Automated visual inspection for injection molded parts should be specified as a part-specific measurement and decision system, not simply as a camera on the line. Buyers in Europe, North America, and the Middle East should first decide which visible conditions matter to product function, regulatory obligations, assembly, branding, or customer perception. A system may be suitable for confirming a clip is present, a mark is readable, or a flash-free edge is within an agreed visual limit, yet be unsuitable for judging a subtle gloss change unless the part, light, and acceptance standard have been developed together.
The most useful specification connects a controlled part presentation, optics and lighting, an approved defect library, a pass/fail rule, and a response when the rule is triggered. It also distinguishes automated visual inspection from dimensional metrology: a camera may screen an edge profile or a visible feature, but it should not be assumed to replace a validated gauge or CMM. This framework helps buyers request proportionate injection molding quality control evidence without implying that every molded part can receive the same detection capability.
Specify the defect before specifying the camera. Define the surface zone, allowable condition, viewing distance or functional consequence, part orientation, and the reference samples that represent accept, reject, and borderline conditions. Then require the supplier or integrator to demonstrate performance on those conditions in the intended production setup.
01
Start with a part-specific inspection charter
Begin with a drawing or controlled visual standard that divides the component into zones. For each zone, state the feature or defect to be checked, such as short shot, flash, gate vestige, sink, weld line, contamination, color mismatch, insert presence, logo legibility, or assembly orientation. Identify whether the check is cosmetic, functional, safety-related, or traceability-related. The same mark can be acceptable on a concealed surface and unacceptable on a customer-facing surface, so a single statement such as “no defects” is not a workable machine-vision requirement.
Turn subjective language into observable rules. Instead of “no visible scratch,” define the inspection face, whether a scratch must be detected by contrast, length, width, interruption of texture, or another agreed characteristic, and how approved comparators will be used. For an injection molding CTQ, record the datum or view used, its limit or attribute decision, and the consequence of a fail. Do not ask a visual system to certify material grade, internal voids, or structural performance.
State the operating context as well: resin and color family, texture or gloss level, cavity identification if used, cycle-rate range, expected part temperature at inspection, and whether parts arrive singly, nested, or on a fixture. Part presentation should be included in the specification, not treated as an afterthought.
02
Match illumination, optics, and handling to the defect mechanism
A camera records reflected or transmitted light, rather than the buyer’s general impression of quality. Lighting geometry should be selected after review of representative production parts. Dark-field or low-angle illumination can make some scratches, raised flash, and texture disturbances more visible. Diffuse illumination can reduce glare on glossy surfaces, while backlighting can support silhouette, edge, or presence checks. A feasibility study should show accepted and rejected samples in the proposed setup [1].
Specify field of view, camera views, resolution at the target feature, lens working distance, exposure strategy, and the acceptable variation in part location and rotation. Where the part has hidden faces, deep ribs, transparent regions, metallic inserts, or highly reflective surfaces, request an explicit coverage map. If no single view reveals the condition reliably, the solution may need multiple cameras, controlled rotation, or a different inspection method. “100% visual inspection” should mean every defined inspection opportunity is presented to the approved vision sequence; it should not imply that all possible defects are detectable.
For robotic part handling injection molding, define the handoff: pick location, permitted gripper contact areas, fixture nests, orientation verification, reject route, and what happens when a part cannot be located. The system should distinguish “not inspectable” from “rejected for a product defect.” Image systems are assemblies of camera, lighting, optics, processing, and line controls; their effectiveness depends on those components and the production environment, not on software alone [2].
03
Define acceptance logic and validate it with challenge samples
The inspection specification should name the decision categories: pass, fail, review or hold, and equipment fault. Define the output required for each category, including a PLC signal, reject confirmation, operator alert, and segregation route. Agree how re-inspection is controlled, who may override a result, and whether a failed image is retained. For multi-cavity tooling, decide whether results must be linked to cavity, machine, lot, shift, or time window. That linkage can help isolate a recurring process change, but buyers should request only the traceability that their product and risk assessment require.
Before routine release, ask for a documented trial using a mutually agreed sample set. Include known good parts across normal appearance variation, known reject parts for each defect class, and borderline conditions that test the decision boundary. Test the intended line speed, orientations, and reasonably expected process conditions. Record false accepts, false rejects, unreadable or uninspected parts, and any manual adjudication. A peer-reviewed example of in-process machine-vision validation notes that measurement-system analysis separates process variation from measurement-system variation and considers accuracy, precision, and stability [3]. The appropriate study design and acceptance target remain program-specific.
AI or anomaly-detection tools may assist with complex appearance patterns, but they do not remove the need for labelled reference samples, version control, and change approval. Require the supplier to identify the algorithm or rule version, training-data ownership and retention where relevant, retraining triggers, and the method for preventing an unapproved update from changing production decisions. If a new resin, colorant, texture, camera, light, fixture, or mold repair materially changes the image, re-evaluate the affected inspection rule.
04
Request usable evidence, maintenance controls, and clear commercial boundaries
A practical approval package may include the inspection charter, coverage map, approved images, defect catalogue, validation protocol and results, system settings under change control, reject-flow test, and a short operator or maintenance instruction. Request examples of the report format: part or batch identifier, result, defect category, image reference where retained, exception reason, and reviewer disposition. Data retention periods, access rights, export format, and cyber-security responsibilities should be agreed commercially, especially where inspection data must cross borders or enter a customer quality system.
Define checks for camera focus, lens cleanliness, light output, fixture wear, image trigger, reject actuator, and time synchronization with the molding cell. A daily or shift-start challenge part can confirm that the controlled setup still recognizes a known condition, but its frequency and content should follow the production risk rather than a universal rule. EMVA 1288 provides a common framework for specifying and measuring machine-vision cameras and sensors, making component comparison more transparent; it does not validate the complete inspection cell or a particular defect-detection claim [4].
Finally, make system ownership and responsibility unambiguous. Identify who supplies the vision hardware, who owns source configurations and image data, who supports remote access, who approves parameter changes, and who investigates an escape or recurring false reject. Scope the cost, schedule, spare parts, service response, and any site acceptance separately in the quotation or quality agreement. Those commitments depend on the chosen architecture, risk, location, and service model and should never be inferred from a general inspection description.
Practical takeaway
Before you move
to the next review.
- Provide controlled drawings and a zone-based cosmetic or functional acceptance standard for every feature the system must judge.
- Supply representative accepted, rejected, and borderline parts, including normal resin, color, texture, and cavity variation where relevant.
- Specify inspection coverage, part orientation, line-speed range, handling interface, and the treatment of uninspectable parts.
- Agree pass, fail, hold, and equipment-fault logic; verify reject confirmation and physical segregation during acceptance testing.
- Request version-controlled settings, validation records, sample images, maintenance checks, data rules, and a change-control process.
FAQ
Common questions
Can automated visual inspection replace manual inspection for molded parts?
It can reduce or replace defined repetitive visual checks when the defect, presentation, and pass/fail criteria are demonstrated on the production setup. It does not automatically replace human review for subjective appearance decisions, unseen surfaces, or conditions the imaging method cannot reveal. Keep a controlled review route for exceptions and periodically confirm the system against approved samples.
What defects can a vision system inspect on injection molded parts?
Potential checks include presence or absence, orientation, readable marks, color or contrast differences, visible flash, edge chips, obvious contamination, surface disturbances, and selected outline features. Detectability depends on contrast, accessible views, part movement, lighting, resin, texture, and the agreed defect definition. A feasibility study with real samples should determine scope rather than a generic defect list.
How should buyers validate an automated visual inspection cell?
Approve a written protocol and test good, bad, and borderline samples at intended operating conditions. Compare the system result with the agreed reference disposition, record false accepts, false rejects, and no-read events, and verify the reject path. Document the final configuration and require change review if material, appearance, tool, lighting, handling, or software changes could affect results.
Does 100% automated inspection guarantee zero defective shipments?
No. It means every presented part receives the defined inspection sequence. Performance still depends on coverage, sampling of challenge conditions, stable presentation, equipment health, and the limits of the chosen optical method. A buyer should define residual risk, escalation, and containment actions rather than treat the phrase as a guarantee.
References
Source material
- [1] KEYENCE — Machine Vision Lighting Selection Guide
- [2] KEYENCE — Visual Quality Inspection: A Complete Guide
- [3] Guha et al. — Application and validation of machine vision inspection for efficient in-process monitoring of complex biomechanical device manufacturing
- [4] European Machine Vision Association — Introduction to Machine Vision Standards
This guide is general information for early procurement and engineering discussion. It does not replace product-specific DFM, qualification, validation, regulatory assessment or legal advice.