Classic wooden flooring creates a warm and inviting atmosphere, and its natural grain and color tones offer a lively effect and add character to a room. Scheucher Holzindustrie GmbH in Mettersdorf, Austria, is now in its fourth generation of manufacturing high-quality parquet floors, combining tradition with modern technology in its production. The company relies on image processing with artificial intelligence (AI) to inspect the glue joints of its multilayer parquet. The system is IDS NXT, an all-in-one system using intelligent cameras.

Application

Not all parquet floors are the same. In addition to solid variants, multi-layer parquet is particularly popular. Several aspects must be considered during its manufacture in order to obtain a high-quality, durable, and user-friendly product. The choice of wood species, for example, influences the stability and appearance of the parquet. As the name suggests, it consists of several layers — in particular a core layer and a wear layer. Scheucher uses spruce or pine as backing. A variety of wood species can be chosen for the wear layer.

A major advantage of "prefinished parquet" is its ease of installation. Many types can be laid as a floating floor, so they do not need to be glued down. With a little manual skill, even inexperienced people can install their own wooden floor. A click system ensures the exact fit of the planks. They are simply clicked directly into each other. This saves time and reduces the workload. It is important that the click system is milled with high precision so that the panels interlock stably and precisely. Incorrect milling can lead to problems during installation and instability. The same applies to the adhesive bond between the core layer and the wear layer of the parquet. The adhesive must be applied evenly to ensure complete coverage and a uniform bond between the layers. Air inclusions or irregular application quantities can lead to cavities and impair the durability.

"In the profiling for our MULTIflor parquet floors, a special click profile is milled into the planks. During this process, two intelligent IDS NXT cameras are used to check the presence of adhesive in the adhesive joint between the core and the backing layer," explains Dipl.-Ing. Klaus Bauer, Research and Development at Scheucher.

Intelligent Quality Control Without a PC 

Scheucher illuminates the already glued parquet planks with UV light. The adhesive fluoresces and adhesive residues become visible. Scheucher relies on image processing with AI to detect faulty floorboards. Two IDS NXT rio cameras are used per system. The intelligent models become powerful inference cameras by loading a neural network. They detect the fluorescent adhesive joints and process the image data directly "on the edge" via field programmable gate array (FPGA) and pass the result on to the downstream process. The embedded solution makes the host PC redundant and reduces energy consumption and system costs. Thanks to screwable and IP-protected connectors and a robust housing, they are also absolutely suitable for industrial use. The standardized OPC UA industrial protocol enables direct communication with machines and control systems.

Using AI Without Prior Knowledge of Deep Learning

The complete inference camera solution for AI applications in industrial image processing contains all the components that a user needs to create, train, and execute a neural network in a productive environment. Users are able to concentrate fully on the planned application without having to worry about deep learning, image processing, or camera programming.

"In our profiling system we had not used any image processing, neither rule-based nor intelligent, for the quality control of our bonded joints until now. Nevertheless, it was our goal, even as a layman, to solve this requirement using AI. As we are looking for errors that only occur extremely rarely, our challenge was to find a sufficient number of 'bad partial images' to train the neural network," explained Bauer. "And we succeeded. In the end, we needed around 60 good and bad part samples before the AI had learnt to recognize the threshold between OK and not OK."

Faulty panels can be reliably sorted out and reworked. The system prevents planks with defects in the parquet bonding from being delivered to customers. "This makes us one of the few manufacturers on the market to achieve a 100% inspection in this regard," said Bauer. Especially since it takes less than 90 milliseconds to check a floorboard.

Outlook

In this example, the inspection of the glued joint could also be solved using classic, rule-based image processing, though with increased programming effort. However, it was important to Scheucher to find an intelligent solution.