EMUG Tech delivers seven computer vision service areas: surface defect detection using CNN and YOLO v8 models for scratches, dents, porosity, and coating voids; dimensional inspection and measurement using 3D structured light and stereo vision; assembly verification and presence detection for missing components and incorrect assembly; weld quality inspection using visual, thermal, and X-ray imaging; remote and drone visual inspection for infrastructure and energy assets; safety monitoring and PPE compliance detection; and custom computer vision solutions for non-standard inspection requirements. All programmes follow the EMUG FOCUS Framework — five phases covering Frame, Obtain, Configure, Unite, and Sustain.

EMUG FOCUS is the five-phase computer vision delivery methodology: Frame — vision use-case definition, hardware specification, and production line integration design; Obtain — image dataset collection, defect labelling, and data pipeline setup; Configure — model architecture selection, training, and acceptance criteria validation; Unite — edge hardware deployment, SAP QM and MES integration, and production go-live; Sustain — model performance monitoring, retraining from production feedback, and programme expansion. FOCUS ensures vision AI is validated against production acceptance criteria and integrated with enterprise quality systems before go-live — addressing the most common failure mode of vision AI programmes: impressive demo accuracy that degrades in real production lighting and throughput conditions.

Training dataset requirements for manufacturing defect detection typically range from 500 to 5,000 labelled images per defect class, depending on defect visual complexity, required detection accuracy, and variability in production conditions. EMUG FOCUS Obtain phase covers structured image acquisition protocol to capture the full range of production lighting, camera angle, and surface condition variation; annotation and labelling workflow using CVAT or Roboflow; data augmentation techniques (rotation, brightness, blur, noise) to extend limited defect datasets; and transfer learning from pre-trained industrial vision models to reduce the labelled data requirement for common defect types. For programmes with very limited defect history, EMUG recommends a structured data collection phase before model development begins.

Production line speed deployment requires three components: camera and lighting hardware designed for the production line throughput and field of view; edge inference hardware (NVIDIA Jetson Orin, Jetson AGX Xavier) positioned at the production station for sub-100ms inference latency without reliance on network connectivity; and model optimisation using TensorRT, quantisation, and pruning to fit the model within edge hardware latency and memory constraints. EMUG FOCUS Frame phase designs the complete hardware architecture — camera specification, lighting design, edge hardware selection, and mechanical integration — before model development begins, ensuring the vision system is designed for production latency requirements from the start rather than retrofitting an accurate but slow model onto a fast production line.

EMUG FOCUS Unite phase integrates computer vision with SAP QM and MES: every defect detected by the vision system automatically creates a SAP QM non-conformance notification with defect classification, location coordinates, confidence score, production lot, and inspection image attached as a quality record. The SAP QM integration uses standard BAPIs or the SAP Quality Management API. PLC pass-fail signals prevent defective parts from advancing to the next production station. MES production dashboard integration provides real-time yield visibility. For IATF 16949 programmes, the integration creates a complete digital traceability chain from inspection image through SAP QM defect record to corrective action workflow.

IATF 16949 requires measurement system analysis (MSA) for inspection equipment, including automated vision systems. EMUG FOCUS validates vision inspection using gauge R&R equivalent methodology: repeatability testing (same part, same conditions, multiple inspection passes), reproducibility testing (same part, different production conditions, lighting variations, camera positions), and bias assessment against reference standards. Validation is performed using a statistically significant sample of known-good and known-defect parts across the full range of production conditions. Documentation covers system specifications, validation methodology, results, and acceptance criteria sign-off — meeting IATF 16949 MSA requirements for inspection equipment qualification.

A focused single-use-case vision deployment — for example, surface defect detection on one production line — typically reaches production go-live in 10 to 16 weeks from Frame phase start to Unite phase completion. The timeline breakdown is typically: Frame and Obtain (3 to 4 weeks for hardware specification, dataset collection, and labelling), Configure (3 to 5 weeks for model training, optimisation, and validation), Unite (3 to 5 weeks for hardware installation, system integration, and go-live). Complex deployments involving multiple defect classes, multiple camera positions, 3D measurement, or SAP QM integration with custom workflows may take 16 to 24 weeks. Data collection is the most common schedule constraint for programmes with limited defect image history.

Computer vision programmes are delivered across automotive OEMs and Tier 1 suppliers, aerospace and defense manufacturers, industrial machinery producers, energy and process industries, and high-tech electronics firms in 20 countries: Germany, France, UK, Netherlands, Sweden, Italy, Spain, Poland, Czech Republic in Europe; India, Japan, South Korea, China, Malaysia, Thailand in Asia-Pacific; UAE and Saudi Arabia in the Middle East; USA, Canada, Mexico, Brazil in the Americas. Computer vision delivery from Hyderabad, Germany, and Dubai with on-site hardware commissioning, camera installation, and production line integration capability at client manufacturing facilities globally.