*3.1. Disassembly Planning Using an Adaptive Planner*

The disassembly task planner presented by Choux et al. [13] consists of four steps: (1) image capturing by the installed vision system, (2) detection of the different components of the battery, (3) autonomous decision making about the possible disassembly sequences, and (4) position and path calculation to remove the components by the mechanical system of the disassembly station. Considering the identified issues in the previous section, we present our design for an adaptive disassembly planner (see Figure 3). Thereby, we extend the steps presented by Choux et al. with a disassembly strategy optimizer. However, disassembly strategy optimization requires a lot of data about the product, the disassembly process, and the market. For this reason, the disassembly strategy optimizer can be ignored if these data are not available.

Nevertheless, these gaps can be minimized in a disassembly factory after gaining some experience, e.g., by performing several disassembly experiments with autonomous decision-making or by taking advantage of the expertise of disassembly experts gained from past similar batteries. Once these data are available, an optimal disassembly strategy can be determined. In this way, the autonomous disassembly decision can be avoided. The component i to be disassembled is thus specified by the optimizer. After the calculation of the positions, disassembly operations can be performed.

If the disassembly proceeds normally, the disassembly strategy only needs to be calculated once. In case of failure, due to the many uncertainties during the disassembly process, or when complications occur that complicate the execution of disassembly with the existing strategy or pose any safety risks, alternative actions must be initiated. Here, two cases can be distinguished. The first case occurs when there is no optimized disassembly strategy due to incomplete data. In this case, the disassembly task planner must make an additional autonomous decision, such as changing disassembly tools or adjusting the disassembly sequence. The second case occurs if the disassembly strategy optimizer is active. The status data will then be updated if any components were damaged during the disassembly action. This is an important step, since it significantly impacts the optimized disassembly planning, as the appropriate circular economy strategy depends on the component condition. After that, a new disassembly strategy can be calculated to proceed with the disassembly by removing component i. If the disassembly is not yet complete, new images are captured to detect component i + 1 of the computed disassembly strategy. Thereby, it is necessary to check if the previous disassembly step of component i was performed correctly. If any connections were damaged or if the component i was destructively removed, that would require an update of the disassembly strategy. In this case, new precedence constraints may arise that need to be calculated. Moreover, the status data must be updated to calculate

the next steps of the disassembly subsequently. In the next section, the design of the disassembly strategy optimizer implemented in this work is addressed.

**Figure 3.** Concept of the adaptive planner.
