This year's MODEL

`The model is the message,' declares Sandro Macchietto, punctuating his remark with a palms-down gesture towards the tabletop. The dapper Italian professor of process systems engineering at Imperial College is clearly keen to explain his subject. `What we do here is to improve processes; and we do that by making models. Once we've done that, only then can we evaluate and optimise.'

Macchietto heads what he claims to be the largest concentration of research into the process industries on the planet. With 100 full-time researchers, the Centre for Process Systems Engineering (CPSE) is about three times larger than its nearest rivals, he says. Around half of its £4 million annual turnover comes from the Engineering and Physical Sciences Research Council (EPSRC), the rest coming from the European Union and industry.

Perhaps best-known as the birthplace of the best-selling process simulation package, SPEEDUP, which it sold to Aspen Technology in 1991, the CPSE is an interdisciplinary research centre. Its research staff include chemical engineers, electrical and control engineers, computer scientists specialising in both software and hardware, `and even a couple of MBAs,' Macchietto notes. `It's no good coming up with all these wonderful ideas if they don't make commercial sense.' A consortium of companies support the centre, including ICI, DuPont, and Honeywell.

Macchietto is therefore ideally placed to predict the future for process plant design. `At the moment, we still see companies going for economies of scale, and for certain products that will continue. But things are changing. There's an emphasis now on customisation of products. What will become increasingly important is the design of multipurpose plants that would be flexible and able to cope with a wide range of products. We'll see economies of scale giving way to the economies of flexibility.'

But there are still problems, he says. Multipurpose plants are in a near-constant state of flux, and the current generation of process simulation packages is very good at providing models of steady-state processes, but they can't cope when the process is starting up or shutting down, or when conditions are changing to make a new product grade. `There are a huge range of variables to cope with,' explains Macchietto. `Some vary over time, like temperature, giving us very large partial differentials that have to be integrated. Others are discrete variables - for example, a valve can be open or closed, a stirrer on or off. What we need - and what we're working on now - are new languages and environments that will allow us to optimise these dynamic processes.'

The centre's new baby is known as the general process modelling system, or gPROMS. The project is led by Costas Pantelides, who helped develop SPEEDUP in the 1980s and now hopes to cap this achievement.

`Most of the things that are do-able with simple models have already been done,' Pantelides comments. Older simulation tools may be able to optimise processes up to 97-98 per cent of their maximum theoretical efficiency, he states, `but for a pharmaceutical firm, that last 2-3 per cent could represent hundreds of millions of pounds in profit.'

For example, current modelling packages can only deal with perfectly-mixed reactors. `This doesn't happen in the real world. Products are often distributed - by size and shape in the case of crystallisation, which is very common in pharmaceuticals; by molecular weight in polymerisation reactors; by spatial position in tubular reactors or packed bed absorption systems.'

The system works by modelling both the processes inside the plant and its operating procedures in terms of what Pantelides calls `hierarchies'. For example, a separation unit might consist of four linked distillation columns handling different mixtures. These have top, middle and bottom sections where different processes are at work. Each of these contains a series of linked trays, with vapour passing up and liquid running down. And the action of each tray can be described in terms of the formation and behaviour of bubbles of vapour. Together, these stages give a very accurate view of something that, in the past, would have been treated as a `black box', with so much of a mixture going in and so much of its components coming out.

Similarly, process operations can be broken down. Starting up the separation system consists of a series of operations - filling up a reboiler tank, opening a reboiler steam valve, and so on. `There can be tens of thousands of steps, and each will have a distinct effect on the plant's dynamic operating profile,' comments Pantelides. g28

The gPROMS team has recently finished a project to make a virtual model of a plant making the plasticiser dioctyl phthalate (DOP) Mitsubishi Chemical. Its production process, which involves the esterification of 2-ethylhexanol (2EH) with phthalic anhydride (PA) with a homogeneous catalyst in a batch reactor, is several decades old.

Mitsubishi was suffering from a problem quite common in the speciality chemicals sector - batch processes tend to operate to tight margins. In Mitsubishi's case, the margins on the DOP process were down to 1 per cent of the price, leaving it close to making a loss.

The process has complex kinetics. Two reactions occur: a fast, irreversible reaction between PA and 2EH to form (mono)octyl phthalate; and the reversible formation of the diester from this and more 2EH, which proceeds by competing catalytic and non-catalytic pathways.

The gPROMS model had to simulate three components: the reactor, the separator and the reflux drum. There were just six factors the operators could vary: the rates of addition of PA, fresh 2EH and recovered 2EH to the reactor; the rate of steam supply to the heating coil; the pressure inside the reactor; and the amount of catalyst. Despite this, however, the model consisted of over 2000 differential and algebraic equations.

The result showed that the PA should initially be fed into the reactor simultaneously with recovered 2EH and, as the reaction proceeds and the PA feed finishes, the recovered 2EH is replaced with a fresh supply. Meanwhile, the optimum temperature profile keeps within the bounds specified by the product purity, but goes as close as possible to the upper boundary as esterification begins. This cut the operating costs by 1 per cent - enough to double the potential profits from the process.

gPROMS is currently being tested by chemical engineering departments at 87 universities. It is being used to model systems like continuous polymerisation, where the conditions are changed frequently to make different grades, and also by unit operations designers, who are using it to make detailed models of what happens inside new equipment. `I can see a time where a single package like this could be used to design an entire complex - from each piece of equipment to the final formulation and distribution network,' Pantelides predicts.