Materials analysis software QAI

Overview
QAI is an AI-based image evaluation module integrated into QpixControl2 for automated indentation detection and evaluation in hardness testing. It supports Vickers, Knoop and Brinell methods and is optimized for challenging, low-contrast, etched or rough surfaces. The AI runs locally within QpixControl2 and requires no internet connection.

Key benefits

• Automatic, high-accuracy detection of hardness test indentations (Vickers, Knoop, Brinell).
• Higher recognition and hit rates compared with classic image recognition algorithms.
• Reduced need for manual intervention and verification.
• Improved repeatability and reduced systematic deviation of measurement results.
• Consistent results for identical indentation images.

Materials and surface types
QAI increases detection performance on difficult surfaces. It is particularly effective on rough, ground, etched and otherwise low-contrast or irregular surfaces where classical algorithms often fail.

Examples (selected test images and conditions)

• Low contrast on steel — Hardness: 725 HV1; Preparation: grinded P1200 / polished 1 µm.
• Low contrast on etched steel — Hardness: 309 HV0.5; Preparation: grinded P1200 / polished 1 µm.
• Low contrast on etched carbon steel — Hardness: 121 HV1; Preparation: polished 1 µm.
• Low contrast on etched construction steel — Hardness: 235 HV0.5; Preparation: grinded P1200 / polished 1 µm.
• Etched steel — Hardness: 305 HV0.5; Preparation: polished P1200 / polished 1 µm.
• Low contrast on etched steel — Hardness: 837 HV0.5; Preparation: grinded P1200 / polished 1 µm.
• Large deformation/bulging on steel — Hardness: 263 HV10; Preparation: polished 1 µm.
• Small indentation on cast iron — Hardness: 361 HV0.01; Preparation: polished 1 µm.
• Rough surface on steel — Hardness: 287 HV10; Preparation: grinded P80.

Evaluation & performance comparison
QAI replaces the classic image recognition in QpixControl2 and provides measurable improvements. Representative test results for 90 hardness points on a test block (target HV1 ~701):

Classic Evaluation — summary table
Mean value | Range
700.04 | 24.90

Hardness min. | Hardness max.
688.80 | 713.70

Standard deviation | Results OK
5.88 | 90

QAI Evaluation — summary table
Mean value | Range
701.55 | 16.40

Hardness min. | Hardness max.
692.50 | 708.90

Standard deviation | Results OK
3.47 | 90

Data handling and operation
All QAI processing and AI inference run locally within the QpixControl2 environment on the PC; no cloud or internet connection is required. The model does not self-train on the instrument. Model updates and retraining are performed and validated by QATM to maintain compliance with hardness testing standards.

Characteristics / technical specifications

• Commercial name: QAI
• Integration: Fully integrated into QpixControl2 operating software
• Supported measurement methods: Vickers, Knoop, Brinell
• Operation mode: 100% local / offline inference within QpixControl2 (no cloud required)
• Model updates & retraining: Managed by QATM (no autonomous learning on the device)
• Primary benefits: increased recognition/hit rate, higher repeatability, lower standard deviation vs classic evaluation
• Observed improvements: reduction in standard deviation (classic ~5.88 → QAI ~3.47) and narrower result range in tests
• Suited for: etched, grinded, polished, rough and low-contrast surfaces
• Image types: supports typical optical images produced by Qpix systems (camera/lens/magnification unchanged)