Process design and optimisation

In short

Process design converts a product requirement into a complete, buildable and profitable plant: a flowsheet, mass and energy balances, equipment duties and sizes, utilities, an estimate of capital and operating cost, and a profitability number. Optimisation is the quantitative half of it - choosing operating conditions, sizes and structures that minimise total annual cost or maximise net present value, subject to balance, equilibrium, safety and statutory constraints. Almost every design decision is a trade-off between capital cost and operating cost, and the examinable skill is writing that trade-off as an objective function and finding its optimum. It is a heavily weighted GATE (CH) section ("Plant Design and Economics") and the core of the design project in every B.E./B.Tech chemical curriculum. Design, techno-economics and simulation roles at EIL, Technip Energies, L&T Hydrocarbon, Toyo, Jacobs, Tata Consulting Engineers, Reliance, IOCL/BPCL/HPCL, UPL and Aarti Industries interview directly on this material.

Explained

What process design covers, and in what order

• Stages: process research and lab data → conceptual (feasibility) design → basic/front-end engineering design (FEED, including PFDs and heat and material balance) → detailed engineering (P&IDs, datasheets, layout) → procurement, construction, commissioning.
• Design classes: *new (grassroots)* design versus *retrofit/revamp* (an existing plant constrains what can be changed - most Indian work is revamp and debottlenecking).
• Onion diagram - the order in which decisions must be taken, each layer fixed by the one inside it: reactor → separation and recycle → heat recovery network → utilities → effluent treatment. Changing the reactor changes everything outside it, so never optimise an outer layer first.
• Design basis document fixes plant capacity, on-stream factor (typically 0.90-0.95, i.e. about 7,900-8,300 h/yr), feed and product specifications, site conditions, utility conditions and battery limits. Every later number depends on it.
• Design margin (over-design): equipment is sized above the balance duty - commonly 10% on heat-exchanger area, 10-20% on pump head and flow, extra trays on columns. Margins compound, so they must be applied once, deliberately.
• Hierarchy of objectives: technically feasible → safe and statutorily compliant → environmentally acceptable → economic optimum. An "optimum" that violates a constraint is not a solution.

Flowsheet synthesis and degrees of freedom

• Douglas's hierarchical decision procedure: (1) batch or continuous; (2) input-output structure of the flowsheet (feed purification, by-product handling, purge); (3) recycle structure and reactor choice; (4) separation system synthesis (vapour and liquid recovery); (5) heat exchanger network and utilities. Economic potential is evaluated at each level and clearly bad branches are pruned early.
• Key structural decisions: whether to purify feed before the reactor; whether to recover and recycle unconverted reactant or purge it; gas recycle and purge rate; number of product streams; by-product as product, fuel or waste.
• Reactor-separator-recycle trade-off: low single-pass conversion means a smaller, cheaper reactor but a larger recycle, larger separation and higher pumping/reboiler duty - so there is an economic optimum conversion (often 95-99% for a cheap reactant, much higher when the reactant is costly).
• Degrees of freedom (design variables) = number of unknown variables − number of independent equations. Only that many values may be specified; more makes the problem over-specified (inconsistent), fewer leaves it under-specified (no unique answer). A simple mixer, a flash, a column - each has a countable DOF, and in simulators the DOF are exactly what you type in.
• Separation sequencing heuristics (for an N-component mixture there are many possible column sequences - heuristics avoid enumerating all):
  • remove corrosive, hazardous or thermally unstable components first;
  • do the easiest separation (largest relative volatility) first;
  • remove the most plentiful component early;
  • favour near-equal-sized splits (50:50) when relative volatilities are similar;
  • leave the most difficult / highest-purity separation last;
  • avoid refrigeration and vacuum if a simple change of pressure will do.
• Short-cut then rigorous: size with short-cut methods (Fenske-Underwood-Gilliland for columns, U∆T_lm for exchangers) to screen alternatives cheaply, then confirm the surviving flowsheet rigorously in a simulator.
• Inherent safety and environmental limits enter at synthesis, not afterwards: minimise inventory of hazardous material, substitute a less hazardous solvent, moderate conditions, prefer closed-loop recovery to end-of-pipe treatment (detailed hazard study belongs to the HAZOP/PSM topic).

Capital cost estimation

• Estimate classes (AACE-style, names vary by company): order-of-magnitude/study (±30-50%, from capacity ratios), preliminary/budget authorisation (±20-30%, from equipment list and factors), definitive (±10-15%, from quotations and take-offs), detailed/contractor (±5-10%).
• Equipment purchased cost from vendor quotes, in-house databanks or published correlations (Peters & Timmerhaus, Sinnott/Towler, Turton), always with a stated cost basis (year, material, pressure, capacity range) plus material and pressure factors.
• Capacity scaling - the six-tenths rule: cost ∝ capacityⁿ with n ≈ 0.6 typically (range roughly 0.4-0.9; n = 1 for identical parallel trains). Valid only within the correlation's size range - this is the *economy of scale*.
• Time escalation with cost indices: CEPCI (Chemical Engineering Plant Cost Index, published monthly), Nelson-Farrar for refineries, Marshall & Swift. In India, WPI for machinery and project-cost indices are also used. Index values move every month; always quote the index value you used rather than relying on a remembered number.
• Fixed capital investment (FCI) = ISBL (inside battery limits: equipment, erection, piping, instruments, electrical, civil, structures, insulation, painting) + OSBL (offsites: utilities, storage, roads, buildings) + engineering and construction supervision + contingency (typically 10-15%).
• Factorial methods: *Lang factor* FCI = f_L × Σ(purchased equipment cost), with f_L ≈ 4.74 for a fluid-processing plant, 3.63 solid-fluid, 3.10 solid processing (Peters & Timmerhaus set, including engineering and contingency). *Hand/Guthrie factors* apply a separate factor to each equipment type - more accurate.
• Working capital (raw material and product stocks, spares, receivables, cash) ≈ 10-20% of FCI; recovered at the end of the project. Total capital investment = FCI + working capital (+ land, + start-up cost).
• In India, GST on plant and machinery is generally 18% (slabs are 5%/18%/40% since 22 September 2025) but is largely recoverable as input tax credit for a taxable manufacturer, so the estimate should be built on pre-tax cost with non-creditable taxes and duties shown separately; customs duty on imported equipment is not creditable in the same way and must be added.

Operating cost and profitability

• Cost of manufacture (COM) = direct costs (raw materials, utilities, operating labour and supervision, maintenance, operating supplies, laboratory) + fixed charges (depreciation, local taxes, insurance, rent) + plant overheads; total product cost adds general expenses (administration, distribution and selling, R&D).
• Turton's convenient factored form uses FCI, operating labour, utilities, waste treatment and raw materials with fixed multipliers (see formula box); raw material and utility cost usually dominate in bulk chemicals.
• Depreciation: straight line, declining balance or sum-of-years-digits in textbooks; in Indian practice, book depreciation follows the Companies Act, 2013 (useful-life basis, Schedule II) and tax depreciation follows the written-down-value blocks under the Income-tax Act, 2025 (which replaced the 1961 Act from 1 April 2026 and renumbered the sections) - quote the provision, not a remembered section number.
• Profitability measures: simple payback, return on investment (ROI), net present value (NPV), discounted cash flow rate of return (DCFROR = IRR), equivalent annual cost. Non-discounted measures ignore the time value of money and favour short projects; NPV and IRR are what a board actually sees.
• Decision rules: accept if NPV > 0 at the company's cost of capital; rank mutually exclusive options by NPV, not by IRR (IRR can mislead for different scales or non-standard cash-flow signs). Sensitivity analysis on price, feed cost, capacity and capital (spider/tornado plots) is expected in any design report.
• Cash-flow diagram for a project: land and FCI outflow during construction, negative cash flow at start-up, positive cash flow through the operating life, working capital and salvage recovered at the end.

Optimisation: formulation and methods

• Statement: minimise (or maximise) an objective function f(x) over decision variables x, subject to equality constraints h(x)=0 (balances, equilibrium, design equations) and inequality constraints g(x) ≤ 0 (capacity, purity, safety, environmental, bounds).
• Objective choice matters: total annual cost, NPV, profit per tonne, yield, energy per tonne, CO₂ per tonne. The optimum moves when the objective changes - a yield optimum is not a cost optimum.
• Classification: unconstrained vs constrained; continuous vs integer vs mixed; LP (linear objective and constraints - simplex method, global optimum at a vertex), NLP (SQP, reduced-gradient, interior point), MILP/MINLP (structural choices such as "build this column or not" - branch and bound, outer approximation), stochastic/global methods (genetic algorithms, simulated annealing, particle swarm) for non-convex or black-box problems.
• Single-variable optimum: set df/dx = 0 and check d²f/dx² > 0 for a minimum. Where a cost is the sum of a term falling as 1/x and a term rising as x, the optimum occurs where the two terms are equal.
• Constrained analytical optimum: Lagrange multipliers (∇f = Σλ∇h); the multiplier is the *shadow price* - the change in the optimum objective per unit relaxation of the constraint. KKT conditions extend this to inequalities.
• Search methods without derivatives: golden-section and Fibonacci search (single variable, unimodal), univariate/pattern search, Nelder-Mead simplex, response surface from designed experiments.
• Convexity: a convex objective over a convex feasible region has one optimum which is global. Chemical processes (recycles, bilinear blending terms, equilibrium) are usually non-convex, so a gradient method finds a *local* optimum - start from several points.
• Flowsheet optimisation in practice: sequential-modular simulators (Aspen Plus, Aspen HYSYS, PRO/II, CHEMCAD, the open-source DWSIM) with a built-in optimiser; equation-oriented solvers (gPROMS) for large simultaneous problems; Excel Solver, Python (SciPy, Pyomo) and GAMS for the algebraic models asked in coursework.
• Real-time optimisation (RTO) sits above the control layer: a steady-state model re-optimises set-points every few minutes and hands them to the regulatory controllers; data reconciliation and parameter fitting precede it.
• Multi-objective problems (cost vs emissions) have no single answer - produce a Pareto front and let the decision maker choose.

Classic design trade-offs worth memorising

• Reflux ratio: R near R_min needs infinite trays (high capital); large R needs huge reboiler and condenser duty. Economic optimum typically R ≈ 1.1-1.3 R_min.
• Heat exchanger: large area = high capital, low ∆T and low utility cost; the optimum approach temperature (∆T_min / EMAT) is typically 10-20 °C for liquid services, smaller with cheap area or costly refrigeration.
• Pipe diameter: small pipe is cheap to buy but costs more to pump (∆P ∝ 1/d⁵ roughly); there is an economic diameter.
• Insulation thickness: thicker insulation costs more but saves heat; optimum where marginal costs are equal.
• Multiple-effect evaporators: each extra effect cuts steam roughly in proportion to 1/N but adds a full set of capital; optimum N is usually small (2-6).
• Batch cycle time, reactor volume vs conversion, catalyst life vs regeneration frequency, compressor interstage pressure ratio (equal ratios minimise work for equal-efficiency stages), turndown vs efficiency.
• Optima are usually flat: total cost near the optimum changes little, so the practical choice is the nearest standard size (standard pipe NB, standard motor rating, standard vessel diameter) and robustness against uncertainty matters more than the last decimal.

Indian statutory and practice context

• Environmental clearance for new or expanded chemical plants under the EIA Notification, 2006 (as amended) - Category A (MoEFCC) or B (SEIAA), with public hearing and an EIA/EMP report; capital and operating cost of pollution control must be in the estimate.
• Consent to Establish and Consent to Operate from the State Pollution Control Board under the Water (Prevention and Control of Pollution) Act, 1974 and the Air Act, 1981; CPCB/SPCB discharge and emission standards and the Hazardous and Other Wastes (Management and Transboundary Movement) Rules, 2016 fix the effluent layer of the onion diagram.
• Labour and safety: the four Labour Codes are in force from 21 November 2025 and replace the Factories Act, the Minimum Wages Act, the Industrial Disputes Act and the BOCW Act - manning, shift patterns and welfare facilities in a design basis follow the OSH, Working Conditions and Safety Code. Boilers come under the Boiler Act and IBR; LPG, ammonia and solvent storage under PESO (SMPV, Gas Cylinders and Petroleum Rules) licences.
• Codes and standards feed the cost estimate: BIS standards for pipes, flanges, vessels, structures and electricals; ASME, API and TEMA where the client specifies them; IS 1893 and IS 875 for seismic and wind loads on structures. Many older IS numbers have been withdrawn or revised - confirm the current number and edition before citing it.
• Energy and incentives: Perform-Achieve-Trade (PAT) targets under the Energy Conservation Act for designated consumers, BEE and PCRA energy-audit practice, and the PLI schemes for bulk drugs and chemicals all change the economics of a design option in India.

Key points

• Onion diagram - decide reactor, then separation/recycle, then heat recovery, then utilities, then effluent; outer layers follow inner ones.
• Douglas hierarchy - five levels of decisions, with an economic screen at each level to prune alternatives.
• Degrees of freedom - variables minus independent equations; exactly that many specifications, no more.
• Six-tenths rule - cost₂ = cost₁ (Q₂/Q₁)^0.6; economy of scale, valid only in the correlated size range.
• Cost index - escalates a cost from one year to another (CEPCI, Nelson-Farrar); always state the index values used.
• Lang factor - FCI from total purchased equipment cost; about 4.74 for fluid-processing plants.
• FCI vs working capital - working capital is 10-20% of FCI and is recovered at project end; together they are total capital investment.
• Estimate accuracy - ±30-50% order of magnitude, ±20-30% preliminary, ±10-15% definitive.
• COM - raw materials, utilities, labour, maintenance, depreciation and overheads; raw materials usually dominate in bulk chemicals.
• NPV - discounted net cash flows minus investment; accept if positive at the cost of capital. IRR/DCFROR - the rate making NPV zero.
• Payback - FCI ÷ annual cash flow; ignores time value and life beyond payback.
• Objective + decision variables + constraints - a problem is not an optimisation problem until all three are written.
• LP - linear; optimum at a vertex; simplex finds the global optimum. MINLP - needed when structure (yes/no) is a variable.
• Lagrange multiplier = shadow price of the constraint; KKT generalises to inequalities.
• Convexity guarantees a single global optimum; process models are generally non-convex, so local optima are the norm.
• Optimum reflux ≈ 1.1-1.3 × R_min; optimum exchanger approach typically 10-20 °C.
• Flat optimum - round to standard sizes and test sensitivity rather than chase precision.
• Constraints first - a cheaper design that breaches a CPCB limit, a code or a safety requirement is infeasible, not optimal.

Formulas and examples

Example 1 - scaling and escalation. A 50 m² stainless shell-and-tube exchanger cost Rs 12.0 lakh when the plant cost index was 556.8. Estimate a 120 m² unit at an index of 800 (n = 0.6).

• Size factor: (120/50)^0.6 = 2.4^0.6 = e^{0.6 × 0.8755} = 1.691.
• Cost at old index: 12.0 × 1.691 = Rs 20.29 lakh.
• Escalate: 20.29 × (800/556.8) = 20.29 × 1.4368 = Rs 29.2 lakh (≈ ±30%, an order-of-magnitude estimate).

Example 2 - Lang factor and annualised capital. Purchased equipment for a solvent recovery plant totals Rs 8.0 crore (fluid processing, f_L = 4.74). Working capital 15% of FCI. Life 10 years, i = 12%.

• FCI = 4.74 × 8.0 = Rs 37.9 crore; working capital = 0.15 × 37.9 = Rs 5.69 crore → TCI = Rs 43.6 crore.
• (1.12)¹⁰ = 3.1058 → CRF = 0.12 × 3.1058/(3.1058 − 1) = 0.37270/2.1058 = 0.1770 /yr.
• Annualised capital charge = 0.1770 × 37.9 = Rs 6.71 crore/yr, the figure to add to operating cost in a TAC comparison.

Example 3 - profitability. FCI Rs 40 crore, net cash flow Rs 9 crore/yr for 10 years, discount rate 12%.

• AF = (1 − 1/3.1058)/0.12 = (1 − 0.3220)/0.12 = 5.650.
• PV of inflows = 9 × 5.650 = 50.85 → NPV = Rs 10.85 crore (> 0, accept).
• Simple payback = 40/9 = 4.44 years; discounted payback: cumulative PV reaches 37.00 at year 6 and 41.07 at year 7, so ≈ 6.7 years.
• IRR: AF required = 40/9 = 4.444. At 18%, AF = 4.494; at 19%, AF = 4.339 → interpolating, IRR ≈ 18.3%.

Example 4 - optimum insulation thickness (analytical optimum). For a steam line, annual heat-loss cost ≈ 7,200/x (Rs/yr, x in cm of insulation) and annualised insulation cost = 800x (Rs/yr).

• TAC = 7,200/x + 800x; dTAC/dx = −7,200/x² + 800 = 0 → x² = 9 → x_opt = 3 cm.
• TAC = 7,200/3 + 800 × 3 = 2,400 + 2,400 = Rs 4,800/yr; the two costs are equal at the optimum, and d²TAC/dx² = 14,400/x³ > 0 confirms a minimum.
• At 2 cm TAC = 3,600 + 1,600 = Rs 5,200; at 4 cm = 1,800 + 3,200 = Rs 5,000 - only 4% worse, i.e. a flat optimum, so the nearest standard thickness is acceptable.

Example 5 - linear programming (product mix). Products A and B give contributions of Rs 400/t and Rs 300/t. Reactor time: 2 h/t (A), 1 h/t (B), 100 h/week available. Separation: 1 h/t (A), 2 h/t (B), 80 h/week available.

• Maximise Z = 400A + 300B subject to 2A + B ≤ 100, A + 2B ≤ 80, A, B ≥ 0.
• Vertices: (50, 0) → Z = 20,000; (0, 40) → Z = 12,000; intersection of the two lines: B = 100 − 2A, so A + 200 − 4A = 80 → A = 40, B = 20.
• Z = 400(40) + 300(20) = 16,000 + 6,000 = Rs 22,000/week at A = 40 t, B = 20 t; both constraints are tight, so both have non-zero shadow prices - extra reactor or separation hours are worth paying for.

Where it is used

• Design consultancies and EPC: Engineers India Ltd, Technip Energies, L&T Energy Hydrocarbon, Toyo Engineering India, Jacobs, Thyssenkrupp UDI - roles in process design, techno-economics and proposals; cost estimates for refinery and fertiliser revamps.
• Refining and petrochemicals: IOCL, BPCL, HPCL, Reliance, HMEL, GAIL, ONGC Petro additions - LP models for refinery planning and crude selection, and optimum conversion/recycle studies in crackers and aromatics plants.
• Bulk chemicals, agrochemicals and pharma: UPL, Aarti Industries, Deepak Nitrite, SRF, Divi's, Sun Pharma - batch-versus-continuous decisions, solvent recovery economics, campaign scheduling, PLI-backed capacity expansions.
• Fertilisers, cement and metals: optimum number of evaporator effects, waste-heat recovery projects, PAT energy targets, de-bottlenecking to raise capacity at minimum capital.
• Job roles: process design engineer, process simulation/modelling engineer, cost estimation and proposals engineer, technical services and energy-efficiency engineer, real-time optimisation and APC engineer, and project feasibility/DPR preparation for banks and investors.
• Simulation and modelling tools used day to day: Aspen Plus, Aspen HYSYS, PRO/II, CHEMCAD, DWSIM (free), Excel Solver, Python (SciPy, Pyomo), GAMS, and Aspen Capital Cost Estimator for estimates.

Interview questions

• Q: What is the onion diagram and why does the order matter? A: It is the layered order of design decisions - reactor, then separation and recycle, then heat recovery, then utilities, then effluent treatment. Inner decisions set the duties and streams for the outer ones, so optimising an outer layer around a wrong reactor choice wastes effort.
• Q: State the six-tenths rule and its limitation. A: Cost scales as capacity^0.6, so doubling capacity raises cost by about 52% - economy of scale. It is a correlation valid only within the size range and technology it was derived from, and breaks down once parallel trains are needed.
• Q: A pump cost is known for 2015; how do you bring it to today? A: Multiply by the ratio of the current plant cost index to the 2015 index (CEPCI or an Indian machinery index), then apply material, capacity and pressure factors. The index value used must be stated, since it changes monthly.
• Q: What is the Lang factor and what does it include? A: A multiplier on total purchased equipment cost to obtain fixed capital - about 4.74 for a fluid-processing plant - covering installation, piping, instruments, electricals, civil work, engineering and contingency, but not working capital or land.
• Q: Why prefer NPV over payback? A: NPV discounts all cash flows over the whole project life at the cost of capital, so it measures value added; payback ignores the time value of money and everything happening after the payback point.
• Q: Why is the optimum reflux ratio above the minimum? A: At R_min trays become infinite, so capital is unbounded; raising R cuts trays but raises reboiler, condenser and column diameter costs. Total annual cost minimises at roughly 1.1-1.3 R_min.
• Q: How do you decide degrees of freedom for a flowsheet? A: Count unknown stream and unit variables, subtract independent equations (balances, equilibrium, specified relations); the remainder are the design variables you may specify, and specifying a different number makes the problem over- or under-specified.
• Q: What does a Lagrange multiplier mean physically? A: It is the shadow price - the improvement in the objective per unit relaxation of that constraint, for instance rupees per extra hour of reactor time or per extra tonne of steam.
• Q: When is an MINLP needed rather than an NLP? A: When the decision includes discrete structure or integer quantities - whether to install a column, how many effects or trains, which feed to use - alongside continuous operating variables.
• Q: Your gradient optimiser gives different answers from different starting points. Why? A: The model is non-convex (recycles, bilinear and equilibrium terms), so the solver converges to local optima; use multiple starts, good initialisation or a global/stochastic method and compare objective values.
• Q: Cheapest design versus the one you would build - which do you pick? A: The one that meets all constraints - safety, codes, SPCB consent limits, turndown and operability - at the lowest total annual cost; a lower-cost infeasible or unoperable design is not a candidate at all.
• Q: How does an Indian plant's statutory environment change the economics? A: Environmental clearance under the EIA Notification, 2006 and SPCB consents under the Water Act, 1974 and Air Act, 1981 fix effluent and emission limits, so treatment capital and operating cost are part of the base case; PAT energy targets and PLI incentives shift the optimum towards energy recovery and larger domestic capacity.

Common mistakes

• Comparing a capital cost with an annual operating cost directly - capital must be annualised with a CRF (or operating cost brought to present value) before adding.
• Escalating cost with an index but forgetting capacity scaling, or vice versa; or using an index from a different country or sector without saying so.
• Over-specifying a flowsheet in a simulator (more specifications than degrees of freedom) and then blaming the solver for not converging.
• Applying design margins at several stages so a 10% margin silently becomes 40%, inflating both equipment size and cost.
• Reporting a factored estimate to the rupee - a ±30% number should be rounded and its accuracy class stated.
• Treating a local optimum, or the best of a handful of hand-picked cases, as *the* optimum, and skipping the sensitivity analysis.
• Optimising one unit in isolation - minimising reflux in a column while ignoring the extra trays, taller shell and the heat-integration opportunity outside it.

Related topics

Pinch analysis and energy integration · Process equipment design · Process flow diagrams and P&IDs · Mass transfer: distillation and absorption · Chemical reaction engineering: reactor design · Plant safety: HAZOP and PSM