Distillation: So simple and yet so complex... and vice versa Sigurd Skogestad Norwegian University of Science and Technology (NTNU) Trondheim, Norway

Outline When use distillation Increase in heat input decreases temperature?? Complex model but simple dynamics usually Control: Get rid of some myths! Complex column configurations (Petlyuk/Kaibel) save energy as well as capital

BASF Aktiengesellschaft

F V L B D

Alternative: Packed column

When use distillation? Liquid mixtures with difference in boiling point Unbeatable for high-purity separations because Essentially same energy usage independent of (im)purity!  Going from 1% to % (1 ppm) impurity in one product increases energy usage only by about 1% Number of stages increases only as log of impurity!  Going from 1% to % (1 ppm) impurity in one product increases required number of stages by factor 3  log(1.e-6)/log(1.e-2)=3 Well suited for scale-up  Columns with diameters over 15 m Examples of unlikely uses of distillation:  High-purity silicon for computers (via SiCl 3 distillation)  Water – heavy-water separation (boiling point difference only 1.4C)

Reflux gives strange effects

INCREASED HEAT INPUT ) LOWER TEMPERATURE TOP SO SIMPLE.... and yet SO COMPLEX

Simple to model

Model stage i Usually most important!

Simple to model... yet difficult to understand SIMPLE TO MODEL  1920’s: Models known. Graphical solution (McCabe-Thiele)  1960’s: Simulation with computer straightforward No need for more work!? BEHAVIOR NOT SO SIMPLE TO UNDERSTAND  Mathematician: Large number of coupled equations Nonlinear equations (mainly VLE) Complex behavior expected  Simulation and experience Not so complex Dynamic response: simple  More simulations: Maybe not so simple after all Instability Multiple steady states

Dynamic behavior simple! Example: Composition response of propane- propylene splitter with 110 stages and large reflux

Propane-propylene with 110 stages. Increase reflux. Simulated composition response with detailed model. XDXD XBXB 2000 min X feed stage Observed: “Simple” first-order responses with time constant about 6 h = 400 min

Increase in reflux. Mole fraction propylene on all stages feed stage All stages have a very similar slow first-order response! Behaves like “a single mixing tank” Why? Reflux gives strong interactions between the stages SO COMPLEX (model)... and yet SO SIMPLE (response)

Dynamic behavior simple? 1970’s and 1980’s: Mathematical proofs that dynamics are always stable  Based on analyzing dynamic model with L and V [mol/s] as independent variables In reality, independent variables are  L w [kg/s] = L [mol/s] ¢ M [kg/mol]  Q B [J/s] = V [mol/s] ¢  H vap [J/mol] Does it make a difference? YES, in some cases!

Molar and mass reflux t=0: z F is decreased from 0.5 to L w [kg/s]= L[mol/s]/M where M [kg/mol] is the molecular weight, Data: M L =35, M H =40. What is happening? Mole wt. depends on composition: more heavy ! M up ! L down ! even more heavy...) Can even get instability! With M H =40, instability occurs for M L <28 (Jacobsen and Skogestad, 1991)

Instability for “ideal” columns: Many people didn’t believe us when we first presented it in 1991! Likely to happen if the mole weights are sufficiently different

I IIIII I IV II Reflux back again.... but not composition !? Reflux Top composition SO SIMPLE.... and yet SO COMPLEX

Multiple steady state solutions IIIIV I II V

I IIIIV II V

V IV I

Actually not much of a problem with control! This is why you are not likely to notice it in practice... unless you look carefully at the reflux....will observe inverse response in an unstable operating point (V) V IV I SO COMPLEX (no control).... and yet SO SIMPLE (control)

Myth of slow control Let us get rid of it!!! Compare manual (“perfect operator”) and automatic control for typical column: 40 stages, Binary mixture with 99% purity both ends, relative volatility = 1.5  First “one-point” control: Control of top composition only  Then “two-point” control: Control of both compositions

“Perfect operator”: Steps L directly to correct steady-state value (from 2.70 to 2.74) Disturbance in V Want x D constant Can adjust reflux L Myth about slow control One-point control

“Perfect operator”: Steps L directly Feedback control: Simple PI control Which response is best? Disturbance in V CC x DS Myth about slow control One-point control

SO SIMILAR (inputs)... and yet SO DIFFERENT (outputs)

Myth about slow control Two-point control “Perfect operator”: Steps L and V directly Feedback control: 2 PI controllers Which response is best? CC x DS: step up CC x BS: constant

Myth about slow control Two-point control SO SIMILAR (inputs)... and yet SO DIFFERENT (outputs)

Myth about slow control Conclusion: Experience operator: Fast control impossible  “takes hours or days before the columns settles” BUT, with feedback control the response can be fast!  Feedback changes the dynamics (eigenvalues)  Requires continuous “active” control Most columns have a single slow mode (without control)  Sufficient to close a single loop (typical on temperature) to change the dynamics for the entire column

Complex columns Sequence of columns for multicomponent separation Heat integration Pressure levels Integrated solutions Non-ideal mixtures (azeotropes) Here: Will consider “Petlyuk” columns

Typical sequence: “Direct split” A,B,C,D,E,F A F B C D E

3-product mixture A+B+C A+B A B C 1. Direct split A+B+C A+B A B C B+C A+B+C A B C B+C 3. Combined (with prefractionator) 2. Indirect split

Towards the Petlyuk column A+B A B C B+C A+B A B C B+C A+B A B C B+C 4. Prefractionator + sidestream column liquid split vapor split 5. Petlyuk 30-40% less energy A+B+C 3.

Implementation of Petlyuk in single shell A A+B B B+C A+B+C B (pure!) C 6. DIVIDED-WALL IMPLEMENTATION in single shell! Gives about 40% savings also in capital thermodynamically equivalent (both about 40% savings in energy) C A+B+C A SO COMPLEX.... and yet SO SIMPLE 5. PREFACTIONATOR IMPLEMENTATION “Thermally coupled” with single reboiler and single condenser

Montz GC – Chemicals Research and Engineering Dividing Wall Columns Off-center Position of the Dividing Wall ≈≈

V min -diagram for Different Distillation Arrangements  = D C1 /F V T /F P A/B P B/C P A/C V min (C1) V min (Petlyuk + ISF/ISB) C1 C21 SO COMPLEX.... and yet SO SIMPLE (to estimate enrgy) V min (A/B) V min (B(C)

Divided wall columns: starting to catch on 1940’s: first patent 1960’s: Thermodynamic analysis (Petlyuk) 1984: First implementation (BASF) 2005: BASF has about 50 divided wall columns  also in Japan, South Africa... Control issues still not quite solved  but I think it should be rather easy

4-product mixture A,B,C,D A B C D A – iC5 B – nC5 C – iC6 D – nC6 Direct optimal extension of Petlyuk ideas requires two divided walls. Will look for something simpler Conventional sequence with 3 columns

4-product mixture: Kaibel column A+B A B D C+D ABCD C D A B (pure!) C (pure!) Alternative 3-column sequence Kaibel: 1 column!! More then 50% capital savings Also saves energy (but maybe not exergy) A – iC5 B – nC5 C – iC6 D – nC6

Control of Kaibel column Prefractionator: Close 1 “stabilizing” temperature loop Main column Close 3 “stabilizing” temperature loops Close a “stabilizing” temperature (profile) loop for each split D SO COMPLEX.... and yet SO SIMPLE (to operate)

H=6m D=5cm F S1 S2 B D

Conclusion Distillation is important Distillation is unbeatable (in some cases) Distillation is fun Distillation is complex yet simple... and vice versa

column, uses, when use? strange responses... increase heat.. T drops model complex: would expect complicated behavior... yet simple: show typical response e.g all stages response simple: expect always stable... yet complex: can be unstable with mss (Lw V) NEW column configurations... “easy first” Petlyuk. Kaibel. make drawing of how it evolves Better. heat-integrated Petlyuk (prefrac). Hidic \item BATCH DISTILLATION (Reflux) \item MODEL, DYNAMICS (Feedback) \item CONTROL (Steady-state misleading) \item MULTIPLE STEADY STATES AND INSTABILITY (Nonlinearity and feedback) \item INTERLINKED COLUMNS (Parallel paths) \item BATCH DISTILLATION AGAIN \item SYSTEMS VIEW \item CONCLUSION

The response is nonlinear....

The response is nonlinear.... but this can be corrected by taking log X D = ln(x DL /x DH )xDxD SO SIMPLE.... and yet SO COMPLEX

Distillation control CC LV Two-point LV TC TsTs xBxB CC xDxD

Refinery Main Fractionator

Can make problems... Detuned controller gain V V

Multi-Effect Prefractionator Additional large energy savings