KOMPASS AI brings foreign object detection to exposed product surfaces and features visible through suitable transparent materials. Check unwanted particles, residue and inclusions against your quality criteria, then flag affected units before assembly, sealing or packing makes the area harder to inspect.
Foreign object detection is most useful when the unwanted material is still visible. A fragment on an exposed part may be easy to isolate now, then become hidden inside an assembly or package a few seconds later.
Our AI Product Inspection system, KOMPASS™, checks captured images for unwanted visible material and agreed inclusion defects. Modular vision hardware provides the views and lighting needed for the product, while the trained model separates suspect features from acceptable appearance.
The inspection starts with a usable view of the target material. Define the visible defect, the acceptance rule and the response while the unit can still be isolated.
Choose the checkpoint before the relevant surface is covered. Position the camera and lighting to make the target material visible against the product, including difficult low-contrast examples.
Train with actual contamination examples and normal product features. The model checks the visible area against the agreed defect categories and acceptance rules.
Pass the result to the configured rejection or review response. Keep the defect image so quality teams can investigate repeat findings and compare them with earlier inspected units.
Show us the unwanted material you need to find and the product it appears on. We will assess the viewing conditions and the right checkpoint for inspection.
Get In TouchAll camera-based methods need a usable view. The inspection design determines what is exposed, and the decision method determines how acceptable variation is handled.
Shriji's published application identifies contamination, sealing damage and surface irregularities among its inspection challenges. Jidoka deployed KOMPASS on the wad assembly line to classify defects and connect the result to automated rejection and quality reporting.
These results cover the bottle-cap inspection application as a whole. They do not establish a detection rate for every contaminant or material.
Foreign material detection depends on whether the unwanted feature can be distinguished in the captured image. Particles, fragments, fibres or residue may be suitable targets when the product, background and lighting provide enough visible information. Foreign particle detection should be tested using the actual material, its likely positions and the smallest rejectable examples. Define the inspection around those conditions. A material name alone cannot establish whether the system will see it reliably on your product.
Inclusion detection is possible only where the chosen imaging method exposes the feature. A particle visible through clear material presents a different task from one buried inside an opaque component. Surface inclusion detection can assess a visible mark, but cannot establish the full extent or composition of material hidden below it. Share cross-sections, reference findings or other available evidence during feasibility work so the team can identify which requirements need an additional inspection method.
No. Foreign material inspection should match each risk to a method capable of detecting it. Cameras assess appearance within the available view. Other technologies can address particular conductive materials, density differences or concealed features under suitable conditions. Visual contamination detection can add a useful checkpoint where unwanted material is exposed, but its contribution needs validation alongside the rest of your inspection process. Keep the acceptance criteria and response for each method clearly defined.
Train with representative good parts as well as real reject examples. Include printing, mould texture, normal colour variation and the changing appearance of the surface under production conditions. Black spot detection, for example, needs a definition of which dark features are unacceptable and which belong to the product. Test borderline cases separately, then review escapes and false rejects together. This keeps a sensitive inspection from becoming an unnecessary source of scrap or repeated manual checking.
Plastic contamination detection is harder when the fragment has little contrast against the product. Transparency, gloss, thickness, orientation and the background can all change what appears in the image. Embedded particle detection adds the difficulty of the surrounding material obscuring the target. Feasibility needs physical samples under the proposed lighting and camera views. If the required information is absent from those images, changing the model alone will not make the target reliably detectable.
Put it where the relevant area is accessible and a suspect unit can still be isolated. For visible contamination, that may be before a cover is fitted, a package is closed or the product moves into another operation. Consider whether overlapping parts, fixtures or product movement obscure the view. Also allow enough distance and time between capture and the configured response. The right location should reflect both where the risk appears and how the line will act on a finding.
The inspection result must remain associated with the correct unit as it travels to the response point. Optical sorting and other separation arrangements need suitable presentation, tracking and timing. Agree whether the line should reject the unit, hold it for review or take another defined action. Check the complete sequence at the intended production rate, including closely spaced parts and changeovers. A correct image classification is useful only when the right unit receives the right response.
No. Surface contamination inspection establishes a result for the visible features and conditions included in validation. It does not determine chemical composition, detect every microscopic or biological hazard, or prove that concealed areas are clear. Keep that scope explicit in your quality process. For a trial, provide representative samples, a defect acceptance standard, the required views and the response you need on the line. Those inputs establish what this visual checkpoint can contribute.
Share the contaminant, the product surface and the point where your team needs a decision. We will assess visibility, coverage and the response needed to isolate an affected unit.
