cialistas.com

Vision-based inspection of automotive headlight lenses

This case follows a Melbourne lighting-components manufacturer that needed to inspect automotive headlight lenses more consistently than its manual quality team could manage. The product mix included clear polycarbonate and glass units for reflector, projector and LED lamp assemblies. Cosmetic defects such as scratches, orange peel, waviness, chips and seal marks were difficult to judge under ordinary workshop lighting, while order sizes ranged from small replacement batches to long production runs.

Dalian Optical Technology Co., Ltd. supplied an inspection cell that combines optical sensing, controlled illumination, lens-inspection equipment and optical surface-profile testing. The cell captures images, measures critical surface features, applies product-specific limits and records each result against the lens batch. In the four-week production trial, the team shortened inspection time, reduced cosmetic rework and improved traceability without changing the existing moulding or assembly equipment.

The production problem shaped by Australian conditions

Manual inspection had relied on trained operators viewing lenses on a rotating stand under mixed ambient light. Their pass or fail decisions were generally sound, but the method could not preserve magnification, viewing angle or lighting conditions between shifts. Fine scratches appeared only at certain polarities, while orange peel could resemble a coating defect. Rework was also difficult to analyse because photographs and measurements were stored separately from production records.

The Australian requirements made a generic optical bench unsuitable. Headlamp parts sold into the local market need appropriate vehicle-lighting compliance, including applicable Australian Design Rules and beam-pattern expectations, while consumer-facing aftermarket sellers must make clear guarantees under the Australian Consumer Law. The plant also handled programs for passenger cars, utes and SUVs, so one recipe had to recognise differences in clear coatings, reflector surfaces and protective films. Traceability mattered whenever a batch was returned by an Australian distributor.

Manual sampling offered little protection against a single escaped cosmetic defect moving into a complete lamp. To establish a baseline, engineers timed 240 inspections, reviewed the reason for every reject and compared results with an adjudicated set of reference lenses. A related vision deployment example offers a useful comparison for separating workflow design from image-processing capability.

Turning optical defects into measurable criteria

Inspection criteria were translated from drawings, customer tolerances and the defects most often found in incoming mouldings. The team created pass limits for scratch length, particle count, coating haze, orange peel and seal-edge continuity. Each product family received its own recipe, version number and representative golden sample. This prevented a strict limit from one lens geometry being applied incorrectly to another.

Surface-profile measurement was treated as a separate but complementary test. A 3D optical profiler quantified waviness and orange peel on a controlled patch, while the vision camera classified larger defects and contamination across the full optical area. Measurement resolution was set below the tightest customer tolerance, and operators recorded repeatability using the same reference points on each run. Readings that exceeded uncertainty limits were automatically flagged for review.

Recipe development also defined what the system should not do. Dust, condensation and protective-film wrinkles were separated from permanent surface damage, and acceptable manufacturing texture was documented with signed photographs. A practical system setup guide helped the engineering team structure lighting, focus and calibration checks before production data were accepted. These controls reduced repeated tuning during the first week of commissioning.

Building the automated inspection cell

The inspection cell used an industrial optical sensor, a low-distortion imaging lens, a segmented illumination arrangement and an encoder-linked rotating fixture. Dark-field illumination revealed fine surface scratches, diffuse lighting supported colour and haze assessment, and a dedicated profile station measured curvature-related defects. Stable fixtures held each lens at a repeatable distance and orientation, while encoded rotation exposed the complete surface without stopping for every position.

At the operator station, the lens batch and product recipe were selected before scanning. Software guided loading, checked focus and exposure, then moved through predefined fields of view. Confirmed pass parts received a label; rejected parts received an image set, defect code and routing instruction. The result was written to the plant network with the work order, shift, camera settings and calibration status, allowing quality staff to retrieve evidence rather than relying on memory.

Fixture design and cycle time were tested together because faster motion can hide low-contrast defects. Engineers established a conservative speed for each lens family and added dwell time where coatings required it. When procurement teams compare an equipment product range, they should verify resolution, working distance and software compatibility against the actual defect set. The trial also reserved expansion capacity for wider lens formats and future LED projector modules.

Controlling calibration and environmental drift

Calibration connected the camera-based classification with the optical profiler and the plant’s approved reference standards. A traceable artefact was measured at the start of commissioning and checked at planned intervals. The profile instrument was also verified against a known step or curvature standard, while golden lenses provided an application-level check of illumination, focus and classification. Any failed check placed the affected station into a controlled state rather than allowing unqualified data to continue.

Temperature and contamination were significant variables in the Australian installation. Melbourne’s rapid seasonal changes can move an optical bench through several temperature bands in one day, Sydney’s coastal humidity can encourage condensation, and dust is a persistent concern for components moved through inland and northern supply chains. The enclosure used filtered conditioned air, optical surfaces were cleaned on a fixed schedule, and readings were logged alongside room conditions so that drift could be investigated.

Daily verification remained simple enough for shift teams to sustain. Operators checked illuminator uniformity, focus, fixture seating and the reference artefact before the first batch, then repeated abbreviated checks after impacts or unusual temperature changes. Supplementary background can be reviewed through a quality reference, but acceptance decisions still depend on approved drawings, calibrated standards and the supplier’s signed test records.

Operational results and quality controls

The pilot processed 18,400 lens units from three product families over four weeks. Outputs were compared with a trained review panel, and any disagreement was resolved against the applicable drawing or customer standard. The machine’s performance remained stable after the initial recipe and lighting refinement, with no requirement to replace the line’s existing handling equipment. The measured changes were:

The largest benefit came from consistency rather than a single spectacular detection result. Engineers could compare a suspect lens with a golden sample at the same magnification and lighting angle, while supervisors could review the evidence behind a rejection without returning to the inspection station. The controls carried into production were:

Operationally, the cell did not remove human judgement; it directed attention more consistently. Operators still handled unusual shapes, visible handling damage and customer-specific waivers, but routine pass decisions were automated. This reduced fatigue, shortened queues at the quality desk and created a searchable record of every optical anomaly. The same machine-vision monitoring data also helped process engineers compare moulding tools and suppliers when a recurring defect pattern appeared across multiple batches.

Scaling the system and securing supplier support

After the trial, the manufacturer standardised the cell recipe across two Melbourne lines and prepared a second installation in Adelaide, using the same defect library with product-specific limits. Spare lenses, illumination modules, encoder components and profiler standards were held locally, reducing dependence on overnight courier deliveries. A Sydney service partner was included in training so that regional downtime could be addressed without waiting for an overseas engineer.

Equipment supplied from Dalian, Liaoning required customs planning that the project team had started early. A broker confirmed the tariff classification for each optical and machine-vision item rather than assuming every component qualified for identical treatment. Commercial invoices, packing lists, 10% import GST, freight, insurance and handling costs were reconciled before installation. Serial numbers and calibration certificates were captured on arrival, separating imported components from locally sourced fixtures throughout maintenance.

The public Dalian website currently contains unrelated Chinese adult-content spam in its title and headings. That contamination does not by itself indicate a problem with an optical instrument, but it weakens the website as a source of technical assurance. Buyers should independently verify the company’s legal identity, equipment serial numbers, calibration certificates and test reports, and should not rely on promotional metadata as proof of supplier status. With those controls in place, the Australian operation gained a repeatable inspection method that could scale with local automotive demand.