Orthopedic prostheses are among the most demanding quality inspection targets in manufacturing. When the part is surgical-grade stainless steel polished to a mirror finish, every conventional imaging approach fails. Intelgic engineered a purpose-built AOI system — custom optics, custom lighting geometry, motorized multi-angle imaging, and three months of dedicated AI fine-tuning — to achieve inspection accuracy that meets the zero-defect standard this application demands.
The Challenge
The manufacturer produces artificial knee prostheses — femoral components, tibial trays, and associated implant hardware — in surgical-grade stainless steel with a mirror-polished surface finish. Each part must meet strict dimensional and surface quality standards before being cleared for distribution. A surface scratch, machining pit, burr, or coating anomaly that would be trivial on an industrial part is a potential patient safety issue on a load-bearing joint implant.
Their existing process relied on manual visual inspection by trained quality technicians under controlled lighting. While diligent, the process suffered from the fundamental ceiling of human inspection: fatigue, subjectivity, inconsistency between inspectors, and an inability to maintain uniform scrutiny across high-volume production runs. There was also no systematic documentation linking each part to its inspection record — a growing liability in a regulated industry.
- Mirror-polished stainless steel surface that defeats all direct lighting approaches with specular glare
- Complex three-dimensional geometry — condylar curves, chamfers, articular surfaces — creating inspection blind spots
- Multiple product sizes requiring changeover without loss of inspection accuracy
- Manual inspection inconsistency and fatigue on long production shifts
- No per-part inspection traceability for regulatory documentation
- Defects near the specification limit — too fine for untrained human detection
- Automated defect detection on mirror-polished metal surfaces without specular imaging artifacts
- Full coverage of complex 3D prosthesis geometry in a single inspection cycle
- Multi-recipe support for all product sizes — one system, no recalibration per changeover
- Single button press operator interface — minimal training required
- Timestamped, per-part inspection records for full production traceability
- Detection accuracy sufficient for medical device quality programs
Why Polished Stainless Steel Defeats Conventional AOI
Standard automated optical inspection systems direct structured light at a part and capture the reflected image. On matte or low-gloss surfaces, this works reliably. On a mirror-polished surgical implant, it fails completely. The surface acts as a perfect mirror — reflecting the camera housing, the light source itself, and the enclosure walls directly into the lens as overwhelming bright artifacts. Actual defects, which have low contrast against the uniformly reflective background, become invisible. The result is an image full of false positives from reflections and false negatives for the defects that actually matter. Solving this required a fundamentally different optical architecture.
The Solution: Custom AOI Inspection System
Intelgic designed and built the inspection system from first principles — treating it as an optical engineering problem before it was a software or AI problem. Hardware, lighting, mechanics, and AI were co-designed as a single integrated system, each element informing the others.
Lighting Geometry — Defeating Reflectivity
The foundation of the system is a custom lighting architecture specifically engineered to neutralize specular reflection on high-gloss metallic surfaces. Three complementary illumination modes work in sequence during each inspection cycle.
Diffuse Dome Illumination
A custom-built dome floods the part with soft, omnidirectional light from all angles simultaneously — eliminating all specular hotspots and making the mirror surface inspectable.
Dark-Field Low-Angle
Grazing-angle illumination causes surface topographic features — scratches, pits, burrs — to scatter light brightly against a dark background, making fine defects visible against the polished surface.
Sequenced Multi-Mode
The system captures images under each lighting condition per inspection position. The AI receives a complementary image set that reveals different defect types — no single condition can mask another.
Motorized Multi-Angle Imaging
The complex three-dimensional geometry of femoral components and tibial trays — condylar curves, posterior chamfers, articular surfaces, fixation features — cannot be fully inspected from a single camera position. Intelgic designed a precision motorized stage that indexes the part through a programmed sequence of positions and rotation angles, with the camera capturing the full image set at each station.
This eliminates blind spots entirely. Every surface zone the specification requires to be inspected is imaged from the optimal angle and under the optimal lighting condition for that zone. The complete motion sequence, lighting switching, and image acquisition run automatically after the operator presses start — no operator action is required during the cycle.
Multi-Recipe Architecture
The manufacturer produces the same implant design across a range of patient sizes — each with different dimensions and slightly different surface geometry. Intelgic built the system around a configurable recipe architecture: each product type and size has a stored inspection program containing the full motion sequence, lighting configuration per station, camera parameters, AI model weights, and acceptance thresholds.
Switching between product sizes is a single operator selection on the touchscreen. The system reconfigures itself completely — no physical setup, no calibration, no delay. This protects the hardware investment as the manufacturer adds new implant sizes and variants over time.
"The lighting geometry was the breakthrough. Once we solved reflectivity, the AI had clean, consistent image data to learn from — and it learned with remarkable speed."
Inspection Workflow
Operator Loads the Part & Selects Recipe
The operator places the prosthesis into the precision fixture on the motorized stage, then selects the product recipe from the touchscreen — a single step that loads the complete inspection program for that implant size and type.
Single Button Press Initiates the Full Automated Cycle
One button press begins the inspection. From this point, the system operates fully autonomously — the operator does not need to intervene at any stage of the cycle.
Motorized Stage Indexes Through Inspection Positions
The precision mechanical stage moves and rotates the part through a programmed multi-position sequence. At each station, the lighting system cycles through its configured illumination modes while the high-resolution camera captures a complete image set under each condition.
Intelgic AI Analyzes Every Image
As images are captured, Intelgic's fine-tuned vision AI models analyze each frame in real time — classifying regions, identifying anomalies, and cross-referencing findings across complementary lighting conditions to distinguish genuine surface defects from optical artifacts or reflections.
Defect Classification, Location Mapping & Pass/Fail Decision
Every detected anomaly is classified by defect type, measured for size and severity, and mapped to its precise location on the part geometry. The system renders an annotated defect map and issues a Pass or Fail decision against the recipe acceptance criteria.
Complete Inspection Record Logged Automatically
Every inspection generates a timestamped record — part ID, recipe used, full image set, defect classifications, location data, severity scores, and pass/fail result — providing the per-part traceability required under FDA quality system regulations.
The AI Fine-Tuning Program
Stainless steel prosthesis surfaces present AI with an unusually demanding visual dataset. Complex curved geometry produces lighting variations between parts of the same design. Deliberate surface textures in fixation zones must be distinguished from defects. And the population of genuine defects is small and subtle — sitting near the boundary of the acceptance specification.
Intelgic ran a structured three-month AI fine-tuning program in parallel with production. The system inspected actual production parts, and AI models were iteratively trained against expert-graded ground truth data. This approach produced a model robust to real production variability in a way no pre-trained general model could achieve.
AI Accuracy Development — 3-Month Fine-Tuning Program
Defect Types Detected
The system is trained and validated across the full spectrum of surface defects relevant to surgical-grade orthopedic implant manufacturing:
Surface Scratches & Abrasion Marks
Machining Pits & Micro-Voids
Edge Chips & Micro-Fractures
Burrs at Machined Transitions
Surface Contamination & Staining
Coating Adhesion Failures
Results
Following deployment and completion of the three-month AI fine-tuning program, the Intelgic AOI system became the primary quality gate for all finished orthopedic implant production. The results across detection accuracy, traceability, and operational consistency exceeded the manufacturer's requirements.
The facility moved from sampling-based manual inspection to 100% automated inspection — every prosthesis receiving equal scrutiny regardless of production volume or shift. Detection accuracy reached 98.6%, validated against expert-graded ground truth. Zero defective parts have been shipped to distribution since deployment.
The per-part inspection record generated for every unit resolved a longstanding traceability gap in the manufacturer's quality system. Documented proof of inspection, complete with defect maps and AI classification data, is now automatically available for every part — eliminating the manual documentation burden and providing a robust audit trail for FDA quality system compliance under 21 CFR Part 820.
The multi-recipe architecture has proven its value as new implant sizes have been added to the production line. Each new recipe is developed, validated, and deployed to the existing hardware — extending the system's productive life without additional capital expenditure.
"Purpose-built optics, a motorized imaging stage, and AI trained specifically on surgical stainless steel. There are no shortcuts to this level of inspection accuracy — and no substitute for it."
