Unfortunately, a lot of our revenue comes from investigating and fixing poor designs. In this case the impact of a late design change was masked by the manner in which the process simulations were constructed. The process was in fact inoperable with no clean solution.
Like any offshore platform the aim was to produce a stabilised oil. Disposal of the condensate was initially recognised as a potential problem. It was part of the original design to re-inject the condensate into the reservoir. In order to cut costs this idea was revisited during the infamous ‘value engineering’ phase and removed.
On start-up the gas compression KO drums flooded due to the accumulation of condensate recycling around the compressors. John was asked to review the design and suggest possible ways forward.
Sample analysis confirmed the well stream compositions corresponded to those used in the design. However, the steady state process simulation model provided was very slow and difficult to converge. This is generally a warning sign.
The model convergence appeared to be due to excessive use of ‘recycle convergence blocks’. Each recycle was independently configured a dedicated convergence block.
Checking the overall component mass balance also showed the cumulative effect of each recycle blocks convergence criteria produced small but significant discrepancies in the mass balance of some key components. Rationalising the convergence blocks significantly reduced their numbers and scope for errors.
The revised model was still difficult to converge, requiring manual intervention, it did however now show a substantial increase in the condensate recycles sufficient to overwhelm the compressor KO drum condensate level control valves and drain lines. After start up it wasn’t practical to increase the valve and drain line sizes to the required values.
Reconfiguring the scrubber drains and purging some light condensate to flare, whilst relaxing the crude TVP and export gas specifications provided a less than ideal solution. In other words an offshore process had been built that simply didn’t work.
The problem arose due to not recognising a potential issue when a model is difficult to converge. Each use of a convergence block introduces convergence criteria that either compounds the error in mass balance or stops the model from converging. In this case, the errors were sufficient to mask the problem of removing the condensate injection.
Models that don’t converge easily can usually be put down to one or more of four causes:
- Convergence blocks are incorrectly or inefficiently placed.
- The model contains ‘Adjusts’ that iterate some operating conditions to provide a solution to some desired criteria. ‘Adjusts’ are unreliable, especially when nested.
- The process has significant issues that must be resolved. These may be manifest by slow convergence and relative large changes in the flow rate of some streams when relatively small changes to temperature and pressure specifications are made.
- The simulation package contains an error.
We expect the process simulator to provide a perfect mass balance (Input = Output), they don’t. There are usually some small insignificant imbalances. In models that have several recycle convergence blocks and/or high recycle flow rates the imbalances may become significant.
If a steady state model is indicating convergence problems, investigate the cause.