Proceedings of the

The 33rd European Safety and Reliability Conference (ESREL 2023)
3 – 8 September 2023, Southampton, UK

Optical Surface Analysis with Support Vector Machines based on Two Different Measurement Techniques

Marcin Hinz1, Jannis Pietruschka2,a and Stefan Bracke2,b

1Chair of Artificial Intelligence in Mechanical Engineering, Department of Mechanical, Automotive and Aeronautical Engineering, Munich University of Applied Sciences, Germany.

2Chair of Reliability Engineering and Risk Analytics, Faculty of Mechanical and Safety Engineering, University of Wuppertal, Germany.


The surface topography as well as the optical perception are important features for evaluating the quality of fine grinded knives. Parameters as the surface roughness, gloss or coloring are used for the quantification of these features. The measuring is implemented by the use of traditional methods, which are manual, time-consuming and cost-intensive. On top of that, the application of these methods for the condition monitoring of the ongoing process is rather limited. Therefore, a new, faster and more cost-effective approach is needed to improve the classical measurement methods. A conceivable approach could be based on image analysis.

Over the past years, different contactless image analysis based approaches have been developed to simplify the traditional roughness measurement methods. Some studies propose picture pre-processing and feature extraction in combination with machine learning algorithms.

The overall goal of the presented research activities is the development of a condition monitoring tool which can be implemented in the ongoing grinding process of the knives. It should be used to ensure the knives quality and to reduce rejects by an immediate detection of deviations of the target values and the possibility to adapt the production process accordingly. For this reason, a data set based on cutlery samples has been generated and analyzed. The extraction of features of the data set is presented for a better understanding of the training process. The features are used to train various machine learning algorithms with and without a combination of logged process parameters to evaluate the surface roughness.

Within this study the image of each grinded surface is analyzed regarding its measured arithmetic average roughness value (Ra) by the use of Support Vector Machine (SVM) and Support Vector Regressor (SVR) algorithms.

Keywords: Machine learning, SVM & SVR, Supervised learning, Surface topography, Condition monitoring.

Download PDF