AI-based assistance system for manual assembly in loudspeaker production

Assembly is an essential part of the industrial value creation process. In Germany in particular, many products continue to be assembled predominantly by hand due to small batch sizes, high variety of versions, and complex manufacturing steps. This also applies to electroacoustic devices, where precision, experience, and sensitivity play a decisive role.

This requirement is particularly evident in high-quality loudspeakers: Different chassis, delicate components, and precision joining operations require qualified personnel and detailed expertise. Only after complete assembly can acoustic, electrical, and sensory tests reliably assess whether a system is working flawlessly. If deviations are detected, an often time-consuming troubleshooting process begins—including possibly multiple rework cycles.
 

Project goal: AI-supported assistance for troubleshooting and rework

Against this backdrop, the project aims to develop an AI-supported production assistant that supports humans in manual assembly, precisely analyzes deviations, and identifies their causes in order to provide targeted recommendations for efficient rework.

Unlike traditional quality assurance systems, the solution does not exclusively evaluate the overall quality of the final products. Since high-quality loudspeakers often require rework, the assistance system intervenes earlier in the process: it is designed to identify deviations in a targeted manner, thereby significantly reducing the number of necessary production loops.

The innovative core consists of linking measurement and process data as well as manual inspections by specialists (visual, acoustic, haptic) in a multimodal way for the first time and deriving interpretable recommendations for rework processes from this. The combination of physical, manufacturing, historical, and sensory data creates an AI-based assistance system that not only recognizes that there is a fault, but also why.

To do this, the production assistant processes various data sources and analysis steps:

• Physical test variables such as transfer functions, resonances, or typical error patterns (e.g., knocking, vibrations, air pockets)
• Manufacturing information such as process times, screw torques, or soldering temperatures
• Visual, acoustic, and haptic evaluations done manually by experienced employees
• Acoustic and electrical measurement data, supplemented by intelligent filtering methods to suppress noise and structure-borne sound

Building on this, model-based evaluation algorithms, domain-specific knowledge, and explainable AI methods are used. They allow reliably narrowing down the causes of faults, revealing correlations, and deriving specific recommendations for rework.
 

Responsibilities of Fraunhofer IDMT

Fraunhofer IDMT is responsible in particular for the following aspects of the project:

• the development of AI-based classification and diagnostic models,
• the multimodal linking of acoustic, physical, and process-related data,
• the analysis of previously unknown cause-effect relations using cluster and correlation analyses,
• the development of explainable AI methods that make decision-making processes transparent,
• and the integration of the developed AI into a prototypical production assistant.

Fraunhofer IDMT's expertise in audio analysis, intelligent sensor signal processing, and explainable AI is resulting in a solution that digitizes the knowledge of experienced specialists and effectively supports production-related decisions.
 

The future of manual assembly: AI assistance systems for more efficient processes

The AI-supported production assistant shows how intelligent assistance systems can transform manual assembly in the future. Particularly in the manufacture of high-quality loudspeakers, the combination of measurement data, specialist knowledge, and explainable AI opens up completely new possibilities: errors can be diagnosed more quickly, rework processes can be carried out in a more targeted manner, and quality standards can be maintained at a consistently high level. The production assistant is designed as a supportive assistance system that complements the work of skilled workers rather than replacing it.

The solution goes far beyond classic quality assurance. It digitally maps experience-based knowledge, recognizes complex relationships in assembly, and supports skilled workers with precise, comprehensible instructions. This creates a modern, future-proof assembly process in which humans and AI work together to ensure more efficient processes and sustainably higher product quality.