Individualization of an assistance system

Assistance systems in manual assembly have the potential to relieve employees both cognitively and ergonomically and at the same time improve the efficiency and quality of assembly processes. Due to the increasing individualization of products and the associated growing complexity of modern assembly tasks, assistance systems are becoming increasingly important. However, many systems lack the necessary flexibility to switch dynamically between different process variants or to react to unexpected process steps that are unknown to the system. Especially in variant-rich assembly with small batch sizes, existing systems require a high level of manual adaptation. In addition, many assistance systems do not have the infinitely variable dynamics required to adapt to the individual skills and needs of employees in real time.

Against this background, there is therefore a need to develop an adaptive assembly assistance system that offers continuous and dynamic support based on the individual performance of the employees. The aim is to meet the requirements of modern assembly processes in order to increase both productivity and flexibility in industrial assembly in the long term.

Digital data processing

A standardized database is crucial for the development of a functional automated assistance system. The required data includes, in particular, a parts list of the components to be assembled, including their positions and orientations (poses), specifications of the individual components, for example in ECLASS format, as well as information on the external geometries of the components in a CAD format. A clustering algorithm is used to process and link this data, which categorizes the components based on the ECLASS and CAD information and generates assembly-relevant attributes, such as the required assembly method, for each component. In addition, the developed algorithm uses the parts list and the CAD model of the overall assembly to determine the respective position and orientation of the individual components. On this basis, a trained machine learning algorithm uses the categorization, attributes and pose information to create an optimal assembly sequence.

Both natural language processing (NLP) and automatically generated renderings of the components are used to display the assembly instructions. In combination with the assembly-relevant attributes and a database of labeled pictograms, the system makes it possible to generate a visual explanation of the assembly steps in addition to the automatically generated textual instructions.

Automatically recognize assembly progress

To ensure a continuous assembly process without manual acknowledgement of the completed steps, the system is equipped with sensors and cameras that continuously monitor the assembly process. Based on the information about the external geometries and the derived poses of the individual components, the system can recognize the current assembly step and evaluate it accordingly. The sensors compare the actual position of the assembled component with the target position, while the camera compares the features of the assembled component with the synthetically generated digital image to check whether the correct component has been assembled. The feedback on the completed assembly step is displayed visually on the screen for the assembler. If the assembly step is correct, the next step is visualized automatically.

Individualized assistance

To enable individualization of the assembly process, all assembly steps are recorded in a log file. This contains information on the assembled component and assembly, the time required, the error rate, the type of error and an anonymous employee ID. Based on the digital product data and the recorded assembly data, an AI-based, continuous instruction adaptation is developed. This analyzes and clusters the collected data of the assembly steps performed. The result of this analysis enables the system to automatically adapt the level of detail of the instructions so that employees are optimally supported without experiencing oversaturation or a lack of information. In addition, employees can provide feedback on each assembly step, which is also incorporated into the dynamic, continuous adjustment of the instructions and individually tailored to the needs and abilities of the employees.

Jonas Viehöver M. Sc., Prof. Dr.-Ing. Bernd Kuhlenkötter, Chair of Production Systems