A pilot plant exists to answer the questions the laboratory cannot. Its product is data. A pilot campaign that makes good material and leaves those questions open has not done its job.
Post 1 showed why size changes behaviour, using a reaction scaled from 1 L to 10,000 L. This post covers the middle step, the 100 L pilot reactor: what it should be designed to find out, how large it should be, and what to have in hand before it is shut down. Post 3 uses the results.
Start from the questions
The pilot plant sits between two lists. The first is the list of questions the lab work left open. The second is the list of numbers the commercial design needs. Each pilot run should move an item from the first list to the second.
Table 1. Eight questions a pilot plant commonly has to answer
| Question | What to measure | Design decision it feeds |
|---|---|---|
| How fast is heat released, and can any reactant build up unreacted? | Heat flow and conversion against time during the addition | Addition time, cooling system, relief design |
| Is the result sensitive to mixing? | Yield and impurities at two or three agitator speeds | Agitator type and power, feed point location |
| How well does the product filter, wash and dry? | Filtration rate, cake moisture, drying curve | Filter and dryer type and size |
| What do recycled streams do over many batches? | Impurity levels in recycled solvent over consecutive runs | Purge rate, solvent make-up, purification steps |
| Which materials resist the process fluids? | Corrosion coupons and gasket samples exposed in the reactor | Materials of construction |
| Does commercial-grade raw material give the same product? | Runs using the grade and supplier planned for the plant | Raw material specifications in the design basis |
| What leaves as waste and emissions? | Quantity and composition of every waste stream and vent | Treatment systems and permit applications |
| Can operators run and clean it? | Blockages, fouling, cleaning time, manual interventions | Operating procedures, staffing, cycle time |
Not every project has all eight. A project should write down its own list before the pilot plant is designed, because the list decides which instruments, sample points and equipment the pilot plant needs.
How large should the pilot be?
The pilot should be large enough to show the effects of scale and small enough to change cheaply. The step that is safe to take depends on what is in the process.
Table 2. Typical size of one scale-up step
| Type of operation | Typical step in volume | Why |
|---|---|---|
| Single-phase liquid or gas reactions with known kinetics | 100 to 1,000 times | Behaviour can be calculated with confidence |
| Liquid reactions limited by mixing or heat removal | 50 to 100 times | Results depend on vessel geometry |
| Gas-liquid and liquid-liquid reactions | 20 to 100 times | Contact area between the phases is hard to predict |
| Solids: crystallization, filtration, drying | 10 to 50 times | Particle size and cake behaviour change with scale |
These ranges are a guide and vary with how well the process is understood. In the example, each step is 100 times: 1 L to 100 L, then 100 L to 10,000 L. If the product had to be crystallized and filtered, that part of the process would justify a smaller step or a separate test on vendor equipment.
Run until the results repeat
One good run shows that the process can work. A design needs to know how it works on an ordinary day. That takes consecutive runs under the same conditions, with every input and output weighed.
In the example, ten consecutive runs gave yields between 85% and 89%, with a mean of 87.2%. The commercial design in Post 3 uses 87%. Using the best run, 89%, would understate raw material use and waste for every year of operation.
What to have before you shut the pilot down
Table 3. Data to collect before the pilot campaign ends
| Data | Test of completeness |
|---|---|
| Mass balance | Everything in and out weighed. Totals agree within about 2%. |
| Energy balance | Heat of reaction and heat transfer coefficient measured on the pilot reactor |
| Yield and quality | At least five consecutive runs at the design conditions, using commercial-grade raw materials |
| Operating limits | Runs at the edges of temperature, addition rate and concentration, showing where quality is lost |
| Recycle effects | Recycled solvent used for enough runs to show whether impurities level off |
| Separation data | Filtration, washing and drying rates measured on representative material |
| Materials | Corrosion rates from coupons exposed for the whole campaign |
| Waste and emissions | Each stream measured and analysed |
| Product samples | Enough material for customer approval |
The first two rows matter most. A pilot plant without flow meters, load cells and temperature measurements on the jacket cannot produce a mass or energy balance, however well it runs.
Five common mistakes
- Building the pilot to make samples. It is designed around the product and has too few instruments to give design data.
- Using lab-grade raw materials. Trace impurities in commercial grades are a frequent cause of surprises at start-up.
- Stopping after the first good run. The design is then based on the best result.
- Running once through with fresh solvent. The plant will recycle, and impurities that build up never appear in the pilot.
- Keeping no record of problems. Blockages, foaming and cleaning difficulties are design information and should be logged like any other result.
Takeaway. List the questions before designing the pilot plant, instrument it to close a mass and energy balance, and run it until the results repeat. In the example, ten runs give a design yield of 87%.
The reaction, the vessels and the costs in this series are an illustrative example. The calculations are screening estimates and do not replace process safety testing or a design by a qualified engineer.
