheet storage methods
requires considerable manual effort and discipline. At the
very least, a list of locations to visit and the points to
measure is required. Pipelines usually require hundreds or
even thousands of points to be measured at one time. To do
this manually is often a costly exercise. One multinational
company identified the cost to be as high as AUD$16.00
per point per annum.
The situation is more complicated by the fact that the
measurements must also be entered back into the long-term
storage system and calculations made to predict alarm dates.
The
challenge is then to decide which points to measure next.
This
is where Signature Point Referencing™ has made a
difference. Most systems provide a simple selection of a
route of points that has become due according to some simple
scheduling mechanism(s).
Signature
Point Referencing™ provides an intelligent algorithm that
allows selection of points from a range of criteria. The
internal expert analysis and selection criteria allow great
reduction of the number of points to be physically measured
by a ratio of hundreds or even thousands to one. Values for
points not actually measured on any one venture out into the
plant are derived from mathematical relationships to those
that are measured, so condition values are still available
for those points too.
The
greatest problem of all for management of the vast amount of
data is the speed of calculating results. When millions of
readings are accumulated over time the reporting engine must
be optimised so that reports run quickly.
A powerful dbms such as OracleTM in conjunction with an
optimised database design will render results in less than a
few minutes for even the largest enquiry.
Condition
monitoring systems have mechanisms for showing at what time
a point being monitored will reach an alarm condition. The
alarm value is chosen to enable sufficient early warning to
allow for planning and scheduling of maintenance. The
algorithms employed are generally a straight line fit to the
data with strange terms such as short and long prediction to
indicate a best-case to worst-case range of failure dates.
These
calculations are often quite simple and somewhat inaccurate
with the difference between the short and long predictions
being many years apart. This of course is of no use
whatsoever to a maintenance planner who needs to schedule
maintain a particular plant item. Choosing the earliest date
is a waste of capital invested in the plant, and the long
date might be too late.
It
is usually futile to attempt to use linear regression across
the long-term history of readings because sections of
pipelines are often changed out without notifying the
planner. When an item of plant is repaired or changed out
the straight-line mechanism fails because of the saw-tooth
effect in the graph (see Figure 2). In the pipe thickness
example, what was once thin is now thick, and to draw a
straight line through this produces a nonsense result.
The
following diagram indicates the results achieved over a long
period when one incorrectly assumes that a component hasn't
been changed.
Figure
1. Saw Tooth Effect
Unless
you are aware of the plant repair or change-out, serious
miscalculations will result. This problem is very easily
understood and even fixed if you are monitoring just a
handful of points but becomes severely time consuming when
there are tens or hundreds of thousands of monitored points.
The systems generally provide some mechanism to avoid this
problem by using an historical loss rate if the plant has
been changed.
Signature
Point Referencing™ has overcome the “plant-altered”
problem by presenting to the user
all
points in a batch of measurements that fall outside user
nominated parameters.
Ranges of anomalous conditions are checked and the user is
asked to choose how to handle them. By this method a single
point on a pipe spool can be measured over a period spanning
alterations to the plant with minimal loss of accuracy. Any
time a plant is marked as being “Altered” either
manually, through the data import processing or from the
CMMS interface the historical
loss of material is used to determine the point at which
alarm conditions will be met.
This allows you to predict through the step in the graph.
3.3.
Computerised
Maintenance Management Systems
Computerised
Maintenance Management Systems (CMMS) often have modules
available for storage of condition monitoring data, however,
they are generally not well suited to the specialised
requirements for analysis of condition monitoring data. CMMS
are designed to manage information…information is more
than just large chunks of raw data as is collected in the
field.
Support systems for the CMMS should aggregate and filter out
data that is of little benefit in the CMMS and only provide
the information that is needed to support decision-making.
Typical
problems experienced when storing condition monitoring data
in CMMS are;
Adding
otherwise unnecessary complexity to location and equipment
registers making the system more difficult for regular CMMS
users. CMMS work best when structured to manage work on ”maintainable
items” such as pumps, motors, screens and valves.
Condition monitoring requires at least an order of magnitude
more detail than this.
The
large volume of data involved and the frequency of it being
updated. CMMS usually provide a screen for input of data,
manual entry of thousands or even hundreds of measurements
is tedious, error-prone and labourious.
The volume of condition monitoring data is likely to be as
large as the rest of the entire CMMS database combined.
Lack
of suitable prediction algorithms.
Special analysis will need to be done via specially written
programs or implemented using the reporting tool.
The latter is especially troublesome because of the
complexities in dealing with effects such as the saw-tooth.
Inappropriate
display and reporting mechanisms. A CMMS might provide a
screen for displaying generic condition monitoring data
however this is unlikely to suit the requirements of all
kinds data.
This makes interpretation of the results of monitoring
difficult if not impossible.
Report writers are typically designed to list off multiple
rows of information and provide many levels of aggregation.
It is also usually possible to write more complex queries
directly in SQL. Even so, it is very difficult and therefore
costly to develop reports in this fashion.
Lack
of interfaces to dataloggers.
Dataloggers used to store field measurements are able to
upload data to a personal computer. CMMS generally do not
provide seamless interfaces to the various datalogger brands
and models available in industry.
The
application of a condition monitoring system in a
traditional maintenance management
environment
is illustrated in Fig.1. Maintenance management decisions
are supported by theCMMS,
which may be linked to or integrated with a number of other
support systems.
CMMS
are highly vertically integrated and typically allow
management to:
- Organise
and track inventory
- Manage
equipment costs
- Track
equipment history
- Schedule
maintenance tasks
- Manage
and maintain maintenance resource utilisation records
- Generate
work orders, purchase orders, requisitions etc.
- The
difficulty in interfacing CMMS to special purpose
condition monitoring systems is always in the level of
detail. The latter often provide equipment structures
that are an order of magnitude more detailed than in the
CMMS.
Figure
2. Typical CMMS environment
Because
special purpose condition monitoring systems are continually
under development to meet the needs of specific industries,
they also incorporate many of the features of the
traditional CMMS. There is therefore a significant overlap
in capability provided by these systems with CMMS.
Unfortunately, the overlap causes difficulty in deciding
where the interfaces should occur and complicates the
decision for an organisation to invest in such a system.
For
these reasons the pipe management system is often left
non-integrated with the CMMS and the benefits are thus lost.
3.4.
Signature
Point Referencing™
Signature
Point Referencing™ is unique in that it uses predictive
intelligence
technology to track any kind of measurement data.
It
works with most existing data measurement and collection
devices and learns the relationships between monitored
components as successive measurements are taken.
 Proprietary
algorithms embedded within the system offer quick and
accurate predictions based on the aggregation of fewer and
fewer real world measurements.
Points
nominated as Signature points are monitored at a higher
frequency than the points they control. This provides
frequent and automatic compensation for fluctuations in
operating conditions and derives thickness values on
components that would otherwise be uneconomical to measure.
The
technique produces a saving in time and resources compared
to traditional manual methods that employ exhaustive
testing. Signature Point Referencing™ not only recognises
trends over a period of time, it continues to add data and
question its own results. This presents a case for
maintenance action that is based on scientific outcome, not
personal objectivity.
The
algorithm includes a user definable weighted average for the
rate of deterioration to smooth oscillations that typically
occur in field measurements.
The
selection of points for download to a datalogger for
measurement is handled dynamically. The user can select from
a matrix of choices that allows him to download historical
data into a datalogger, with each measured point standing on
its own. The user can choose “Signature Points”, “Calculated
Points” or “All Points” within selected plant areas
and /or routes, and for each selects points that will
surpass one of the alarm levels at a nominated date, or “All
regardless of alarm condition”. The points can then be
arranged into any desired order to create a kind of instant
measurement route.
3.5.
Unrelated
Units of Measure
The
algorithm allows you to measure one variable to derive the
value of another.
An example of this would be recording kilometres travelled
by a haul truck to show the expected wear on the tyres.
Signature
Point Referencing™
provides a mechanism to reduce the number of real-world
measurements required without increasing the overall risk of
failure. The reduction in the number of physical
measurements required allows more of the plant to be
monitored at a lower cost.
It does this by using patented technology that has been
shown to produce a massive return on investment.
Copyright
2002 (c) SigPoint Pty Ltd
|
