AI Quality Inspection for Concrete Paving Block and Kerb Production

The measuring station is installed on the pallet conveyor between the vibro-press and the curing chamber. As the pallet passes, a laser line is projected across the blocks and an image of the top face is captured at the same time. Image capture is triggered by a pallet sensor or the conveyor encoder, so every image line is tied to a row on the pallet. A global shutter camera and short exposure keep the image of the moving pallet free of distortion.

Height is measured by laser line triangulation: because the camera views the laser line at an angle, the line shifts in the image when it falls on a higher surface. Calibration converts this shift into millimetres, and the height of every block is calculated relative to the pallet surface. Keeping block height consistent and close to the minimum the standard allows affects both quality and concrete consumption directly; a drift in height is usually the first sign of a change in filling, vibration time or mix.

Surface defects are found with a deep learning model: broken or missing corners, edge and surface cracks, incomplete filling, pores and voids, and spalling or colour differences in the coloured face mix. Because wet concrete reflects light and its tone changes as it dries, rule-based methods struggle here; the model is trained for your product types and colours on your own good and defective blocks. A separate recipe is kept for every product type (mould).

Results are logged with the pallet number, mould and shift; the operator screen shows the defect map and height trend of the latest pallets. An alert is raised when height approaches the minimum or when a defect repeats in the same mould cavity, so the press setting can be corrected or the mould checked before the problem spreads through a whole batch. Defective blocks can be flagged with pallet and position data for removal at the cubing stage after curing.

The laser line projector and the global shutter camera are mounted above the pallet conveyor in a housing protected against dust and moisture; images reach the edge AI computer over GigE. The Nexcom COEUS-3721T in stock (NVIDIA Jetson Orin Nano) runs the height profiling and the surface model under Linux; the global shutter Crevis MG-A121K-9 in stock is a suitable camera for moving pallets. Wide pallets or multi-row layouts use several cameras and lasers.

Results are written to the operator screen and a database. A link to the press PLC (pallet number, mould data, alert signal) can be made via Modbus TCP, OPC UA or a digital signal; which data is shared with the press and how the alert is used are decided together in the project.

What decides the success of an AI project is your product, your lighting and your line speed, not a catalogue figure. That is why every project starts with your own samples: first, images are collected on the bench with the right camera, lens and lighting, an initial model is trained, and the false reject and miss rates are measured on your parts. If the result meets the acceptance criterion, the system runs on the line as a pilot, initially in a mode that only logs and makes no decisions.

For paving block projects, feasibility is carried out with pallets from your own line: images are collected from different product types and colours and from good and defective blocks; height measurement is compared with reference blocks of known size and with calliper measurements after curing. During the pilot the system only records for a while; alerts are switched on after the results have been compared with your quality team's checks.

The system is added to a running vibro-press line, on the pallet conveyor between the press and the curing chamber; the press itself and the moulds are not touched. A bridge for the camera and laser above the pallet conveyor, a pallet detection sensor and an operator screen are all that is needed.