Photonics and fiber optic manufacturing

In production it covers the assembly and surface checks that decide whether an optical component performs to spec: alignment of transmit and receive sub-assemblies in a transceiver, epoxy distribution and tilt where a fiber array is bonded, surface roughness on a waveguide that would show up as scattering loss, and fractures, delamination or micro-voids in fiber ribbon. These are visual defects with optical consequences, which is what makes them worth catching before the part is packaged and tested.

No, and the distinction matters when you are choosing. A fiber inspection scope is a handheld or benchtop microscope a technician points at a connector end face to check it in the field or at a workstation, one connector at a time. This is fixed in-line inspection: a camera mounted over the process, triggered by the line, deciding on every part and signalling the result to the PLC. If you need a portable scope for field work, that is a different product category and we do not make one.

Because the defects vary in ways a programmed rule cannot enumerate. Epoxy does not spread the same way twice, a fracture in glass has no fixed shape, and surface roughness is a texture rather than a measurement against an edge. Rule-based tools want highly repeatable features and precise thresholds, which is why they struggle here. A model trained on examples of acceptable and unacceptable parts learns the range instead, which is what makes these defects practical to automate rather than theoretical.

With lighting geometry rather than more light. Glass is both transparent and specular, so a single diffuse or single direct source produces hotspots that hide the defect and reflections that look like one. Multi-angle capture, sometimes called 2.5D, illuminates from several directions and combines the frames so the system reads surface slope instead of raw brightness. Dark field is often the right answer for scratches and surface roughness, because a smooth surface stays dark and only the defect scatters light into the lens.

Often yes, if you are measuring rather than only detecting. A fiber array is a set of features that must be compared across the width of the frame, and a standard lens views the outer positions slightly from the side, so pitch and position pick up an error that grows toward the edges. Telecentric optics remove that perspective entirely. If the check is simply whether epoxy has voids or a fiber is missing, a standard lens with the right lighting is usually sufficient and considerably cheaper.

Less than most teams expect. A clear-cut classification often trains usefully on 8 to 20 examples of each class. Where failures are rare by design, which is typical in photonics, anomaly detection trains on good parts only. For a specific rare defect that has not occurred yet, synthetic defect generation places realistic examples onto images of your own parts so the model can be trained and validated without waiting for scrap.