senswork GmbH will demonstrate AI-assisted surface defect recognition and high-resolution 3D quality-control systems at the International Manufacturing Technology Show (IMTS) 2026 in Chicago. The German instrumentation supplier will occupy booth space at the September show to present two complementary inspection technologies aimed at automated production environments.

What is senswork bringing to IMTS 2026?

The company's AI-powered surface inspection portfolio combines 2D imaging with trained neural-network classifiers to detect cosmetic and functional defects on machined, cast, and additively manufactured parts. In parallel, the firm will show 3D measurement systems that deliver sub-millimetre geometric reconstruction of complex components for in-line dimensional verification.

The pairing targets the same end-user problem: qualifying parts without removing them from the production cell. Production engineers who currently rely on manual visual inspection or coordinate measuring machine (CMM) batch sampling face throughput limits once defect frequency rises above a few parts per thousand.

How does the inspection physics work?

2D surface inspection relies on photometric contrast: scratches, pores, inclusions, and weld spatter produce local intensity gradients under controlled illumination. Traditional rule-based algorithms compare pixel statistics against fixed thresholds. Neural-network classifiers instead learn feature hierarchies from labelled training datasets, letting them generalise across surface textures and lighting variations that defeat thresholding.

The underlying physics does not change, but the decision boundary does. A convolutional network trained on tens of thousands of annotated images can resolve sub-100-micrometre defects on reflective metal surfaces under diffuse LED lighting, the typical shop-floor configuration. Vendors do not always publish the training corpus size, the false-accept rate at a defined defect size, or the minimum contrast ratio. Buyers should request those numbers under controlled demo conditions before committing capital.

Which standards govern 3D quality control?

3D quality control on the shop floor usually relies on one of three sensor families: laser line triangulation, structured-light projection, or photometric stereo. Each reports point clouds with quoted resolution and accuracy, but the meaningful number is volumetric accuracy under production conditions, where temperature drift, vibration, and ambient light all degrade the headline specification.

ISO 10360 governs acceptance and reverification tests for coordinate measuring machines, while VDI/VDE 2634 covers optical 3D systems. The relevant clause for in-line systems is the capability test against an artefact of known geometry, repeated at production temperatures rather than the 20 °C laboratory reference. Vendors who quote only 20 °C accuracy leave the buyer to extrapolate to factory conditions, often a 10 to 30 percent penalty.

What separates vendor claims from measured performance?

Datasheet headline numbers typically cite best-case conditions: stationary target, controlled temperature, single-shot measurement. Production environments introduce conveyor vibration, part-to-part positional variation, and thermal gradients that erode those numbers. The standard remedy is a periodic calibration interval — daily, weekly, or per-shift — using a calibrated artefact traced to a national metrology institute.

Buyers evaluating AI-based inspection should request:

• False-accept and false-reject rates at a defined defect size
• Training dataset composition and refresh policy
• Inference latency at full production throughput
• Calibration interval under documented production conditions

These figures rarely appear in glossy product literature and require direct technical questioning.

What adoption question does this raise?

The development raises the deployment question that always follows a successful trade-show demo: can the system maintain its metrological performance across shifts, materials, and lighting changes that characterise real manufacturing? A system that holds 95 percent detection accuracy on a controlled line may drop into the high 80s once ambient light, coolant splash, and operator-induced part variation enter the picture. The answer depends less on the algorithm than on the integration discipline the end user applies during ramp-up.