Data-driven detection of moving bottlenecks in multi-variant production lines

Because bottlenecks limit the throughput of production systems, it is important to correctly detect and control them. This task is especially demanding in high-speed dynamic production environments within asynchronous production lines. A change in conditions often shifts the bottleneck from one process to another. This paper proposes a data-driven concept for the detection of dynamic bottlenecks in multi-variant production lines. We build on and contribute to the literature of bottleneck detection methods. We propose a novel concept that dynamically and automatically detects bottlenecks using cycle time data from shop-floor machines. The cycle time distribution of produced batches is translated into the cumulative probability function, which is used to detect the moving bottlenecks.