Fast solution strategies through artificial intelligence
Automated precision assembly is complex and benefits from our targeted deployment of AI
The assembly of optical systems is subject to high accuracy requirements: Even small deviations in the micrometer range change the optical quality of the entire assembly. The Fraunhofer IPT is using artificial intelligence to develop automated assembly processes. The positioning process can be optimized automatically and swiftly / quickly on the basis of self-learning algorithms. This holds out enormous potential for precision assembly of all kinds of components. Our AI approach can be adapted to suit any assembly scenario.
Conventional development approaches are time consuming and fraught with risks
Active and relative precision assembly is characterized by closed-loop positioning. Here, the error in the position and orientation of the components to be assembled is continuously detected by sensors such as cameras and corrected by applying suitable strategies involving the use of micromanipulators. The process calls for (um wiederholung von require zu vermeiden comprehensive expert knowledge in the evaluation of sometimes / often complex beam shapes and in the implementation of suitable strategies for optimization. In addition, in-depth understanding of the product and the corresponding automation competence are required. If the task cannot be performed adequately, the use of time-consuming brute force methods is necessary. Here, the best possible signal from a global scan is used, although this has dramatic effects on the cycle time.
Artificial intelligence loosens the interdependence between equipment manufacturer and operator
Comprehensive knowledge drawn from a range of disciplines including optics, computer science, measurement and control engineering, is essential for the development of an automated optical system positioning process. This creates significant hurdles in the development process and prevents sustainable, economical operation. System manufacturers and operators are then reliant (um Wiederholung von dependent zu vermeiden) on an intensive, cross-domain transfer of knowledge. This dependency often remains even after delivery has been completed if the systems are to be continuously optimized. This is the starting point for the AI solutions of the Fraunhofer IPT:
Methods from the field of machine learning, such as reinforcement learning, offer the possibility of learning suitable assembly strategies automatically on the basis of objectives and boundary conditions. This approach enables the plant operator to use their own production data to continuously improve the development as data-based training. Case studies have shown that data-based processes achieve optimal results in just a few steps resulting in optimized. Thus, development time and development risk as well as cycle time can be optimized.
The Fraunhofer IPT has the expertise and experience to make self-learning, adaptive processes based on machine learning usable for precision assembly. The use of artificial intelligence is also suitable for the assembly of non-optical components and systems. The processes are tested and evaluated using either our own machinery or the facilities provided by our customers.