greek english
intermek-logo solakis-logo


Automation of the Optical Screening of Marble Tiles

"RESEARCH - CREATE - INNOVATE" 8.6 Components and systems
Project Code T1EDK-00706

The Project.

Scope and Objectives of the project

Various phases of the process of processing marble slabs are currently carried out manually by experienced craftsmen:

  • Their resination in which a craftsman detects structural and appearance defects (e.g. cracks, cavities, etc.) in the plate and covers them with resins.
  • Their visual sorting in the final stage and their categorization into A-B-C qualities.

Although it would be possible to use robotic arms to assist experienced craftsmen (e.g. for resination) or even complete replacement (e.g. in sorting), there is no suitable software available for the visual identification of natural rock defects that would move the arm to process them with the appropriate procedure.

This particular proposal concerns precisely the development of all the individual necessary subsystems and their completion in an advanced machine vision software for the detection and detection of various quality characteristics of marble tiles, with the aim of automating their sorting process..

The proposed software, in the form of an experienced robotic vision system, will direct and control kinematically a general-purpose industrial robotic arm, equipped with a suitable tool in the end-effector (e.g. grab or suction cup).

The process of sorting the marble tiles (described in detail in the attached Annex I of the present) also raises issues of safety of the staff, due to the weight of the tiles and the strain on the employee during their palletization.


Project Implementation Methodology

The project will be implemented in the following distinct research phases:

  • Detailed recording of manual tile sorting. The sorting process for various types of marbles, mainly Greek (e.g. Drama, Kavala, Thassos) will be recorded (with cameras and other equipment). The images will be registered in a database and classified based on their qualitative characteristics.
  • Development of a visual screening algorithm. The recorded images and their quality characteristics will feed into an experienced machine vision system that will be developed with Deep-Learning technologies. The trained system will be computerized and verified.
  • Development of automation of visual sorting. It includes the integration of systems and the integration at laboratory level of mechanical vision automation and robotic control of industrial arm. In the laboratory areas of the TEI AMTH, a pilot application will be created: a production line (final phase of sorting) consisting of:
    • drive belt, where various ready-made tiles will be placed
    • an artificial vision system,
    • industrial robot – articulated arm and
    • hardware-software for the operation of automation and guidance of the arm.

Within the framework of the present proposal, an "intelligent" algorithm for the sorting of marble tiles will be developed, consisting of three distinct methods of processing hierarchically executed.

The algorithm will make use of advanced methods of artificial vision, pattern recognition and machine learning to give the system perception and intelligence capabilities similar to those of an experienced screening technician.

In particular, the algorithm will make use of geometric models, color characteristics and "deep learning" ( deep learning" models. ), each of which will be run in parallel on GPUs (graphics card) to accelerate calculations, ensuring short response time and meeting the real-time operation needs of the sorting system.

Expected Results

The main result of the project will be a complete and complete operational system of automated sorting of marble tiles, level TRL 6, with customization specifically for Greek marbles, consisting of:

  • Algorithm of mechanical vision (pattern recognition) and Deep Learning technology.
  • Integrated system (embedded system) of interoperability of mechanical vision, with robotic arm.
  • User interface software for the configuration, management and operation of the above system.
  • Training Dataset screening images from Patterns of Greek marbles.
  • Validated, verified operation in a laboratory pilot automated sorting device (TRL 6).


1. Supportive Actions.

It concerns actions of preparation of the whole research process and work planning actions, taking into account updated data and data..

  • 5/2018-9/2018

    1.1 Bibliographic Review

    Detailed overview of existing commercial products, technological solutions and scientific research related to the project.

  • 5/2018-11/2018

    1.2 Market Trends

    Recording and analysis of market trends for marble tile applications. Opportunities, growth forecasts per application (bathrooms, floors, kitchens, orthomarbling, residential & industrial users) and area (North America, Europe, Asia-Pacific, Other countries).

  • 8/2018–12/2018

    1.3 Development of Research Methodology

    Theoretical approach to the research process that will be applied by identifying the most important parameters to be considered, objectives and expected results.

  • 5/2018-3/2022
    Project completion

    1.4 Project Management

    Effective management and execution of the project will be carried out throughout the project to ensure the coordination of research and technical activities.

  • 5/2018–3/2022
    Project completion

    1.5 Publicity

    Publicity and dissemination actions, as described in the relevant section of the proposal.

2. Optical Data Collection

It concerns the design and installation - operation of the optical sorting data collection device, in the production line of marble tiles of the company SOLAKIS. Includes the following activities

  • 9/2018–1/2019

    2.1 Layout Design

    Detailed definition of the functional requirements of the optical data collection device. The device will be installed in the production line of marble tiles of the company SOLAKIS. Study of the workflow and development of an image classification protocol. Identification of types of marbles to be recorded and rough determination of their quality characteristics.

  • 11/2018–5/2019

    2.2 Data Collection Software

    Development of software for the automated recording of optical data from cameras and the creation of a Dataset with images that will include other data (eg time - date - type of marble, etc.)

  • 1/2019–7/2019

    2.3 Layout Installation

    Supply of the necessary software, hardware and other components. Construction of mechanical devices, adaptation of the existing production line. Installation with cameras and recorders, staff training, commissioning and instructions for use of the system.

  • 6/2019–7/2020

    2.4 Data Recording

    Operation of the manual recording device for manual sorting of marble slabs. Supervision, monitoring, troubleshooting - technical support. Sampling demo datasets at regular intervals and verifying them. Complete the process and download the final dataset from sorted - categorized images of quality characteristics of marble tiles.

3. Optical Sorting Algorithm.

Conventional visual screening of marble tiles requires special skills (acquired through experience) from the people who perform it, related to their ability to recognize the particular visual characteristics of each tile that determine their quality. To meet the aforementioned requirements, the proposed algorithm will consist of modern methods of artificial vision (computer vision), pattern recognition (pattern recognition) and machine learning (machine learning). With the help of these methods the sorting system will be able to perceive the working environment (sorting table and tiles) and with the appropriate "training" through "deep learning" (deep learning), it will be able to distinguish different qualities of tiles. , in a manner similar to the experienced technician performing the conventional sorting.

  • 7/2018–4/2020

    3.1 Development of a Theoretical Model

    The model that will be developed for the needs of visual screening, will perform the following procedures hierarchically:

    Geometric analysis
    Color analysis
    Resolution distribution of patterns (marble waters)

    Each of the above analysis phases of the tiles will be used for the qualitative categorization of the tiles.

  • 7/2018-6/2019

    3.2 Algorithm Development

    Each of the above three phases of tile processing requires the development of an autonomous and specialized algorithm: (1) for the geometric control of the tiles an algorithm for visual measurement of objects will be developed, (2) for the color control an algorithm will be developed which will process the local histograms of the color set of the color image of the tiles and (3) a "deep learning" algorithm will be developed to estimate the distribution and density of the tiles [4].Specifically, a "convolutional neural network" [5] will be developed, properly trained with all the images that will be collected in the previous section (EU2). It is worth mentioning that the performance of the network will be compared with various classic machine learning models (SVM, Extreme Learning Machines, etc.) that will be developed for the same purpose and which will use various distinguishing features e.g. texture characteristics.

  • 4/2019–4/2020

    3.3 Software Implementation

    The implementation of the proposed algorithm will be done using the Python programming language, while appropriate libraries of artificial vision will be used, e.g. OpenCV, TensorFlow, etc.

  • 7/2018–11/2020

    3.4 Completion of Artificial Vision

    Integration of the individual subsystems (Geometric analysis, Color analysis and Pattern distribution analysis) of the software. Testing the software in a laboratory environment and measuring the performance of the software in relation to its sorting performance (successful recognition) and response time (recognition speed). Adaptation of the software to the working conditions of the final system in its natural application environment.

4. Laboratory Completion

Completion of subsystems in a laboratory model and execution of experiments-tests. Improvements of subsystems and re-run of tests. Technology validation and functionality verification. Laboratory application for the use of the system under development, for demonstration purposes. Reliability check for industrial use. The EU includes the following activities:

  • 9/2019–4/2021

    4.1 Construction of devices

    Design - construction of drive belt, digitally controlled with inverter, supply & installation of other electromechanical components (micro-mounting devices, cameras, switches, etc.), installation of the robotic arm in a suitable position, connection of safety devices, initial kinematic function tests.

  • 6/2020–8/2021

    4.2 Arm Programming

    Development of the hardware and software for connecting mechanical optical vision to the robotic arm for guidance, interoperability and synchronization with the programmable drive belt

  • 1/2020–8/2021

    4.3 Laboratory Tests

    Design and conduct a series of experiments and tests for automatic sorting of marble tiles. Detailed recording and classification of results. Reliability testing for industrial use and TRL (Technology Readiness Level, according to NASA & Horizon 2020) at level 6 (System / subsystem model or prototype demonstration in a relevant environment). Development of training tools (Training Material).

  • 1/2021–2/2022

    4.4 Results – Conclusions

    Elaboration and presentation of the results obtained from the project. Processing and scientific - technological documentation and analysis of industrial research tests. Recording, presentation and analysis of functions that need further optimization.

  • 8/2020–3/2022

    4.5 Technical Application Study

    Analysis study of the required resources, procedures and technical requirements, in order for the results of the industrial research to be applied on a productive scale and to emerge a new product, commercially available in the near future.


Interpretable Deep Learning for Marble Tiles Sorting

Athanasios G. Ouzounis, George K. Sidiropoulos, George A. Papakostas, Ilias T. Sarafis, Andreas Stamkos and George Solakis

One of the main problems in the final stage of the production line of ornamental stone tiles is the process of quality control and product classification. Successful classification of natural stone tiles based on their aesthetical value...


Texture Analysis for Machine Learning Based Marble Tiles Sorting

George K. Sidiropoulos, Athanasios G. Ouzounis, George A. Papakostas, Ilias T. Sarafis, Andreas Stamkos, George Solakis

In this paper, the classification of ornamental dolomitic marble stone tiles, in regard to their aesthetical value, was studied based on the rock’s texture. The stone tiles examined are of a dolomitic marble variety commercially known...


Exploiting Deep Metric Learning for Mable Quality Assessment with Small and Imbalanced Image Data

George K. Sidiropoulos, Athanasios G. Ouzounis, George A. Papakostas, Ilias T. Sarafis, Andreas Stamkos, Vassilis Kalpakis, George Solakis

The classification of ornamental dolomitic marble stone tiles has been an issue in the past years, even more so according to their aesthetical criteria. Quality control and product classification during the final stage...


Marble Quality Assessment with Deep Learning Regression Models

George K. Sidiropoulos, Athanasios G. Ouzounis, G Taxopoulos, George A. Papakostas, Ilias T. Sarafis, Andreas Stamkos, George Solakis

Natural rock tile classification, with the use of computer vision and machine learning techniques, is a methodology well documented in academic literature. The broad variety of textures present on the rock tiles’ surface...


Contact. +30 251 039 2288 +30 251 039 2488
  • Amisiana, Municipality of Paggaio, 64100