In production engineering, the purity of a product and the associated quality control play a central role. In order to ensure maximum material and cost savings, the reliable measurement of the manufactured products as well as the detection of irregularities with regard to the given product specification is essential. The aim of ML-X-RAY, funded by the European Union, is to investigate and further develop a measurement system for the inspection of inhomogeneous cable and pipe products using innovative approaches from the field of Machine Learning (ML) or Deep Learning (DL) in the form of Convolutional Neural Networks (CNN).
Basically, a distinction is made between so-called homogeneous and inhomogeneous products in manufacturing. The group of homogeneous products includes, for example, smooth tubes (rotationally symmetrical hollow cylinder geometry without modulation in the length profile). The focus of this project is on the measurement of inhomogeneous products. Due to the manufacturing processes, inhomogeneities in products, especially in pipes and cables, can basically exist in two planes. On the one hand, these can be found in the direction of production or extrusion, corresponding to a modulation in the length profile as in the case of so-called corrugated pipes. On the other hand, the cross-section of the manufactured product, as in the case of so-called stranded cables, can be asymmetrical in relation to the longitudinal axis due to the presence of more than one core.
The foundation of the project is provided by the multi-axis system X-RAY 6000 of the company SIKORA AG, which consists of X-ray sources and direct, digital X-ray images. This system enables the non-contact measurement of multi-layered products of the wire and cable as well as tube and hose industry. Typically, products made of materials such as nylon, rubber, silicone, polyethylene (PE), polyvinyl chloride (PVC), high-density polyethylene (HDPE) or ethylene propylene diene monomer (EPDM) are measured with regard to diameter, ovality, wall thickness and concentricity.
The inhomogeneities described above result in increased demands on data analysis. Therefore, novel ML methods as well as physical models are to be developed within the framework of the project. The physical models enable the synthesis of training data of different geometric shapes, which can be used for the ML methods based on neural networks. On the one hand, signals are recorded in the form of intensity curves, in which characteristic patterns describe the layer transitions between different materials. In the current software, this pattern recognition or layer transition identification is based on classical curve fitting methods, which are, however, not technically capable of correctly recognising the more complicated patterns of inhomogeneities in a robust manner. On the other hand, patterns in the extrusion direction are to be recognised and evaluated on the basis of the temporally correlated intensity curves.
The participating industrial partner SIKORA is one of the leading companies for measuring and control technology as well as inspection, analysis and sorting systems and one of the hidden champions in the German medium-sized business sector. In the cable industry, SIKORA is the world market leader in the field of measuring technology. The project ML-X-RAY aims at the development of a software innovation based on artificial intelligence (AI), which will significantly extend the applicability of the measuring devices of SIKORA AG.
