To content
Faculty of mechanical engineering


04/01/2021 - 03/31/2024

System for adaptive phototonic surface testing with adaptive image evaluation in combination with a cleaning system

Problem and Motivation

Along with the development of ever smaller and more powerful assemblies and components in mechanical engineering, the automotive and electronics industries, the quality characteristic "technical cleanliness" (TecSa) has developed. It essentially describes the absence of unwanted matter within industrial manufacturing and assembly, which have negative consequences for the performance and reliability of components and the overall system. In addition, particles can lead to unwanted pseudo defects in the surface defect inspection process, which can lead to increased scrap or rework rates.

Currently, cleanliness analysis according to VDA 19.1 is mostly used, but this does not allow direct inspection of the components with simultaneous interpretation of the results. Due to the time latency, it is also not possible to adapt processes between the measurements and the results. Cleaning processes in particular are excluded.


The targeted innovation is based on research into an AI-based control logic that can analyze particulate soiling and scratches. In this project, a sensor will be automatically adapted to different surfaces and geometries and the resulting results of the sensor will be classified in the form of images.

The goal is to develop an assistance system that provides users with real-time information about the properties of particles and their formation in the production process. This will enable users to initiate improvements in the value chain. This is to be combined with a self-configuration of the cleaning system. On the one hand, this is intended to meet the social and ecological requirement of enabling demand-oriented cleaning with minimal use of energy and chemicals and, on the other hand, to reduce the effort required for the configuration of cleaning processes by experts.

[Translate to English:] © [Translate to English:] IPS​/​TU Dortmund
[Translate to English:]

Procedure and Division of Labour

The work plan of the project is divided into four work packages. The content of the research question with regard to the optical sensor will be elaborated by the AI-based control of the optical sensor, the classification of the generated particle images and the control of the demand-oriented cleaning process. In addition, a final validation of the project results and the transfer into practice shall be performed

Initially, the requirements for the overall systems will be collected and defined. With the addition of algorithms that support the automated image acquisition and the self-configuration of the sensor, the research work that concerns the segmentation of the particles on different surfaces will follow. In addition, the incorporation and verification of different classification models, taking into account the required database and the annotation of the images, which allow the interpretability of the sensor results.  In order for the cleaning parameters to be configured to achieve the desired cleanliness at minimal cost, reinforcement learning approaches will be explored. Thus, attention is paid to the sensor's incorporation potentials in cleaning systems with respect to the self-configuration of the overall system. Finally, the research results will be tested with regard to their application. Thereby interfaces for industrial integration and the economic potentials, which can be achieved by the application of the results, will be investigated.

Research-, Development- and Application Partners

The SysPOT research project is being developed by a team consisting of 14 consortium partners (research, development and application partners). PI Innovation GmbH is taking the lead in the consortium, while the Institute of Production Systems is contributing its knowledge on the use of Deep Learning in the field of technical cleanliness analysis, supported by IconPro.

Inline SPOT sensor: PI Innovation, Fraunhofer IPM, Hexagon.

Demand-driven cleaning: Höckh, Gläser, LPW

AI-based control: IPS, IconPro

Associated partners: Audi AG, Hansgrohe, sprintBOX, Walter AG, ZF, MTU Aero Engines


Funding Reference

The project (funding code: 100483490) is funded by the German Federal Ministry of Education and Research (BMBF) as part of the "Computer-Aided Photonics" funding area and is supervised by the project management organization VDI Technologiezentrum.