A Closed-Loop Data-Driven Framework for Value Stream Mapping and Future-State Design
This paper proposes a closed-loop, data-driven framework that integrates process mining, complex network analysis, explainable machine learning, and digital twin simulation to automate Value Stream Mapping and quantitatively optimize Future-State designs, demonstrating significant reductions in lead time, work-in-process, and defects across real-world manufacturing scenarios.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the bustling world of modern manufacturing, companies have long relied on a method called Value Stream Mapping to understand how their products move from raw materials to finished goods. Imagine a team of engineers walking a factory floor with clipboards, timing how long a worker takes to tighten a bolt, noting where parts sit waiting in a pile, and sketching a static picture of the entire process on paper. This traditional approach helps identify waste and delays, but it has a fundamental flaw: it captures only a single moment in time, often missing the hidden chaos of real life, such as unexpected machine breakdowns, rework loops, or the subtle ways different tasks interfere with one another. As factories have become more digital, generating vast amounts of data from sensors and computer systems, the old method of manual observation has struggled to keep pace with the complexity of the data available.
Researchers have been exploring ways to use this digital data to build a more accurate picture of how work actually happens, moving beyond static sketches to dynamic, data-driven models. They have developed tools that can automatically reconstruct the history of a process from digital records, much like reading a diary to understand a person's day. They have also begun using mathematical techniques to see the hidden structure of how tasks connect, and computer simulations to test changes before making them in the real world. The central question for the field has become how to take all these separate digital tools and weave them together into a single, reliable path that leads from raw data to a better-designed factory.
A team of researchers led by Edwin Montes-Orozco has proposed a new framework that connects these digital capabilities into a closed loop, creating a systematic way to redesign production processes. Instead of relying on manual observation, their method starts by feeding raw event logs—digital records of every action taken in a factory—into a system that automatically rebuilds the current state of the process. This system does not just list what happened; it reconstructs the flow of materials and information, identifying exactly where time is lost and where defects occur. It then layers on a structural analysis that looks at how different tasks and resources depend on one another, revealing connections that a simple timeline might miss.
Once the current state is understood, the framework uses explainable machine learning to predict where problems are likely to happen next. Unlike a "black box" computer program that gives an answer without explaining why, this system highlights the specific factors driving delays or errors, allowing human experts to understand the root causes. With these insights in hand, the researchers use a digital twin—a virtual replica of the factory—to simulate different improvement strategies. They can test ideas like adding more workers to a specific station or changing the order of operations, watching how these changes ripple through the entire system in a safe, virtual environment before ever touching a real machine.
The researchers tested this approach using a controlled, synthetic manufacturing environment where they could precisely measure the effects of every change, as well as three real-world datasets from different industries, including business processes and procurement workflows. In the synthetic environment, the complete framework, which combined all the digital tools, demonstrated a powerful ability to improve performance. When compared to the baseline process, the new design reduced the total time a product spent in the system by 32.40 percent and cut the amount of unfinished work sitting in queues by 45.09 percent. It also lowered the rate of defects by 27.63 percent and reduced the need for rework by 45.52 percent, while simultaneously increasing the number of finished products by 71.73 percent.
The study also examined how well the system could predict specific problems, such as rework or bottlenecks, using real-world data. The results showed that adding the structural analysis of how tasks connect significantly improved the accuracy of these predictions. For instance, in one dataset, the ability to predict rework improved dramatically when the system considered the network of connections between activities, suggesting that understanding the shape of the process is just as important as knowing the speed of individual steps. The researchers found that these improvements were not just random fluctuations; statistical tests confirmed that the gains were significant and consistent across the different configurations they tested.
Crucially, the paper argues against the idea that simply having more data or using a single advanced tool is enough to solve these problems. The researchers demonstrated that the greatest benefits came from integrating all the steps into a continuous cycle: reconstructing the current state, diagnosing the causes of poor performance, simulating potential fixes, and then designing a new future state based on the evidence. They showed that treating these tools as separate, isolated technologies misses the opportunity to create a coherent path from observation to action. The framework does not replace human decision-making but provides a rigorous, evidence-based foundation for it, allowing managers to see the consequences of their choices with a clarity that was previously impossible.
While the results are promising, the researchers are careful to note that the most dramatic improvements were observed in a controlled simulation environment designed to validate the method. The real-world tests confirmed that the approach works across different types of industries and data qualities, but the full cycle of designing, simulating, and implementing a new factory layout has not yet been tested in a live industrial setting. The work suggests that by combining process mining, structural analysis, and digital simulation, organizations can move from guessing about improvements to knowing exactly what will work, turning the art of continuous improvement into a precise, data-driven science.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.