Procedure Function Chart

ISA-88.01 §5.5 defines the procedural control model in four levels: Procedure (the top-level intent — "make Product X"), Unit Procedure (work on one unit — "prepare reactor"), Operation (a complete processing activity within a unit — "granulation"), and Phase (the smallest procedurally-controlled action — "charge water").

The PFC is the graphical representation of that hierarchy: rectangles for steps, diamonds for decisions, parallel bars for synchronisation, transitions with named conditions. It is to a recipe what a P&ID is to a process: the canonical readable map.

• Top-down hierarchy — a Procedure box expands into a chain of Unit Procedure boxes, each of which expands into Operations, each into Phases.
• Transitions — named, boolean conditions on the arrows between steps (e.g. "temperature ≥ 65 °C AND hold-timer ≥ 30 min").
• Parallel divergence — synchronised parallel branches (e.g. CIP one reactor while charging another), joined again at a parallel convergence bar.
• Selective branching — one path of several taken based on conditions (e.g. potency-based dispense alternatives).
• Loops — repeat constructs for iterative additions (e.g. "charge ingredient until pH within range").

A recipe expressed purely as code is unreviewable by operators and unauditable by QA. The PFC is the artefact that bridges domains:

• Operators see the path the batch will take and where they will be prompted.
• Process engineers see the synchronisation points where bottleneck delays occur.
• QA sees the decision logic and can challenge each transition condition.
• Auditors see how the master recipe encodes the master production record (21 CFR 211.186).
• Validation can map IPCs and CPPs to specific PFC steps, making PPQ scope unambiguous.

PFCs sit at the recipe layer (MES, batch manager). SFCs (sequential function charts, IEC 61131-3) sit at the PLC layer. They look similar — both have steps, transitions and parallel branches — but their roles differ:

Both can coexist: a PFC phase "Heat to 65 °C" delegates to an SFC inside the reactor's equipment-module logic that ramps the jacket, monitors the loop and signals completion.

• One entry and one exit per construct — no spaghetti control flow.
• Every transition has a boolean condition, named so it can be referenced in deviations.
• Parallel divergence requires explicit parallel convergence — no implicit joins.
• Selective branches must be mutually exclusive at the divergence point.
• Loops have a bounded exit condition — infinite loops fail QA review.
• Steps reference phase classes from the phase class library, not inline logic.
• Synchronisation across units uses named events, not direct cross-PLC writes.

When a batch deviates — operator override, equipment hold, IPC out of trend — the deviation is anchored to a specific PFC step or transition. This anchoring is what makes post-batch review tractable:

• The eBR captures "deviation occurred at Operation 3.2, Step 'Hold for crystallisation'".
• Reviewer can navigate from the deviation entry directly to the PFC node, see the planned path and the actual path.
• Trend analysis groups deviations by PFC node — recurring failures cluster on specific steps, focusing CAPA effort.
• Recipe revision impact assessment is scoped — "this PFC node changed" identifies the precise validation surface.

• OSD pharma — Procedure "Manufacture Tablet X" → Unit Procedures (Dispense → Granulate → Dry → Mill → Blend → Compress → Coat → Pack), each with operations and phases per ISA-88.
• Biopharma — Procedure "Produce Bulk Drug Substance Lot" with parallel branches for inoculum train, production bioreactor and buffer preparation, synchronised at harvest.
• Food — Procedure "Brew Lager Lot" with selective branches by season (winter recipe vs summer recipe) and loops for fermentation duration based on gravity readings.
• Cosmetics — Procedure "Emulsion Cream" with parallel oil-phase and water-phase preparation, synchronised at emulsification.
• Chemicals — Procedure "Polymerisation Campaign" with iterative monomer addition loops driven by molecular-weight IPCs.

• PFCs maintained as Visio diagrams disconnected from the running recipe — divergence guaranteed within months.
• Transition conditions in free text rather than executable expressions — ambiguity at runtime.
• Parallel branches without explicit convergence — race conditions hidden in code.
• Phase steps written inline rather than referencing the phase class library — recipe duplication and validation explosion.
• PFC node naming non-unique — deviations cannot be anchored unambiguously.
• No version diff visualisation — recipe revision review is blind.
• PFC printed only for paper batch records — operators do not see it on the kiosk, defeating its purpose.