In a context of customization, current assembly system configurations typically result in suboptimal use of space and capex, as (1) they are not continuously operating at full potential (e.g., volume changes due to product growth, market fluctuations or end-of-life) and (2) they need to be drastically and regularly redesigned (eg new products or volume changes). Although Reconfigurable Manufacturing Systems (RMS) were proposed in 1999, much less attention has been paid to Reconfigurable Assembly Systems (RAS). This project will focus on such RAS and address the following barriers that hinder its industrial adoption:

• Lack of a decision framework to decide on RAS reconfiguration on three levels: workstation level (1 day to 1 month), system architecture management level (1 day to 1 month) and online task execution level (1 to 10 seconds). The former two respond to production changes, such as product mix, variants, or volume, while the latter responds to incoming assembly orders and possible fulfillment disruptions, such as stock shortages, rush orders, quality issues, and failures.
• Lack of a proof of concept in a relevant environment showing the potential of RAS in terms of optimization of capex & space utilization enabled by a new decision framework and established technologies (related to flexible flow systems, inline kit and picking systems, industrial machine to machine communication protocols and new design concepts for flexible workplaces)

 To do this, the project will deliver both an Assembly Configuration Recommendender (ACR) and an Assembly Execution System (AES), which will work closely together. The ACR will propose reconfiguration based on external triggers extracted from production changes using a two-step approach. In a first step, initial configurations will be proposed based on historical data and configurations. To this end, a graph-based database approach is realized with, among other things, models describing the RAS characteristics and capabilities. The ACR can call on six different optimization modules in two categories (workstation and system architecture) to calculate optimal RAS configurations, cleverly combining (meta) heuristic approximations with generic local search engines. The AES follows a distributed approach where incoming assembly commands are translated into dynamic task execution commands for the respective hardware modules. Each module will continuously indicate its availability and status. Because each module owns its own time slots, fast local optimizations are possible in the event of disruptions, such as outages or stock shortages. The AES will continuously update stochastic models to estimate performance such as task execution times, reduce uncertainty and predict the need for global reconfiguration requests for the ACR.

The results will be validated and demonstrated on increasing Technology Readiness Levels (TRLs) during project execution on the InfraFlex_infra infrastructure using three predefined scenarios. Compared to a benchmark made using state-of-the-art techniques (baseline), the project aims to reduce the Total Cost of Ownership (TCO) and increase the robustness of the RAS for these scenarios, in accordance with the defined expected impact in the Flexible Assembly cluster roadmap