Parts can be differentiated by measurements such as length, width, height, diameter, thickness, hole diameter, distance between holes, slot dimensions, feature spacing, and profile measurements.

For flat parts, calibrated 2D imaging may be sufficient. When height, depth, or full geometry is important, a 3D camera or laser-profile sensor may be required. Telecentric optics may be used in precision applications to reduce perspective error and improve measurement consistency.

Some parts cannot be differentiated by geometry alone. Visual features may include:

Intelgic can combine object-recognition AI with OCR and code-reading technologies when product identity depends on markings.

Three-dimensional imaging captures depth, height, and surface profile. It is valuable for identifying:

A 3D point cloud can help the system understand both part identity and picking orientation.

Parts are supplied to the inspection and picking area through conveyors, bins, trays, pallets, chutes, vibratory feeders, tables, racks, or returnable containers. The presentation method affects visibility, picking access, and system cycle time.

Parts spread across a conveyor may be easier to image and pick than deeply entangled components inside a container. Intelgic evaluates the complete material flow before selecting the automation concept.

Industrial cameras capture the images required for recognition. Depending on the application, the system may use area-scan cameras, high-resolution cameras, color cameras, monochrome cameras, stereo-vision cameras, 3D time-of-flight cameras, structured-light sensors, laser-profile sensors, robot-mounted cameras, fixed overhead cameras, or specialized illumination.

Lighting is engineered to make relevant differences visible while reducing glare, shadow, and environmental variation.

Intelgic's AI analyzes the captured image or 3D data to locate and classify every visible part. The system can use object detection, image classification, instance segmentation, shape matching, anomaly detection, OCR, barcode and QR-code reading, dimensional measurement, and 3D point-cloud analysis.

For mixed components, instance segmentation can separate individual objects even when several parts appear within the same image. Each detected part can be assigned:

Recognizing a part is only one part of the task. The robot must know where and how to pick it.

For simple flat parts on a conveyor, a 2D position and rotation may be sufficient. Random bin picking commonly requires full 3D pose estimation.

After picking, the robot transfers each part to its assigned destination. Destinations may include dedicated sorting bins, buckets, trays, pallets, assembly fixtures, machine-loading stations, packaging containers, rework areas, reject containers, and quality-review stations.

The AI classification result determines the destination automatically. The robot can also place parts in a controlled orientation when the next process requires organized presentation.

The system can verify whether the pick and placement were completed successfully by confirming that the original location is empty, detecting a part in the gripper, checking the destination bin, monitoring bin fill level, counting sorted components, confirming correct placement, and recording unsuccessful pick attempts.

In conveyor applications, the system must locate objects while they are moving.

This configuration is useful when parts are separated sufficiently and production requires continuous flow.

Random bin picking is more complex because components may overlap, interlock, or hide one another. A 3D vision system captures the bin and generates depth information. Intelgic's AI identifies the visible components and determines which part offers the safest and most reliable pick.

After a successful pick, the system captures new data and recalculates the next target. Difficult applications involving highly reflective, flexible, transparent, or heavily entangled parts may require specialized presentation or separation methods in addition to AI and robotics.

Industrial part recognition differs from general image recognition. Components may be reflective, oily, dusty, partially hidden, or visually similar. Their orientation and presentation may change continuously.

The solution may combine deep-learning AI with deterministic machine-vision measurements. AI provides flexibility for complex appearance, while calibrated vision verifies dimensions and specific features.

Part identification and defect inspection can be combined in a single automation cell. Before placing a part into its destination, the system may inspect it for missing features, cracks, surface damage, incorrect dimensions, deformation, contamination, incorrect assembly, or printing or marking errors.

The robot can then sort the component into accepted, rejected, rework, or review bins. This reduces repeated handling and turns the sorting station into an intelligent quality-control cell.

The system can provide operational information such as number of parts identified, quantity by part type, parts sorted per hour, pick success rate, misclassification or review rate, robot cycle time, bin fill level, unknown-part frequency, rejected-part count, downtime, repeated pick failures, and results by batch or shift.

These insights help manufacturers improve part presentation, gripper performance, production planning, and material flow.