Visual inspection software.

Pharmaceutical visual inspection is a control on a product population, and its record has to do more than count rejects. It must show which units were eligible, which method and settings applied, who or what inspected each population, what defects were observed, what was reinspected, and how every unit reached a final state.

This matters most when the obvious summary misleads. A batch can sit below its alert rate while one critical defect cluster maps to a filling interval, component lot or machine head. A high reject rate can be operational noise, or evidence that the batch should not proceed. The decision lives in the pattern and the genealogy.

1The requirement and the population are explicit.

Product, container-closure system, presentation, market, fill characteristics, route and lifecycle stage determine the effective inspection requirements: 100% inspection, sampling obligations, defect classes, acceptance rules, inspection conditions, method eligibility, reinspection limits, qualification needs and release evidence.

The inspectable population connects filled units to the batch, filling interval, container and closure lots, line, machine head or lane, intervention windows, storage and transfers. Units removed before inspection, damaged in handling, used for setup, challenged, sampled or destroyed do not vanish into a yield calculation; their reasons and states stay distinct.

1.1Why teams choose Seal for visual inspection

Visual inspection is usually recorded as reject counts on a batch record page, with the inspection machine’s data, inspector qualifications and defect library held in separate systems. Each observation is linked to the unit population, its filling interval, component lots, inspector or recipe and the challenge state in force, so a defect cluster can be traced to its source before disposition. A revised defect class, challenge set or recipe applies to later inspections, and earlier batches keep the definitions they were inspected against.

2The defect library teaches the boundary.

Defects are defined by product and container context, location, visual characteristic, severity, detectability, probable mechanism, reference images or physical exemplars, and decision rule. Critical, major and minor are governed classifications, not operator preferences, and a crack, missing stopper, fibre, glass particle or cosmetic scuff can mean different things across presentations.

A useful library holds clear positives, acceptable conditions, near-boundary examples, look-alikes and known process artefacts, with provenance, approval and effective dates. Retired or reclassified examples stay traceable to the inspections and qualifications that used them, and a new defect family can prompt a review of previously coded observations.

Swipe to see the full diagram →

3Manual and automated inspection each have a controlled state.

Manual inspection runs under a defined station, background, illumination, viewing distance, agitation, dwell, pace, rest and rotation. The operator confirms readiness and records interruptions, breaks, lighting checks and escalations. Annex 1 expects manual inspectors to be qualified at least annually, including consideration of eyesight checks, and those prerequisites can be configured as a condition of execution.¹

Qualification measures detection performance. Challenge sets of known defective and acceptable units have controlled identity, difficulty, randomisation, reuse and retirement, and each qualification records true positives, false negatives, false positives and classification accuracy. Qualification can vary by product family, container, defect class and method: a person qualified on clear vials is not assumed qualified for amber syringes or difficult lyophilised cakes.

Automated inspection has a recipe and evidence state: cameras, lighting, handling, algorithms, thresholds, station assignments, reject mechanism, software and recipe versions, setup checks, alarms and maintenance. Each production population references the released recipe and configuration actually used, so a changed threshold or disabled station is not buried in machine logs. Qualification studies, setup challenges and routine challenge sequences stay separately identified, and a failed challenge identifies the time window and population inspected under an unacceptable state before routing reinspection or investigation.

4Observations keep unit and process context.

Each reject or classified observation links the unit or bounded population, defect code, location, image where available, inspector or station, timestamp, pass number and equipment context. Because the container’s filling and component genealogy stays queryable, a defect can be viewed by nozzle, stopper bowl, glass lot, intervention, shift, inspector, camera or time interval.

Approved reinspection defines its purpose, authorisation, population, method, independence, maximum passes and reconciliation. First-pass findings are preserved and included in trending, so repeated inspection cannot be used until a desired yield appears. Every move out of a rejected, questionable or accepted state has explicit authority and reason.

Representative rejects can go to confirmatory examination: microscopy, spectroscopy, dimensional measurement, leak testing or destructive evaluation, with chain of custody back to the original observation. The confirmed defect, material identity, origin hypothesis and uncertainty feed the batch investigation and the defect library rather than a disconnected laboratory report.

5Signals open bounded investigations.

A critical defect, an alert or action rate, a cluster, a novel defect, a challenge failure, a machine malfunction or an adverse trend triggers governed assessment. Scope can expand from one unit to an inspection interval, filling interval, component lot, equipment path, related batches or distributed product as evidence develops.

Trending uses comparable denominators: population, pass, method, recipe, container, line, station, supplier lot, shift, inspector and confirmed versus initial classification. Signals distinguish defects from affected units and unique units from repeat observations, and control limits stay versioned so a chart can be reproduced. AQL and other sampling plans complement the 100% inspection result without replacing it; Seal keeps the two distinct while assembling both for disposition.

6Reconciliation closes the population before disposition.

Filled, transferred, inspected, accepted, rejected, sampled, challenged, destroyed, retained and packaged quantities must balance by container state. An unexplained loss, duplicate identity, reject-bin discrepancy or failed reject verification is investigated, with written conclusions, before the batch can close.²

Batch review then assembles the inspection requirements, equipment and personnel eligibility, run conditions, challenge results, accepted and rejected populations, defect distribution, investigations, samples, reinspection, reconciliation and trend signals. Release, additional inspection, restricted use, rejection or recall assessment applies to an exact population and propagates to inventory.

Inspection equipment can remain the authoritative source for high-volume images and classifications. Seal resolves its outputs with manual work, defects, investigations, laboratory evidence, counts and release, so the record serves operations, quality, engineering and supplier quality, not only the person compiling the batch packet.

7Prove one difficult batch end to end.

Follow one sterile vial batch through container genealogy, eligible population, automated setup, challenge verification, 100% inspection, manual confirmation, defect classification, critical-defect escalation, reject examination, a bounded filling-interval investigation, reinspection, AQL sampling, reconciliation and disposition.

Include an unqualified manual inspector attempt, a failed reject challenge, a novel particle, a defect code changed after microscopy, repeat observations on one unit, a missing reject and a cluster linked to one stopper lot. The first usable release must account for every unit and preserve the evidence behind every changed classification.