Making use of new ideas in examining data30 June 2000
An enormous amount of data is provided by plant monitoring equipment. New ways of examining the data can provide useful insights for the operator. Some possible techniques may be imported from other industries.
There is continual pressure to improve production efficiencies in any industrial process and nuclear power is no exception. To compete in a liberalised electricity market, it is vital to optimise reactor performance. Sometimes that means looking for technologies developed in other industries: a new technique for analysing plant data used in process industries may offer nuclear plant operators a new perspective on the variables that impact plant performance and their interaction with each other.
Developed by Curvaceous Software, the Curvaceous Visual Explorer uses a co-ordinate transformation to display an almost unlimited number of variables on a two dimensional graph. In doing so it gets round the problem geometry has suffered since Euclid of existing in three dimensional space. Being able to represent all the variables that are important to plant performance on one graph gives operators a much better perspective on the factors affecting efficiency and quality of production.
In the Explorer, different variables are plotted on axes that are parallel to each other, rather than the perpendicular model more normally used. Points in Euclidian space become lines connecting the parallel axes. Any particular production condition can therefore be displayed as a line connecting all the variables represented on the graph.
By building up a whole series of lines covering a variety of plant conditions over time, patterns emerge which provide important information on plant operations.
STEAM PLANT MANAGEMENT
Robin Brooks, Curvaceous Software managing director, has described the technique by applying it to a steam boiler. He analyses the boiler's performance and asks whether there is any technical or operation improvement that can be made and whether the economic incentives to do so are good enough. The boiler in question produces up to 200tons of steam an hour and operates at 600psi and 700 degrees F.
The boiler is fired with both oil and gas. Feedwater purity is measured by using the electrical conductivity of water. Combustion air is provided by two combustion air blowers feeding into a windbox which has eight damper positions on its input and four controlling its output and hence the internal pressure.
In total the data consists of 17 variables.
Brooks took 1736 spot values every hour for three months. The data was then plotted using parallel co-ordinates. Each of the variables was plotted from bottom to top, with the lowest value at the bottom of the axis. Each value is joined by a line to the value of the variable on the next axis at the same time. Thus each moment represents a line on the graph (see diagram below).
The graph indicates some clear operating patterns. For example it is clear that Air1F and Air2F operate only within certain limits. Blowdown flow is similarly limited, as is fuel oil flow.
USING THE DATA
There are two main ways to examine this type of data. One is to look at unusual events to try to understand how they happened. The other is to look at normal variability during plant operation and compare it with optimum conditions.
By limiting variables to the conditions under which optimum performance is achieved, it is possible to assess how much of the time the plant is operating at maximum efficiency, and which variables have the most impact on plant performance. Operators can therefore concentrate on maintaining plant operations in the most efficient conditions.
A useful insight that comes from representing plant data in this form is the emergence of process 'holes'. By plotting optimum plant performance data in one colour and overall plant data in another, areas where the plant is operating below its best are exposed.
Holes "are a major problem for plant operations as existing process control algorithms are not able to recognise and therefore avoid holes," says Brooks. "The best - perhaps only - way to deal with holes at present is to understand the process sufficiently well that one can decide which of the yellow bands is 'best' and make that the only operating zone through rigourous implementation of operating procedures, process controls and control limits."
Another important insight is the realisation that often optimum performance occurs when the plant is being operated with certain variables at the extreme edge of their range. This is contrary to the way many plants are run, where operators tend to keep plant variables at or near the centre of their operating range.
It is very unlikely that operating a plant where all variables are maintained within their mid-range will produce optimum performance, but the problem for an operator has been that he or she does not know the changing values of the limits of operational variability and therefore sticks to the mid-range for safety. By representing the data in this form, the limits of operability can be more easily understood.
A further dimension to the analysis is added when cost factors are incorporated. In the context of a nuclear power plant considerations would include optimising fuel burn-up and minimising waste production. Curvaceous Software has yet to analyse the performance of a nuclear power plant using the Curvaceous Visual Explorer, but Robin Brooks is interested in applying the technique to all industrial processes and would welcome data from plants which his company could analyse in this way.
With the increasing emphasis on optimising reactor performance and improving fuel management, new ideas in software management like the Curvaceous Visual Explorer may provide important operating tools.