⚡ electrical engineering

Digital Twin-Enabled Dynamic Production Scheduling for Concurrent Multi-Program Manufacturing Operations

This paper proposes a hierarchical Digital Twin-enabled framework that integrates machine health prediction, fuzzy inference, and multi-agent coordination to dynamically optimize production scheduling for concurrent multi-program manufacturing under various disruptions while balancing makespan, energy consumption, and schedule deviation.

Original authors: MD JAHEDUL ISLAM ARIF

Published 2026-10-16

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: MD JAHEDUL ISLAM ARIF

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 heart of modern industry, factories are no longer just rows of silent machines waiting for instructions. They are complex, living systems where dozens of different product lines run at the same time, sharing the same tools, the same workers, and the same power. The challenge for managers is to keep these lines moving smoothly even when things go wrong. A machine might slowly wear out, a critical part might break, or a customer might suddenly demand an urgent order that jumps the line. Traditional methods of planning production often rely on a static map drawn before the work begins. Once the work starts, if a machine slows down or stops, the plan often fails, leading to delays and wasted energy. To solve this, engineers have turned to a concept called the "digital twin." Imagine a perfect, virtual copy of the entire factory floor that updates in real time, mirroring the physical world exactly. By watching this virtual version, managers can see problems before they happen and test new plans without stopping the real machines. This approach allows factories to adapt instantly, shifting resources and changing schedules the moment conditions change, ensuring that production continues efficiently despite the inevitable chaos of the real world.

A researcher at Lamar University has proposed a new way to use this technology specifically for factories running multiple programs at once. The study introduces a layered system that acts like a nervous system for the factory, connecting the health of individual machines to the big picture of the entire production schedule. Instead of waiting for a machine to break completely, the system watches for early warning signs of trouble. It uses a type of computer model trained to recognize patterns in data like motor vibrations and electrical currents to predict if a machine is healthy, starting to degrade, or about to fail. This information is not kept in a separate maintenance log; instead, it is fed directly into the scheduling brain. If a machine is predicted to get sick soon, the system can provide early degradation information to the global scheduling layer, allowing it to move work to another machine before a breakdown or bottleneck forms.

The system is built in two main layers. The first layer consists of "local twins" for each workstation. These digital copies constantly monitor the physical equipment. When they detect a change in the machine's behavior, they pass this information up to a "global twin," which holds the map of the entire factory's current state, including what orders are in progress, where materials are, and which workers are available. To decide what to do next, the system uses a fuzzy logic controller, a type of reasoning engine that handles uncertainty much like a human supervisor does. It doesn't just see "good" or "bad"; it understands shades of gray, such as a machine that is "mostly good but slowing down." Based on this nuanced view, the system decides whether to keep running as planned, call for a human operator to check the machine, or trigger a full rescheduling of the work.

When a change is needed, the system does not just shuffle the schedule randomly. It runs a complex search to find the best possible new plan. This search looks for a balance between three competing goals: finishing all the work as quickly as possible, using the least amount of energy, and keeping the new schedule as close to the original plan as possible to avoid confusion. The system generates many possible solutions and presents the best trade-offs, allowing managers to choose the path that fits their current priorities. Before any new plan is sent to the real factory floor, it is tested in a three-dimensional virtual simulation. This step ensures that the new schedule will actually work in the physical world, checking for collisions, delays, or resource conflicts. Only after the virtual test passes does the system send the updated instructions to the real machines via a secure communication link.

The researcher tested this framework using a series of simulated scenarios to see how it would perform under pressure. The simulations included normal production, gradual machine wear, sudden machine failures, and the insertion of urgent orders. In the virtual tests, the new system showed a clear advantage over traditional methods. When a machine began to degrade, the system managed to reduce the total time needed to finish the work by about 5.4 percent compared to standard methods that simply waited for the machine to fail. In cases of sudden machine failure, the improvement was even more significant, cutting the delay by roughly 13.3 percent. When an urgent order was inserted, the system reduced delays by about 11.8 percent by intelligently reshuffling the schedule rather than just pushing everything back. These results suggest that by combining early health predictions with flexible scheduling, factories can maintain a steady flow of production even when disruptions occur.

The study also highlighted how the system handles the trade-offs between speed, energy, and stability. The simulations showed that the system could find schedules that saved energy, but this often meant the schedule would deviate more from the original plan. Conversely, the fastest schedules sometimes required more energy or larger changes to the workflow. The system successfully identified these compromises, offering a set of options that respected the physical limits of the machines and the availability of workers. The research emphasizes that this is a simulation-based finding, meaning the numbers come from a computer model of a factory rather than a real-world factory floor. While the results are promising, the author notes that the system still needs to be tested with real industrial data and physical experiments to confirm its performance in the messy, unpredictable environment of an actual manufacturing plant.

Ultimately, this work proposes a closed loop where the factory can see its own future. By linking the health of a single machine to the schedule of the entire factory, and by testing every new idea in a virtual world before trying it in reality, the system offers a way to keep production moving smoothly. It moves beyond the old idea of fixing things only after they break, suggesting instead a future where production plans adapt in real time to the condition of the equipment. The framework provides a structured way to handle the complexity of modern manufacturing, where multiple programs compete for resources and unexpected events are the norm. While the full potential of this approach awaits real-world validation, the simulations indicate that such a system could significantly reduce downtime and energy waste, making the complex dance of concurrent manufacturing operations more resilient and efficient.

Technical Summary: Digital Twin-Enabled Dynamic Production Scheduling for Concurrent Multi-Program Manufacturing Operations

Problem Statement
Modern manufacturing facilities increasingly operate multiple concurrent production programs, creating complex coordination challenges involving changing order priorities, machine degradation, workforce capacity, material availability, and delivery constraints. Traditional scheduling approaches rely on static assumptions regarding machine availability, processing times, and routing sequences. These assumptions often fail when disruptions occur, such as unexpected machine failures, gradual equipment degradation, urgent order insertions, or rework requirements. While existing research has addressed adaptive scheduling, multi-agent coordination, and predictive maintenance in isolation, a unified framework that integrates machine-health prediction with global scheduling decisions for concurrent multi-program operations remains limited. Specifically, there is a gap in connecting predictive machine intelligence directly to dynamic rescheduling actions within a closed-loop control structure.

Proposed Methodology
The study proposes a hierarchical Digital Twin (DT) framework designed for dynamic scheduling across concurrent multi-program manufacturing. The methodology is structured into four integrated stages: state acquisition, condition prediction, intelligent scheduling decision, and virtual validation.

Hierarchical Architecture: The framework consists of local workstation Digital Twins and a global Digital Twin.
- Local Workstation DTs: These models mirror the physical state of individual machines, collecting data such as pneumatic pressure, motor current, vibration RMS, and vibration kurtosis. They utilize Multilayer Perceptron (MLP) models to classify machine conditions into three states: Good, Degraded, or Failure. This prediction serves as a direct input for scheduling rather than an isolated maintenance alert.
- Global Digital Twin: This layer aggregates local health states with production-level data, including production rates, program priorities, Work-in-Process (WIP), queues, and resource assignments.
Decision Logic and Coordination:
- TSK-FIS (Takagi-Sugeno-Kang Fuzzy Inference System): The global DT employs a TSK-FIS to determine operational responses based on machine condition and production rate. The system outputs three primary actions: No Operation, Operator Assistance, or Scheduling. This layer provides interpretable decision-making, linking specific fuzzy rules to observed states.
- Multi-Agent System (MAS): A distributed MAS coordinates information and local decisions across workstations, while a high-level orchestration agent manages interdependent operations to prevent unnecessary schedule changes.
Dynamic Rescheduling and Optimization:
- When the orchestration layer triggers a rescheduling event, an evolutionary multi-objective optimizer generates candidate schedules.
- The optimization minimizes three objectives simultaneously: Makespan ( $C_{max}$ ), Total Equipment Energy Consumption ( $E_{all}$ ), and Machine Deviation from the existing schedule ( $D_{all}$ ).
- Constraints include machine availability, routing precedence, labor capacity, material availability, and program priorities. The result is a Pareto set of feasible schedules.
Virtual-to-Physical Validation:
- Before physical deployment, candidate schedules undergo validation via a 3D Discrete-Event Simulation (DES).
- OPC UA is utilized to synchronize data between the physical PLCs and the virtual environment, ensuring real-time state updates.
- Performance metrics include synchronization skew, latency, and the deviation between the Digital Twin's predicted makespan and actual physical execution time.

Key Contributions
The paper outlines five primary contributions to the field of smart manufacturing:

Hierarchical Architecture: A novel structure connecting local workstation intelligence (MLP-based health prediction) with global production scheduling.
Integrated Decision Logic: The fusion of MLP-based machine-health prediction with TSK-FIS decision logic, allowing equipment conditions to directly influence scheduling actions.
Multi-Objective Optimization: Application of multi-objective optimization to concurrent programs that balances makespan, energy consumption, and schedule stability under complex constraints.
Virtual Validation Stage: Introduction of a 3D DES and OPC UA synchronization layer to validate schedules virtually before physical execution.
Comprehensive Evaluation: Assessment of the framework's performance under various disruption scenarios, including machine degradation, failure, and urgent order insertions, measuring metrics such as makespan, energy, and schedule stability.

Results and Illustrative Findings
The study presents simulation-based illustrative findings rather than empirical industrial data. The results compare the proposed Digital Twin scheduler against two baselines: Original Schedule Delay (absorbing disruptions without rescheduling) and Right-Shift Rescheduling (shifting operations without predictive health data).

Makespan Reduction: Under illustrative disruption scenarios, the proposed framework demonstrated potential makespan reductions compared to the right-shift baseline:
- 5.4% reduction under machine degradation.
- 13.3% reduction under machine failure.
- 11.8% reduction following urgent order insertion.
Proactive Response: The framework enables proactive rescheduling by utilizing machine-health data before significant operational changes occur, contrasting with reactive approaches.
Trade-off Management: Pareto analysis illustrates the trade-offs between minimizing makespan, reducing energy consumption, and maintaining schedule stability, allowing operators to select solutions based on current priorities.
Synchronization: The framework successfully models the synchronization between virtual and physical layers, though specific latency values are noted as illustrative.

Significance and Limitations
The paper claims significance in providing a closed-loop decision structure for proactive rescheduling across concurrent manufacturing programs. It addresses the integration gap between predictive maintenance and production scheduling, offering a mechanism to detect operational changes and generate revised schedules dynamically.

However, the author explicitly states limitations:

The study relies on simulation-based evaluation and illustrative scenarios rather than real industrial datasets or measured shop-floor observations.
The framework requires further testing with diverse production layouts, product mixes, and sensor configurations before industrial deployment.
Real-world validation using physical shop-floor experiments and industrial data is identified as necessary for future work.

In conclusion, the study proposes a theoretical and simulated framework that integrates machine learning, fuzzy logic, and multi-objective optimization within a Digital Twin architecture to enhance the resilience and efficiency of concurrent multi-program manufacturing operations.

Technical Summary: Digital Twin-Enabled Dynamic Production Scheduling for Concurrent Multi-Program Manufacturing Operations

More like this