Self-adaptive KANBAN-based Material Supply
- Purpose of the paper:
Since long, KANBAN is used in industry as an effective approach to provide the required materials to the shop floor. In its traditional form, the parameters for KANBAN, that is for each material the number of boxes and the quantity per box, are static, i.e. calculated
initially and, if at all, re-calculated only if someone suspects the parameters to be suboptimal. This paper shows how a traditional KANBAN system can be converted into a selfadaptive KANBAN, where the parameters are dynamically adjusted to reflect the dynamics of the environment. The self-adaptive approach aims to set KANBAN parameters in a way that shop-floor inventory is at a viable minimum without compromising production output.
Design/methodology/approach:
Using a simulation of the process of supplying material, we suggest an optimization method based on Evolution Strategies (ES). The strategy aims at finding the minimum quantity of material supply per box, while maintaining production efficiency, by dynamically and iteratively adjusting (self-adapt) KANBAN parameters.
Findings:
Compared to the traditional KANBAN approach, self-adaptive KANBAN is more flexible and efficient in unstable production environments, in which key variables are volatile, e.g. demand because of changes in the model mix or replenishment lead time due to the variability of capacity in picking and in-house transportation. In these cases, self-adaptive KANBAN can adjust material supply parameters for stabilizing the production output by avoiding material shortages while, at the same time, reducing on-hand inventory.
Value/Originality:
We provide an approach based on evolution strategy (ES) that aims to optimize the KANBAN system parameters automatically. We implemented a simulator for a lab-based manufacturing environment, SimCar, that relies on KANBAN-based material supply. The simulation study shows that the ES algorithm is capable of coping with two target objectives at the same time, productivity as the primary target and shop-floor inventory as the secondary target.
Research limitations/implications:
Data for this study has been collected from a simulation study, resembling SimCar, a lab- based manufacturing environment. More case-study research is required to test the approach in large-scale industrial applications. A closer look at the mutation strength σ, a key parameter in ES, is required to define a strategy for selecting the “best” σ for a certain situation. Furthermore, in essence, the self-adaptive approach is still reactive. For a proactive approach, it would be necessary to establish deep-learning algorithms on extensive data sets (including, but not limited to, availability and utilization of transportation and picking capacity) collected from the manufacturing environment.