Batch Comparison Overlay

A batch comparison overlay is an MES visualization that superimposes time- or phase-aligned traces from multiple batches to compare how a process actually behaved across runs. It is most powerful when it unites ISA‑88 batch context (procedures, unit procedures, operations, phases) with Level 2 historian signals, alarms, calculated parameters, and quality results. Typical uses include comparing current execution to a golden batch profile, visualizing parameter drift across campaigns, and providing evidence during deviation or OOT/OOS investigations.

Overlays differ from single-batch trends by normalizing for differences in timing and execution, handling pauses, restarts, and equipment variability. Sophisticated overlays permit phase-based alignment, normalization to percentage-of-phase, dynamic time-warping windows, and filtering by lot, equipment, material, or environmental covariates. In regulated settings the overlay itself is a controlled, Part 11/Annex 11-governed decision-support record: configurations, limits, and annotations must be auditable.

Batch overlays sit at ISA‑95 Level 3 (MES/operations management) as a composite of execution data (eBMR/eDHR context, recipe versions, operator entries) and Level 2 signals (controllers, historians). The MES resolves batch boundaries, maps signals to context, and supplies user attribution and approvals. Integration often spans OPC UA/DA for live tags, historian APIs for time-series (e.g., query by batch ID or time window), and MES data services for batch genealogy and master data.

ISA‑95 Level Mapping

ISA‑88 modeling anchors alignment: unit procedures and phases provide deterministic anchors for sync markers. When a phase repeats (e.g., hold, ramp, soak), the overlay can segment traces into comparable sub-runs, enabling aggregation or percentile bands across repetitions.

Regulators expect that visualizations used to support batch release, investigations, or CPV are trustworthy, attributable, and reproducible. 21 CFR 211.188 requires complete and accurate batch production and control records; if overlays are referenced in such records, they must be preserved with their generating parameters. 21 CFR Part 11 and EU GMP Annex 11 require audit trails for creation/modification of electronic records, user access controls, validated software, and preservation of original data. MHRA’s GxP Data Integrity guidance reinforces ALCOA+ and the need to retain raw data along with any transformations and derived views.

• Audit trail: who created/modified an overlay view, filters, time windows, and annotations;
• Retain original raw time-series and the exact transformation/alignment recipe;
• RBAC to segregate roles (author, reviewer, approver) and prevent unauthorized changes;
• Validated calculations and overlays (intended use, risk assessment, test coverage) per GAMP 5.

Accurate comparison requires defining batch boundaries, segmenting on ISA‑88 structures, and normalizing time across runs of unequal durations. Common approaches include: (1) phase-percentage normalization (0–100% for each phase); (2) anchor-event alignment (e.g., at peak exotherm or dissolved oxygen threshold); and (3) constrained dynamic time warping (DTW) to allow limited stretching while preserving causality. The MES must preserve the mapping rules and parameters as part of the record.

1. Phase-based alignment: Align start/end of each S88 phase; interpolate shorter/longer runs to common grids.
2. Event-locked alignment: Place t=0 at a reproducible event (e.g., feed start), then overlay subsequent segments.
3. Constrained DTW: Apply a warping window (e.g., Sakoe–Chiba band) to avoid overfitting while handling small timing skew.

Real-world complexities include batch pauses/resumes, partial reworks, equipment unit swaps, and sensor replacements mid-run. Robust overlays surface these discontinuities with markers and optionally exclude incommensurate segments. Calculated tags (e.g., running deltas, rate of change, mass/energy balances) must specify formulas and units, with version control; the audit trail should tie any reprocessing of time-series to the reason, approver, and effect on visual outputs.

Per GAMP 5, overlays are typically Category 4 (configurable) unless bespoke code elevates them toward Category 5. Validation should be risk-based to intended use: if overlays directly influence batch disposition or deviation conclusions, they demand higher rigor. Define URS around alignment logic, supported tag types, security, and audit trails; implement IQ/OQ/PQ with traceability to risks and requirements; and maintain change control for thresholds, filters, and calculation libraries.

• Test cases: alignment accuracy (phase anchors, event locks), interpolation methods, and edge cases (pauses, reworks).
• Security and Part 11: RBAC, e-signature on critical annotations, session timeouts, and audit trail completeness.
• Data integrity: verify fidelity from source historian to overlay (spot-check timestamps, units, and scaling).
• Performance: ensure overlays remain usable with N≥50 batch traces and high-frequency tags (e.g., 1 Hz).

SOPs should define who may create/edit overlay configurations, how overlays are reviewed and approved for use in quality decisions, and how they are archived with batch records. Train users to interpret overlays statistically (confidence bands vs. spec limits) to avoid overreliance on visual impressions alone.

Overlays often include percentile bands (e.g., P10–P90), mean ± 2σ envelopes, or control bands based on SPC models. The golden batch can be a single exemplar or, more defensibly, a statistical band from a population of good batches. Avoid conflating engineering limits (alarms/interlocks) with quality specification limits; overlays should display both distinctly. For CPV, overlays can expose sustained shifts or increasing variance that warrant CAPA or process robustness work.

Practical Practices

• Derive golden bands from a vetted set of conforming lots; exclude known-atypical runs.
• Display 95% confidence envelopes for the mean trajectory and a separate percentile band for variability.
• Flag out-of-trend (OOT) excursions based on residuals from the golden trajectory, not only absolute limits.
• Support multi-parameter views: subtle defects surface only when temperature, delta‑P, and feed rate are seen together.

When overlays are used in investigations, tie visual anomalies to quantitative metrics: duration above band, area between curves, time to return to band, and cross-correlation among signals. Document the thresholds and statistical rationale in controlled procedures; keep the underlying dataset and selection criteria to enable reproduction.

High-quality overlays demand clean interfaces to historians and context sources. Architect for pull (query-by-batch) and push (event-stream) models; normalize tag metadata (engineering units, scaling, calibration status). Merge batch context (recipe version, parameters, material lots, equipment IDs) so users can filter overlays by meaningful strata. Protect interfaces and data at rest/in transit, with full traceability from sensor to screen.

• Signals: controller tags (OPC UA/DA), historian APIs, alarm/event logs, calculated KPIs.
• Context: MES eBMR context, material genealogy, equipment state, operator interventions.
• Quality: LIMS assay and in-process results aligned at sample timestamps; laboratory OOS flags.
• Security: RBAC, network segmentation between L2 and L3, and client hardening; audit of data access.

Data preparation often includes resampling to common grids, unit harmonization, and gap handling. Mark missing data explicitly to distinguish it from in-band behavior. Version control any calculation pipelines (e.g., smoothing filters) and ensure a reviewer can reconstruct the exact view shown at a decision point.

Pharmaceutical biologics: Overlay dissolved oxygen, pH, agitation, and feed rates across fermentation runs; align by inoculation start and feed-phase transitions. Combine with CPV residual analysis to detect slow drift in oxygen transfer. Sterile fill-finish: overlay temperature and differential pressure across line startups to tame instabilities that precede rejects. Radiopharma: overlay synthesis cycles with half-life decay correction and time-critical phase windows, ensuring time bases are tightly synchronized.

Medical devices: For polymer curing or bonding, overlay temperature and humidity soaks and post-cure profiles, correlating to tensile test outcomes. Food/sausage thermal processing: overlay cook and chill core-temperature curves, yield, and lethality metrics; align on lethality integrals to standardize comparison across variable loads. Chemicals and polymers: overlay reactor jacket inlet/outlet temperatures, ∆T, monomer feed, and torque; use event-locked alignment at initiator addition to compare exotherm kinetics.

• Cosmetics and creams: overlay homogenization speed, vacuum level, and viscosity vs. time to control microstructure.
• API wet granulation: overlay binder addition rate and impeller/chopper RPM with LOD and PSD endpoints.
• Plastics/resins: overlay molecular weight indicators from QC with reactor temperature and time-under-heat.

Formalize overlays in SOPs: define intended use, configuration authority, review/approval of new golden bands, and how overlays inform decisions (release, deviations, CAPA, process changes). Train users on interpretation pitfalls and on the separation between engineering alarms and quality specifications. Define metadata to capture in eBMR/eDHR when an overlay informs a decision, including the overlay configuration version and dataset inclusion/exclusion criteria.

• Record retention: retain raw data and reconstructable overlays for at least the product’s GMP retention period.
• Periodic review: reassess golden bands as processes mature or changes occur (per change control).
• Access review: periodically confirm RBAC mappings and deactivate stale overlay configurations.
• Supplier oversight: if using third-party historians or analytics, qualify vendors and their change practices.

Ensure that overlays are referenced—not embedded as untraceable images—in batch records. Where images are stored, link them to their generating query and configuration artifacts, with checksums to detect tampering. For products under enhanced monitoring, define overlays as required attachments to release packages to enable consistent review.

V5 Ultimate implements overlays as a governed MES function tied directly to the batch record. Signals from Level 2 historians, operator inputs, alarms, and LIMS results are aligned on ISA‑88 phases with configurable normalization (phase-percentage, event-anchored, constrained DTW). Every overlay view stores its configuration, filter criteria, and annotations under Part 11/Annex 11 controls, with RBAC and audit trails. Reviewers can pivot from an excursion to the linked deviation, CAPA, or maintenance work order and back, preserving context.

• Golden band management with version-controlled populations and approval workflows;
• Overlay-to-record linkage: each displayed panel is a reconstructable artifact tied to the batch;
• Cross-module joins: weigh/dispense, eBMR/eDHR, LIMS, WMS, and maintenance events in one timeline;
• CPV integration: push overlay residuals and metrics to CPV dashboards for sustained trending.

Overlays amplify insight—but also errors—if foundations are weak. The most frequent failure is time-base mismatch across systems; second is mis-specified alignment leading to false deviations. Data selection bias (excluding bad runs from the golden set) creates over-optimistic bands. Finally, uncalibrated sensors and unit conversion mistakes can masquerade as process drift.

• Time hygiene: enforce NTP, monitor drift, and display time-offset diagnostics in the overlay UI.
• Alignment QA: validate anchors against test datasets; expose alignment parameters to reviewers.
• Golden set governance: define inclusion/exclusion rules in SOPs; revalidate after process changes.
• Calibration traceability: surface calibration due/overdue status on traces; block overlays that mix incompatible calibrations.
• Sampling harmonization: resample with documented methods; visualize data gaps distinctly.
• Change control: treat overlay thresholds and filters as controlled configuration with impact assessment.

When overlays inform release, include them in the batch disposition package with reviewer sign-off, and backstop them with statistical evidence (e.g., quantified residuals). For investigations, export the overlay context—data lineage, transforms, and notes—so independent reviewers can reproduce findings. Where ICS or historian upgrades occur, perform equivalence checks to confirm continuity of overlays across versions.