Enhancing manufacturing processes through control education with MIT FrED platform

1 Introduction

The transition toward advanced and digitalized manufacturing requires not only highly skilled engineers but also well-trained technicians capable of implementing and maintaining complex control systems (Rapitsenyane et al., 2023). Professionals must be able to model, identify, and control complex industrial dynamic processes while adapting to the inherent uncertainties of real-world production environments. As a result, the challenge for engineering education is not only to prepare students in universities but also to provide continuous training opportunities for industry personnel who must operate and troubleshoot advanced manufacturing systems.

Control theory remains a fundamental component of electrical, mechanical, mechatronic, and automation engineering curricula, providing students with essential tools for modeling, analyzing, and designing feedback control systems (Zenger, 2018; Stoica Maniu et al., 2023). However, students often encounter difficulties in understanding theoretical concepts, such as transfer functions, root locus, and frequency-response analysis, and in translating these concepts into controller implementations on real-world systems. Similar challenges exist in industrial environments, where technicians and engineers frequently acquire control knowledge primarily through on-the-job experience, with limited opportunities to connect theoretical principles with practical applications (Abramovitch, 2019). This gap between theory and practice is still relevant because control expertise is essential for maintaining process stability, energy efficiency, and product quality in manufacturing systems ranging from motor drives and robotic manipulators to thermal and process-control applications. The growing demand for professionals capable of designing, implementing, and troubleshooting feedback systems emphasizes the need for educational frameworks that bridge academic knowledge and industrial requirements. Consequently, there is a growing need for educational frameworks that combine rigorous conceptual learning with hands-on experimentation and that are applicable to both academic and industrial training contexts (Chernenko and Zemzyulina, 2024).

These considerations motivate the development of learning environments in which control concepts can be explored through a complete engineering workflow rather than as isolated theoretical or experimental activities. Such environments should enable learners to move from process analysis and modeling to controller design, simulation, physical implementation, and evaluation while maintaining a clear connection with the engineering purpose of the controlled process.

Educational research on engineering has consistently shown that learning is more effective when students cycle between concrete experience, reflection, abstract conceptualization, and active experimentation rather than only listening to lectures or solving pen-and-paper exercises (Kolb, 2015; Prince, 2004). The Fiber Extrusion Device (FrED), developed at the Massachusetts Institute of Technology (MIT), provides the concrete experience, while modeling, controller design, and performance analysis support reflection and conceptual understanding. Laboratory experiments and hardware-in-the-loop activities close the loop through active experimentation and make abstract notions such as stability, overshoot, and robustness observable during system operation.

These ideas are aligned with active-learning approaches that have demonstrated significant gains in performance across science and engineering courses (Freeman et al., 2014). Learning factories and smart-manufacturing labs operationalize this experiential perspective by coupling authentic production scenarios with project-based learning and continuous assessment (Abele et al., 2017). In this work, the FrED platform plays a similar role on a smaller scale: it can be viewed as a compact learning factory for advanced manufacturing and control, where learners experience the complete workflow from system identification to controller deployment on a real process.

To align it with industrial practice, the proposal is structured according to the V-model: (1) process analysis and performance specification, (2) modeling and system identification, (3) controller design and simulation, (4) hardware implementation, and (5) results evaluation and discussion. This structure connects theoretical analysis with the physical process and provides learners with an opportunity to examine the complete control-engineering workflow in a form aligned with real industrial systems.

The V-model is a structured engineering-development methodology that organizes the design process by connecting specification and design activities with corresponding implementation, verification, and validation stages. Although originally associated with industrial engineering and system-development processes, the V-model has also been adopted in engineering education and training. For example, Ponce et al. (2022) used the V-model to structure an electric-drive course integrating MATLAB/Simulink, hardware, and rapid prototyping, while Tiberi (2020) mapped control-theory learning activities onto the V-model for the industrial training of embedded-software developers. These studies illustrate the educational value of connecting learning activities with structured engineering-development workflows.

2 Literature review

Control systems education has traditionally emphasized analytical derivations and theoretical concepts, which are mathematically rigorous and essential for advanced control topics. Although theoretical exercises and computer simulations provide important tools for understanding system dynamics and controller design, they do not fully expose learners to the practical constraints encountered when control strategies are implemented on physical systems.

Studies have therefore explored educational approaches that more directly connect control theory with experimentation and engineering practice. In Singh and Singh (2023), the authors propose a multidisciplinary instrumentation and control laboratory designed to develop practical competencies aligned with industrial requirements, reporting improvements in student performance after the introduction of the experimental prototype. Similarly, the authors in Rivadeneira et al. (2025) introduce portable, low-cost electronic modules for state-space control education, enabling students to implement controllers on physical systems rather than relying exclusively on simulation. The study reports improved student performance following the practical interventions. In Arrieta et al. (2019), the authors address the connection with industrial practice by developing a process-control training system based on level and flow regulation using industrial instrumentation, PLCs, simple-loop controllers, and an HMI. Another study Barbosa (2020) demonstrates how an affordable physical platform can connect modeling, controller design, simulation, and experimental implementation within a digital-control course.

These studies show the increasing emphasis on experiential control education and on learning environments that connect theoretical analysis with physical implementation and engineering practice. The following sections examine these developments in greater detail through hardware-based learning platforms, experimental system identification, and manufacturing-oriented control education and industrial workforce training.

2.1 Hardware-based learning platforms for control education

Advances in affordable computing and experimental hardware have expanded the possibilities for incorporating hands-on activities into control-engineering education. Recent studies show that low-cost platforms can support different levels of learning.

In Marquez-Vera et al. (2023), the authors employ an ESP32 microcontroller for signal acquisition, sensor and actuator interfacing, process-control exercises, remote monitoring, and introductory artificial-intelligence applications in biotechnology engineering. The study shows the versatility of low-cost connected hardware for exposing learners to several technologies relevant to modern automation. However, its educational activities are primarily oriented toward microcontroller programming and individual process-control applications rather than toward a structured control-development workflow integrating modeling, identification, controller design, and experimental validation.

At a more introductory level, an Arduino-based RC circuit experiment is used to connect the theoretical step response of first-order systems with real-time data acquisition and visualization (dos Santos and Perdicoulis, 2024). This approach provides an accessible means of making abstract concepts, such as time constants and transient response, experimentally observable. Similarly, Oliveira et al. (2021) uses the portable Temperature Control Laboratory (TCLab) to provide practical activities involving PID control together with feedforward and cascade-control structures. This work shows how low-cost thermal platforms can support the experimental validation of both fundamental and more advanced control concepts.

The authors in Aviles et al. (2023) propose a low-cost, open-source ball-and-beam platform for control-engineering courses, structured to provide students with practical experience in control-system development. Their methodology guides learners through mathematical modeling, experimental system identification, controller design using MATLAB, hardware implementation, and closed-loop validation, thereby connecting the main stages of control-system development within a laboratory project. However, the methodology is centered on a dedicated didactic control platform and is structured for academic control courses. It does not organize the learning process through an engineering V-model or explicitly connect control-development activities with process requirements and performance evaluation within a manufacturing-oriented environment intended to support both control-engineering education and industrial workforce training.

In Badau et al. (2024), the authors address the transition from theoretical controller design to microcontroller implementation by implementing different control techniques using Arduino hardware. Their approach considers practical aspects such as sensor interfacing, real-time constraints, controller implementation, and experimental performance, providing learners with experience in translating theoretical designs into functioning embedded controllers. The work therefore provides an example of bridging theory and implementation. However, its primary emphasis is on implementing previously designed control algorithms rather than integrating the complete sequence from process analysis and experimental system identification to controller design and validation.

The authors in Krauss (2024) report the iterative development of a low-cost inverted-pendulum project over five years of control education. An important contribution of this study is the demonstration that the availability of physical hardware alone does not guarantee an effective learning experience. Modeling accuracy, real-time communication, programming complexity, and the organization of the control-design activity were shown to influence students’ ability to successfully connect control theory with physical implementation. This experience highlights the educational importance of structuring modeling, control design, and hardware implementation as connected activities.

In Navarro-Durán et al. (2023), the authors propose a set of low-cost didactic stations for undergraduate automatic-control courses, designed to emulate practical situations in which an existing physical process must be identified and controlled. Within the proposed active-learning approach, students identify the system dynamics and then design, implement, and evaluate controllers on the physical stations. An important strength of the study is that the educational intervention was assessed using control and experimental groups, with the students using the didactic stations obtaining higher final course averages. The work provides strong evidence that industry-like physical experimentation can reinforce control-engineering competencies. However, the approach is primarily structured around undergraduate control courses and individual didactic stations rather than a V-model-based learning framework that organizes the complete control-engineering workflow, from process requirements to performance evaluation, for both academic education and industrial workforce training.

An experimental methodology for undergraduate digital-control education based on a thermal process is proposed in Serra (2020). The methodology integrates model estimation from experimental data, model-based PI/PID controller design according to predefined performance specifications, simulation, and real-time implementation. This work demonstrates that a sequence connecting experimental modeling and physical controller implementation can provide a comprehensive practical experience in a control course. Its scope, however, is primarily centered on digital-control laboratory activities in an academic setting and on a specific thermal experiment. In contrast, the framework proposed in the present work extends this concept toward a manufacturing-oriented, V-model-based workflow that begins with process requirements and is intended to support both academic education and industrial workforce training.

The authors in Ayvaz et al. (2024) develop an industrial-scale remote-access control laboratory comprising multiple physical experimental systems. The laboratory supports a structured sequence of control-engineering activities, including mathematical modeling, system identification, controller design, and experimental testing, while allowing students to remotely interact with the physical systems through web-based interfaces. The study shows how industrial automation technologies can support hands-on control education and broaden access to experimental facilities. However, the platform is organized around a collection of control laboratory experiments, whereas the present work embeds the control-engineering workflow within a manufacturing process and explicitly connects process requirements, modeling and identification, controller implementation, and performance evaluation through a V-model framework.

In Uyanik and Catalbas (2018), the authors develop a low-cost Arduino-Simulink laboratory around a DC motor in which students perform mathematical modeling and system identification, controller design using root locus, frequency-response analysis, and physical controller implementation. Importantly, the authors also evaluated the instructional effectiveness of the laboratory using student performance data and surveys in course offerings. Their results provide evidence that low-cost physical experimentation can effectively connect theoretical control concepts with laboratory practice. Nevertheless, the platform is designed around a feedback-control course and a single electromechanical setup rather than around a manufacturing process in which multiple interconnected subsystems and process-quality objectives can be addressed through a common engineering workflow.

In Yang and Zhang (2024), the authors introduce a reconfigurable wheel-leg robot that can represent systems of increasing complexity and support both introductory controllers and advanced techniques such as filtering and adaptive control. However, the study primarily demonstrates the technical capabilities of the robotic platform rather than a complete educational workflow. In Sozanski (2023), the authors develop low-cost Arduino-based and STM32-based digital PID modules to provide students with access to real-time control experiments, emphasizing practical issues such as fixed sampling periods and embedded implementation. These systems reduce the cost barrier associated with commercial laboratory equipment but remain principally focused on digital-controller implementation.

Overall, the reviewed studies demonstrate that low-cost physical platforms can successfully expose learners to modeling, dynamic-system analysis, controller design, embedded implementation, and real-time experimentation. Some contributions already integrate some of these stages within a single laboratory experience. However, the literature also shows that most platforms are organized around particular control topics, individual laboratory experiments, or a specific academic course. This observation motivates the investigation of a broader structured workflow in which process requirements, experimental identification, controller design, simulation, physical implementation, and results evaluation are connected within the same manufacturing-oriented environment and can be adapted to both academic and industrial training contexts.

2.2 Experimental system identification in control engineering education

System identification is an important component of model-based control engineering because it provides mathematical representations of physical systems. From an educational perspective, experimental identification gives learners an opportunity to connect mathematical models with observed system behavior while facing practical issues such as measurement noise, model mismatch, and unmodeled dynamics.

In Barbosa (2020), the authors present an Arduino-based thermal platform for a digital-control course in which students experimentally obtain dynamic models before proceeding to control-design activities. Using measured step responses, learners approximate the thermal dynamics using first-order and first-order-plus-dead-time models and can compare different parameter-estimation approaches, including two-point, tangent-based, and least-squares optimization methods. An important strength of this approach is that the identified models are subsequently compared with experimental responses and used in control activities, allowing learners to observe the relationship between model quality and controller behavior. However, the proposed activities are primarily structured around a specific academic digital-control course and a dedicated thermal educational platform.

An instructional framework specifically focused on experimental process identification using first-order-plus-time-delay models is presented in Aliane (2024). The approach guides students through data acquisition and preprocessing, application and comparison of different identification methods, and model validation using experimental data from a physical temperature-control system. The learning experience combines simulation-based prelaboratory activities with hands-on experimentation and includes assessment of student learning outcomes and feedback collected over three academic years. This work is centered on the identification stage rather than integrating these activities into a broader control-engineering workflow.

The use of the Arduino-based Temperature Control Laboratory (TCLab) across different undergraduate and graduate control courses is reported in Guzmán et al. (2022). The platform supports open-loop and closed-loop experimentation, model calibration using least-squares estimation, and subsequent control design activities. Student feedback indicated that working with the platform was challenging but motivating and that it provided a suitable environment for understanding control-engineering concepts. Although the platform supports identification and control activities across different academic levels, its reported implementation remains primarily oriented toward university curricula.

The study in Serra (2020) integrates model estimation from experimental data with model-based digital-controller design and real-time implementation on a thermal process, while Uyanik and Catalbas (2018) incorporates mathematical modeling and system identification of a DC motor into a low-cost Arduino-Simulink laboratory followed by controller-design and experimental activities. These contributions demonstrate that experimental system identification can effectively serve as a connection between mathematical modeling and physical controller implementation rather than as an isolated theoretical topic.

The reviewed studies show that system identification can be successfully incorporated into hands-on control education and connected with subsequent control design and implementation activities. Therefore, the remaining opportunity is not simply to introduce experimental identification into control education, but to embed it as one stage of a broader and explicitly structured engineering workflow. In the framework proposed in this work, identification is linked to process analysis and requirements specification and to subsequent controller design, simulation, physical implementation, and experimental evaluation within the same manufacturing-oriented platform. This structure also extends the intended application beyond conventional academic laboratory courses toward industrial control-training activities.

2.3 Industrial workforce training in control engineering

The transition toward digital and advanced manufacturing requires educational and training approaches that develop competencies aligned with the practical needs of industrial environments. The study in Rapitsenyane et al. (2023) examines the relationship between digital-manufacturing skills developed in higher education and those required by small and medium-sized manufacturing enterprises. Their results showed general alignment between educational and industrial needs, but also indicated that exposure to several digital-manufacturing technologies remained predominantly at an awareness level rather than at the level of practical competency. This shows the importance of learning environments in which participants can actively apply engineering manufacturing concepts.

The role of learning factories in developing competencies required for the transition toward Industry 5.0 is examined in Lagorio and Cimini (2024). Using a physical smart-manufacturing learning factory, the authors identified learning contents across several industrial-engineering areas and associated them with technical, methodological, personal, and social competencies. The study highlights the potential of learning factories to bridge academic education and industrial needs by providing realistic manufacturing environments in which learners can interact with industrial technologies and processes. The proposed framework focuses on the identification of learning contents and competencies.

In Quinn et al. (2025), the authors present a learning factory developed around three complementary pillars: education, industry engagement, and research. It provides students with hands-on experience in an industry-relevant manufacturing environment while also supporting workforce training developed in collaboration with industrial partners. It demonstrates the value of using a common manufacturing infrastructure to connect academic education with industrial upskilling. However, its scope is primarily centered on Industry 4.0 technologies and systems integration rather than on a structured control-engineering learning workflow.

In Chelini and Richert (2023), the authors propose a project-based learning approach using a cyber-physical learning factory to develop engineering-design skills for Industry 4.0 environments. Rather than training students on isolated technologies, learners design and implement the control and automation infrastructure of a manufacturing system, including PLC-level control, industrial communication, and manufacturing execution system functions. The approach also incorporates engineering requirements, design documentation, and factory-acceptance activities, providing a realistic manufacturing context for developing transferable skills relevant to industrial practice. However, its workflow is primarily oriented toward automation-system and Industry 4.0 engineering rather than the complete control-engineering workflow.

A relevant contribution to industrial control training is presented in Tiberi (2020), where a control-theory course was specifically developed for embedded-software developers in industry. Rather than replacing an established industrial development process, the proposed approach maps control-theory activities onto the V-model already familiar to the trainees, with individual stages of the control-development workflow represented as educational modules. The effectiveness of the course was evaluated through questionnaires and changes in software-delivery quality. This work demonstrates the value of aligning control education with workflows already used in industry. However, its implementation is oriented toward embedded-software development and conceptual control-theory training rather than toward the execution of a complete control workflow on a physical manufacturing process.

Manufacturing-oriented engineering education has also adopted structured development methodologies to expose students to problems resembling professional practice. In Ponce et al. (2022), the authors present an electric-drive course organized using the V-model, MATLAB/Simulink, low-cost hardware, and rapid prototyping. Students progress from fundamental electric-machine concepts toward the development of an electric-vehicle control prototype, providing a structured connection between theoretical knowledge, simulation, and implementation. The approach demonstrates how industrially motivated engineering problems can be incorporated into undergraduate education. Nevertheless, its scope is centered on an academic electric-drives course and product-development activities rather than on a reusable workflow intended for both academic learners and industrial personnel.

The study in Bradley et al. (2025) employs the FrED platform as the context for developing low-code/no-code digital-manufacturing systems. The proposed activities enable students to design manufacturing-execution applications such as machine monitoring, inventory tracking, and digital work instructions while reducing the software-development burden. This study shows the educational value of FrED as a realistic manufacturing environment, although its emphasis is on digital manufacturing and manufacturing-execution systems rather than on dynamic modeling and feedback-control design.

In a complementary direction, Ponce et al. (2025) uses the FrED spooler subsystem to investigate PID tuning according to industrial operating scenarios. The proposed method relates control-performance priorities to industrial requirements and illustrates the use of FrED as a testbed for manufacturing-oriented control design. However, the study addresses controller optimization and simulation rather than the organization or evaluation of a complete control-engineering learning process.

These studies show two complementary directions in the literature. Industrial training approaches demonstrate the importance of aligning control education with established engineering workflows, while manufacturing-oriented educational platforms provide realistic contexts in which theoretical concepts can be connected with practical engineering problems. Previous FrED studies further demonstrate the platform’s suitability for both digital-manufacturing education and industrially motivated control research. Nevertheless, these contributions address different portions of the overall process. This motivates a framework that combines these perspectives by structuring process analysis and requirements specification, experimental system identification, controller design, simulation, physical implementation, and results evaluation within a common manufacturing-oriented workflow intended for both academic education and industrial workforce training.

Table 1 summarizes the representative studies discussed in the previous sections and compares their scope with that of the present work. The comparison considers whether each contribution incorporates system identification, controller design, physical hardware implementation, and an explicit orientation toward industrial or manufacturing practice. The terms Yes, Limited, and No indicate whether each aspect is explicitly incorporated, only partially addressed, or not considered in the reported implementation, respectively. The final column highlights the main scope of each study and its distinction from the framework proposed in this work.

Table 1. Comparison of representative studies in control engineering education and training.

Throughout this paper, the term learners is used to collectively refer to students, technicians, and engineers engaged in the proposed educational and industrial training activities. This terminology emphasizes the applicability of the framework across both academic and workforce-development contexts.

2.4 Research gap and contribution

The literature reviewed in the previous sections shows that physical platforms can effectively support hands-on control education through activities involving modeling, experimental system identification, controller design, simulation, and hardware implementation. Some studies have already integrated some stages of this process within academic laboratory activities, while other contributions have shown how structured development approaches, including the V-model, can be used to align control education with academic or industrial engineering practice. Previous FrED-based studies have also demonstrated the potential of the platform for digital-manufacturing education and for control design under industrially motivated scenarios.

However, these contributions address different portions of the control engineering and training process. The reviewed literature does not provide a manufacturing-oriented framework that explicitly organizes process analysis and requirements specification, experimental system identification, controller design, simulation, physical implementation, and experimental evaluation as a single V-model-based workflow intended for both academic education and industrial workforce training. This represents the research gap addressed in the present work.

Accordingly, the following research question is formulated:

How can a V-model-based learning framework be implemented on the FrED platform to integrate the complete control-engineering workflow within a manufacturing-oriented environment for control engineering education and industrial workforce training?

To address this research question, this work proposes an educational and industrial training framework implemented on the MIT FrED platform. The main contribution is a structured learning framework that integrates the different stages of the control-engineering workflow within a single manufacturing environment, connecting process requirements and mathematical modeling with experimental identification and validation, control design, simulation, physical implementation, and performance evaluation. The FrED spooler subsystem is used solely as a representative case study to demonstrate the execution of the proposed workflow; however, the framework is designed to be extended to other FrED subsystems and adapted to comparable physical platforms and engineering processes.

3 Methodology

3.1 Fiber extrusion device (FrED) and proposed framework

The FrED platform developed at MIT is a versatile platform for control education. Its design emulates an advanced manufacturing process, such as optical-fiber fabrication, and its subsystems are oriented toward this purpose. The FrED consists of the following modules:

A graphical user interface running on the Raspberry Pi allows learners to monitor process variables and interact with the control system during the educational activities. Figure 1 shows the FrED platform and its modules.

• Extrusion subsystem: This subsystem incorporates a stepper motor to feed material into the system and a hot-end to heat and melt the material. The hot-end includes a thermistor to measure temperature, which is a relevant process variable for regulating fiber diameter.
• Cooler subsystem: It consists of two fans that cool the material according to the desired process conditions.
• Diameter measurement: This module uses a camera to measure the diameter of the produced fiber. The measured diameter can be used as feedback for closed-loop regulation.
• Spooler subsystem: It contains a DC motor and gearbox used to wind the produced fiber and generate the tensile force required during the fiber-drawing process. This force influences the resulting fiber diameter.
• Microcontroller: A Raspberry Pi 4 and a custom PCB are used to control the FrED platform. Python programs executed on the Raspberry Pi implement the control and data-acquisition routines. The PCB provides the power interface between the Raspberry Pi and the actuators and also acquires sensor signals for transmission to the Raspberry Pi.

With this structure, learners interact not only with an isolated subsystem, but with a complete manufacturing system that includes tension control, diameter monitoring, and thermal regulation. Although the spooler subsystem is used as the representative case study in this work, all FrED subsystems play important and interconnected roles in the manufacturing process, and each can be incorporated into the proposed educational framework depending on the learning objectives and control topics being addressed.

The overall workflow follows the V-model structure shown in Figure 2, which represents the development cycle from process analysis to results evaluation across the FrED subsystems.

• Process analysis and requirements specification: definition of quality and performance targets in fiber diameter, extrusion temperature or spooler speed.
• Modeling, data acquisition and system identification: obtaining experimental data, deriving the mathematical models and estimating parameters of the subsystem.
• Control design and simulation: synthesis of the control strategies and simulation using MATLAB.
• Hardware implementation: deployment of the controller to the real system.
• Results evaluation and discussion: comparison of simulation and experimental outcomes, interpretation of system behavior, and reflection on engineering decisions.

Each stage of the V-model is associated with a concrete educational and experimental activity. During the analysis and requirements phase, students define the control objectives related to fiber manufacturing. In the identification phase, data from sensors are used to build and validate models of the corresponding FrED subsystem. The control design phase involves tuning and simulation in MATLAB or Simulink, followed by hardware implementation using the Raspberry Pi control system. Finally, in the evaluation phase, performance metrics such as IAE, overshoot, and control effort are computed and interpreted in terms of industrial relevance. This alignment allows learners to go through the entire control engineering cycle, from theoretical analysis to experimental validation, within a single educational framework.

MATLAB and Python were chosen because they are widely used in academia and industry for modeling, analysis, and control implementation. MATLAB provides specialized toolboxes for system identification and control design, while Python offers flexibility, open-source accessibility, and good integration with hardware platforms such as the Raspberry Pi.

The methodological approach of this work is based on the development of a learning framework and its demonstration through an experimental case study using the FrED spooler subsystem. Simulation is not used as the sole methodological approach, but rather as an intermediate stage of the proposed control-engineering workflow. Experimental demonstration includes physical data acquisition, system identification and model validation, hardware implementation of the designed controllers, and comparison between simulated and experimental system responses.

This workflow is defined at the platform level and can be applied to any FrED subsystem. Therefore, the spooler subsystem case presented in this paper should be interpreted as one representative application of the proposed methodology rather than as its sole intended use.

Figure 2 illustrates the conceptual structure of the proposed educational and industrial training framework and its practical implementation using the MIT FrED platform. The V-model organizes the control-engineering learning process into sequential and interconnected phases that reflect an engineering development cycle: process analysis and requirements specification, modeling and system identification, controller design and simulation, hardware implementation, and results evaluation. At this level, the figure establishes the pedagogical structure of the framework, emphasizing that control education should not be reduced to isolated controller-tuning exercises but should also guide learners through the complete engineering workflow from problem definition to experimental validation. Each phase is explicitly associated with concrete learner activities, including data acquisition, least-squares identification, transfer-function estimation, controller tuning, implementation, and performance analysis using industrially relevant metrics. Thus, the V-model is translated from an engineering-development abstraction into a sequence of activities that can be implemented experimentally on the FrED platform. In this way, Figure 2 illustrates both the overall educational and industrial workflow and the manner in which each stage can be practically implemented on the FrED platform. Together, these elements show that the proposed framework is not merely theoretical but is intentionally designed to connect conceptual analysis with practical development. This representation provides a basis for structuring engineering-education activities in which learners can first examine the general engineering process and then progressively engage with each stage through guided experimentation. From an industrial workforce-development perspective, it also provides a way to align educational activities with structured processes used in real manufacturing environments.

The modular architecture and workflow of the proposed framework allow it to be easily extended to new control strategies in any of the FrED subsystems. Each subsystem can be modeled, identified, and controlled following the same methodological steps: data acquisition, model estimation, controller design, simulation, and validation, thereby maintaining methodological consistency across different implementations. The modular and reusable structure of the proposed framework allows it to evolve into a broader educational and research environment supporting interdisciplinary activities in control, automation, and advanced manufacturing. Within this structure, the framework represents the main contribution of this work, whereas the spooler subsystem serves as a representative case demonstrating how the methodology can be extended to other FrED subsystems.

• Hot-end temperature regulation: This extension focuses on the extrusion subsystem, where a heater and thermistor are used to regulate the temperature profile required for consistent fiber formation and extrusion. The same modeling and control framework can be applied by first identifying the thermal dynamics and then designing and tuning a controller to minimize temperature oscillations and energy consumption. This enables students and engineers to explore time-delay systems and energy efficiency optimization in thermal systems, concepts that are central to modern manufacturing control.
• Optical fiber diameter control: In this case, a vision system measures the diameter of the extruded fiber. The control objective is to maintain the target diameter by adjusting the spooler speed, hot-end temperature, or extrusion rate. By reusing the same workflow, users can perform image-based data acquisition, identify the dynamic relationship between pulling velocity and diameter variation, and design a feedback controller. This implementation introduces learners to advanced sensing and process-quality control techniques relevant to precision manufacturing and industrial scenarios.
• Cascade diameter control: This more advanced configuration integrates mechanical and optical subsystems in a cascade architecture. The inner loop regulates fiber tension through spooler-velocity control, whereas the outer loop regulates fiber diameter by adjusting the hot-end temperature. The framework supports this multi-loop design by combining the previously identified models, allowing learners to analyze interactions between subsystems and evaluate control strategies typical of industrial fiber drawing towers.

Although the proposed framework is intended to be applicable to any FrED subsystem, the spooler subsystem is used in this work as a representative case study. Therefore, this component is described in greater detail below.

The spooler subsystem collects and winds the extruded fiber while simultaneously generating the tensile force required for the fiber-drawing process. By varying the spooler’s rotational velocity, the system directly adjusts the tensile force applied to the fiber as it exits the extrusion subsystem. This tensile force is a key process variable that determines the fiber’s final diameter: higher tension leads to a smaller fiber diameter due to elongation, whereas lower tension results in a larger diameter because the material undergoes less stretching.

From an educational point of view, the design of the spooler subsystem provides a clear physical demonstration of feedback-control principles within a manufacturing context. The subsystem illustrates the integration of electromechanical actuation, sensor feedback, and control algorithms, allowing learners to observe how mechanical tension interacts with material flow and temperature to influence final product quality. By adjusting control parameters and monitoring the resulting motor response, learners can engage in hands-on control implementation while exploring the connection between theoretical concepts of system dynamics and a subsystem that directly influences the fiber-drawing process.

For the educational activities, the spooler DC motor is controlled through Python scripts executed on the Raspberry Pi, allowing learners to modify inputs, collect experimental data, and implement control algorithms during system operation. PWM signals are used to regulate motor speed. Figure 3 illustrates the spooler subsystem of the FrED platform.

With this platform, learners can address practical challenges associated with implementing control concepts in real-world systems, including encoder quantization, signal noise, and mechanical backlash. In addition, the Python scripts can be readily modified to accommodate different experimental tests and control strategies.

By providing direct access to a physical plant, measurable system dynamics, and low-cost hardware, the FrED platform helps connect abstract control-theory concepts with tangible engineering applications. In addition, the relatively low voltage and torque levels facilitate its use in educational laboratory environments.

3.2 Fiber diameter and motor dynamics relationship

The fiber diameter in the extrusion process is directly affected by the mechanical tension applied to the filament, which is generated by the DC motor that controls the spooler speed. When the motor increases its angular speed, the pulling tension on the extruded fiber rises, leading to a reduction in fiber diameter due to material elongation and conservation of mass flow. Conversely, a decrease in motor speed reduces the pulling tension, resulting in a larger fiber diameter. This relationship connects the electrical and mechanical dynamics of the motor with a physical characteristic of the manufactured fiber, providing a relevant scenario for examining the interaction among control input, system response, and process outcome.

From an engineering-education perspective, this relationship provides students, engineers, and technicians with an opportunity to connect theoretical control models with manufacturing-process behavior. By comparing different control strategies and fiber-diameter dynamics, learners can explore how subsystem-level control performance is related to quality metrics such as uniformity and fiber diameter. This connection provides an opportunity to relate abstract control-theory concepts to the practical interactions among motor dynamics, tension regulation, and thermal conditions in the fiber-drawing process.

The relationship between spooler speed and fiber diameter provides an intuitive way for learners to connect control-system behavior with manufacturing quality. By analyzing how changes in motor actuation can affect the resulting product characteristics, learners can examine how feedback-control decisions influence industrial processes. This experience helps bridge theoretical concepts such as system dynamics and controller tuning with practical considerations related to product quality, process stability, and manufacturing performance. Deviations in the speed profile of the spooler translate into fluctuations in diameter, which may cause fibers to fall outside specification limits or require rework and additional inspection effort. By explicitly relating speed control performance to diameter variability, the FrED platform illustrates how control performance at the subsystem level can propagate to a quality-related variable of the manufactured product. This type of connection between engineering activities and realistic manufacturing processes is consistent with the learning-factory approach, in which practical production environments are used to support engineering education and competence development (Abele et al., 2017; Zhang et al., 2019).

It should be emphasized that the experimental case study presented in this work is limited to closed-loop velocity control of the spooler subsystem. Fiber diameter is discussed here to establish the manufacturing relevance of spooler dynamics and to illustrate how this subsystem can subsequently be integrated into broader FrED control applications; closed-loop diameter regulation using the vision system is not implemented or evaluated in the present study. Extending the framework to vision-based diameter control would additionally require consideration of the camera sampling rate, image-processing time, and communication latency on the Raspberry Pi, since these factors would determine the effective feedback rate of the outer diameter-control loop. Such an implementation is beyond the scope of the representative spooler experiment considered here and constitutes a natural extension of the proposed framework.

3.3 Educational and industrial training framework

FrED was developed as a learning and research platform that bridges advanced manufacturing and control education. It provides a modular environment in which learners, including students, technicians, and engineers, can explore modeling, system identification, controller design, and experimental validation within realistic manufacturing contexts.

The framework is designed to be deployed in both academic courses and industrial training programs. In a typical academic setting, it could be implemented in an intermediate control-systems course for undergraduate students in electrical, mechanical, or mechatronics engineering. Class sizes usually range from 20 to 30 students, who have prior exposure to basic signals and systems concepts and introductory programming. The FrED-based modules could be delivered as 4–6 laboratory sessions of 2–3 h each, covering modeling, system identification, controller design, and hardware-in-the-loop validation.

The same workflow can be adapted for reskilling or upskilling technicians and practicing engineers in industrial contexts. In this case, the activities are organized around realistic process specifications and constraints provided by the plant, and the emphasis is placed on translating formal control concepts into actionable tuning and troubleshooting skills. Positioning the framework simultaneously as an academic and workforce-training tool supports the broader objective of preparing the industrial workforce for digital and data-driven manufacturing systems (Kaur et al., 2023).

The proposed framework is organized as a sequence of interconnected learning activities that guide learners through the complete control-engineering workflow, from system identification to experimental validation.

• Process analysis and requirements specification: analysis of the manufacturing process and definition of performance specifications, quality requirements, and operational constraints that will guide the control-system development workflow.
• Modeling and system identification: Learners obtain a dynamic model of the subsystem, estimate its parameters from experimental data, and validate the resulting model.
• Controller design and tuning: Using the identified model, learners design and tune controllers according to performance specifications, such as settling time and maximum overshoot.
• Simulation and validation: Learners evaluate controller performance in simulation and compare the expected behavior with theoretical predictions.
• Hardware implementation: The designed controllers are implemented on the FrED platform, enabling experimentation and observation of control performance on a physical system.
• Experimental comparison and discussion: Learners compare simulated and experimental responses, analyze performance metrics, and discuss discrepancies arising from practical engineering constraints.

3.3.1 Modeling and system identification using the DC motor

The dynamic behavior of DC motors is commonly represented using first-order and second-order models, both of which are widely available in standard control systems textbooks and are typically covered in undergraduate control courses (Ogata, 2010; Dorf and Bishop, 2021; Kuczmann, 2024; Sami et al., 2021). For this reason, the complete derivation of these models is omitted in this work. Learners are expected to be familiar with the underlying modeling procedure from prior theoretical instruction or may obtain it directly from the literature. Within the proposed framework, these models serve as educational tools that allow learners to investigate the relationship between model complexity, identification accuracy, and controller performance. The first-order model provides a simple representation suitable for introductory analysis and controller design, while the second-order model captures additional system dynamics and enables a more detailed evaluation of control strategies. The resulting transfer functions are therefore used as the starting point for the system identification, simulation, controller design, and experimental validation activities described in the following sections.

The first-order model is represented by the transfer function:where ω(s) is the rotor angular speed in rad/s, v(s) is the applied voltage, km is the motor gain, and τ is the system time constant, with km>0 and τ>0.

ω(s)v(s)=kmτs+1,(1)

The second-order model is represented by:where ω(s) is the rotor angular speed, v(s) is the applied voltage, J is the rotor moment of inertia, B is the viscous-friction coefficient, Ke is the back-EMF constant, L is the armature inductance, R is the armature resistance, and Kt is the torque constant. Dividing all terms by JL yields the normalized form:where:The first-order and second-order model representations of the DC motor provide learners with models of different complexity levels that can be used throughout the proposed educational workflow. Although the derivation of these models is not presented in this paper because it is widely available in standard control systems textbooks and commonly covered in undergraduate control courses, understanding the underlying modeling process remains an important learning objective in control education. Relating physical principles to mathematical models provides learners with an opportunity to explore system dynamics and examine how model parameters influence transient and steady-state behavior. Within the proposed framework, the first-order and second-order models allow learners to investigate the relationship between model complexity, identification accuracy, and controller performance, while providing the basis for system identification, simulation, controller design, and experimental validation activities. In industrial settings, these skills support the effective implementation, tuning, and troubleshooting of control systems in manufacturing and automation processes.

G(s)=ω(s)v(s)=Kt(JL)s2+(JR+BL)s+(BR+KtKe),

G(s)=b0s2+a1s+a0,(2)

b0=KtJL,a1=JR+BLJL,a0=BR+KtKeJL.

The sampling interval is an important consideration in both system identification and digital control implementation because it must be sufficiently short relative to the relevant dynamics of the system. In the spooler subsystem, a nominal sampling time of Ts=20 ms, corresponding to a sampling frequency of 50 Hz, was selected. This interval is considerably shorter than the dominant dynamics of the identified motor models. The first-order model in Equation 4 has a time constant of τ=0.1477 s, corresponding to approximately 7.4 samples per time constant. Similarly, the dominant pole of the second-order model in Equation 5 is approximately −7.52s−1, corresponding to a dominant time constant of approximately 0.133 s and about 6.7 samples per time constant. Therefore, the selected sampling interval provides sufficient temporal resolution for the dominant spooler dynamics considered in the present educational experiments. The Raspberry Pi implementation uses Python under a standard Linux operating system and is therefore not intended to provide hard real-time timing guarantees. Operating-system scheduling may introduce timing jitter. However, for the reported spooler experiments and the identified dominant dynamics considered in this study, the nominal sampling interval of Ts=20 ms provided sufficient temporal resolution for data acquisition, system identification, and controller implementation. This conclusion is specific to the experimental conditions evaluated in the present work and should not be interpreted as a general timing guarantee for other FrED configurations or faster system dynamics.

3.3.2 Learning system identification through experimental data

As the system-identification stage of the proposed learning framework, learners experimentally characterize the selected FrED subsystem using measured input-output data. In the representative spooler case study, learners excite the DC motor using different input-signal profiles, such as step and ramp signals, which can be used to sufficiently excite the system dynamics. The corresponding motor-speed response is then recorded, and the resulting experimental data are used to estimate and validate mathematical models for subsequent controller design. In this work, two models of the system are considered, a first-order and a second-order model, which were previously introduced in Equations 1, 2.

In the proposed framework, the least-squares method is used to identify the system parameters. This method is widely employed in the system identification of industrial processes to estimate unknown parameters of mathematical models from experimental data (Ljung, 1999). The objective of this activity is to develop a mathematical representation of the subsystem from experimental data, allowing learners to examine how models can be used to describe and predict the behavior of real engineering systems.

For educational purposes, the identification activity employs simple ARX (AutoRegressive with eXogenous Input) models, which provide a transparent relationship between measured input–output data and the estimated model parameters. First-order and second-order discrete-time representations are considered, allowing learners to investigate how model complexity influences identification accuracy and subsequent controller design. The continuous-time models introduced in Equations 1, 2 are converted into the corresponding ARX structures using a backward-difference approximation. This discretization is intentionally selected because of its simplicity and because it allows learners to directly relate the continuous-time model parameters to the coefficients of the discrete-time identification model. Other discretization methods may also be employed within the proposed framework according to the learning objectives, sampling conditions, and required model accuracy. Therefore, the procedure presented here should be interpreted as a representative implementation rather than as a required discretization method.

For the first-order model in Equation 1, the corresponding time-domain equation is:Using the backward-difference approximation:and rearranging the resulting expression gives:which has the first-order ARX structure used in the following identification procedure.

τω˙(t)+ω(t)=kmu(t).

ω˙(t)≈ω[k]−ω[k−1]Ts,

ω[k]=ττ+Tsω[k−1]+kmTsτ+Tsu[k],

Similarly, the normalized second-order model in Equation 2 is represented in the time domain as:In this case, the first and second derivatives are approximated as:After substitution and rearrangement, the resulting discrete-time model is:which provides the second-order ARX structure employed in the identification activity. In these models, y[k]≡ω[k] denotes the measured motor speed at sample k, u[k] represents the control input, θ is the vector of unknown parameters to be estimated, and Φ is the regressor vector constructed from previously measured input and output samples. The first-order and second-order ARX structures considered in this work are given by: ?>whereThe corresponding second-order ARX structure is:which can be written in regression form as:whereThese model structures are intentionally selected because of their simplicity and interpretability. Learners can directly relate the estimated parameters to the observed system behavior and evaluate how different model structures affect simulation accuracy and controller performance. The use of both first-order and second-order models also allows learners to investigate the trade-off between model simplicity and model fidelity. More advanced identification techniques may be incorporated as extensions once the fundamentals of data-driven modeling have been understood.

ω¨(t)+a1ω˙(t)+a0ω(t)=b0u(t).

ω˙(t)≈ω[k]−ω[k−1]Ts,ω¨(t)≈ω[k]−2ω[k−1]+ω[k−2]Ts2.

ω[k]=2+a1Ts1+a1Ts+a0Ts2ω[k−1]−11+a1Ts+a0Ts2ω[k−2]+b0Ts21+a1Ts+a0Ts2u[k],

y[k]=θ1y[k−1]+θ2u[k],

y[k]=[y[k−1]u[k]]θ,

θT=[θ1θ2],Φ=[y[k−1]u[k]].

y[k]=θ1y[k−1]+θ2y[k−2]+θ3u[k],

y[k]=[y[k−1]y[k−2]u[k]]θ,

θT=[θ1θ2θ3],Φ=[y[k−1]y[k−2]u[k]].

The unknown parameter vector is estimated using the least-squares method, which determines the parameter values that best fit the experimental input–output data according toThis simple ARX structure provides an explicit, transparent link between the sampled input and output data and the coefficients of a discrete-time model. Students can directly see how each term in the regression vector contributes to the predicted output and how parameter estimates change when the experimental conditions or data quality vary (Ljung, 1999). This transparency is very valuable when introducing system identification to learners with a limited mathematical background.

θ^=(ΦTΦ)−1ΦTY.(3)

More advanced identification techniques like instrumental variables, subspace identification, or prediction error methods can be introduced as extensions once learners are comfortable with the basic least-squares method. In this way, the FrED-based activities strike a balance between conceptual clarity and realism; they expose learners to issues such as noise, aliasing, and unmodelled dynamics while keeping the identification algorithms simple enough to implement and interpret during a laboratory session.

Experimental data are collected by exciting the motor with input signals such as steps, ramps, or combinations of both. The resulting input-output measurements are used to estimate the parameters of the first-order and second-order ARX models using the least-squares method. To assess model fidelity, the identified models are validated using an independent dataset not employed during parameter estimation. Figure 4 presents a representative validation example, allowing learners to compare the measured response with the model predictions, evaluate the impact of model complexity on approximation accuracy, and examine the importance of model validation before controller implementation.

After estimating the ARX coefficients through Equation 3, the continuous-time model parameters can be recovered by inverting the relationships introduced during discretization. For the first-order model, the estimated coefficients satisfy:For the second-order ARX model, with the sign of the second autoregressive coefficient included in θ2, the corresponding continuous-time parameters are recovered as:These parameters are then substituted into the continuous-time model structures of Equations 1, 2, providing the transfer functions used for subsequent controller design. This procedure explicitly connects experimental data acquisition, discrete-time ARX identification, continuous-time model recovery, and controller design while retaining a mathematically accessible workflow for learners.

τ=Tsθ11−θ1,km=θ21−θ1.

a1=−θ1+2θ2θ2Ts,a0=−1−θ1−θ2θ2Ts2,b0=−θ3θ2Ts2.

For this representative example, the identified first-order transfer function is:whereas the identified second-order transfer function is:This procedure provides learners with an opportunity to examine how measurement noise, encoder quantization, and unmodeled dynamics affect the quality of experimental data and the resulting model estimates. During the data-acquisition and system-identification activities, motor speed is obtained from the incremental information provided by the encoder; therefore, its finite resolution and the finite sampling interval can produce discrete variations in the estimated speed, especially at lower rotational velocities. Learners are encouraged to inspect both raw and preprocessed signals and, when necessary, apply simple filtering techniques such as moving-average or low-pass filtering. By comparing raw and filtered measurements, they can observe the trade-off between noise attenuation and signal responsiveness and recognize that excessive filtering may introduce additional delay or suppress relevant transient information. Students then use the experimental data to construct the vectors Φ and Y, estimate θ^ for the first and second order models, fit the resulting models to the acquired data, and validate them using different experimental responses from the FrED spooler subsystem. This workflow connects practical data acquisition and preprocessing with system identification and model validation while providing opportunities to practice activities relevant to industrial control applications.

Gf(s)=0.49540.1477s+1,(4)

Gs(s)=1731.3048s2+472.6205s+3495.7927.(5)

3.3.3 Controller design as a learning activity

Despite the availability of more advanced control strategies, PI and PID controllers are still the most widely used controllers in industrial practice due to their simplicity, interpretability, and effectiveness across a broad range of applications. Consequently, the ability to tune, implement, and evaluate these controllers remains an important learning objective in control engineering education and workforce training. Within the proposed framework, learners use the identified models to design PI and PID controllers according to performance specifications such as settling time and maximum overshoot. This activity allows them to connect theoretical concepts introduced in control courses with practical implementation and performance evaluation on a real manufacturing-oriented platform.

The MIT FrED platform supports this learning process by enabling learners to apply controller-design concepts to a real manufacturing-oriented system. Once a model has been obtained through the identification activity, learners can design and tune controllers according to performance specifications such as maximum overshoot (Mp) and settling time (Tset). By comparing simulation and experimental results, learners can investigate the relationship between modeling accuracy, controller tuning, and system performance. In this way, FrED serves as an integrative learning environment that connects theoretical concepts with practical implementation and provides opportunities to practice activities relevant to industrial process control.

Characteristic-polynomial matching is adopted in this work as the controller-design method because it provides a systematic and transparent procedure for translating desired closed-loop performance specifications into controller parameters. By relating quantities such as maximum overshoot (Mp) and settling time (Tset) to the desired damping ratio and natural frequency, learners can define the desired closed-loop characteristic polynomial and obtain the corresponding controller gains through coefficient matching. Within the proposed framework, this approach provides an accessible connection between system modeling, closed-loop performance specifications, controller design, simulation, and experimental implementation. The theoretical foundations of classical controller design based on desired closed-loop pole locations and characteristic-equation coefficient matching are extensively covered in standard control systems textbooks (Ogata, 2010; Dorf and Bishop, 2021; de Oliveira, 2017; Chen, 2013). Therefore, the emphasis here is placed on the main design steps required to obtain the controller gains and on their integration within the proposed experimental learning workflow.

Once the desired maximum overshoot (Mp) and settling time (Tset) have been defined, learners can translate these performance specifications into controller parameters. The corresponding damping ratio (ζ) and natural frequency (ωn) are obtained using classical second-order system relationships. These quantities serve as the basis for calculating PI and PID controller gains for the identified first-order and second-order models.

ζ=−ln(Mp)π2+ln2(Mp),

ωn≈4ζTset.

The main steps required to obtain the controller tuning expressions are presented here to provide a reproducible design procedure. A more detailed theoretical treatment of characteristic-polynomial-based controller design and classical PID and PI tuning can be found in standard control systems textbooks and is commonly covered in undergraduate control courses (Ogata, 2010; Dorf and Bishop, 2021; de Oliveira, 2017; Chen, 2013). Within the proposed framework, the emphasis is placed on connecting performance specifications, controller tuning, simulation, and experimental validation rather than on the mathematical derivation itself.

3.3.3.1 Controller design using the second-order model

For the second-order model, learners can obtain PI and PID controller gains directly from the desired performance specifications. The resulting tuning expressions are presented below and are subsequently used for simulation and experimental implementation on the FrED platform.

For the second-order model, the PID and PI controller gains are obtained by matching the closed-loop characteristic polynomial with a desired third-order characteristic polynomial comprising the dominant second-order dynamics and an additional real pole.

For the second-order model, the PID controller is defined as:Considering the normalized second-order plant in Equation 2, the closed-loop characteristic polynomial under unity feedback is:Since the closed-loop characteristic polynomial is of third order, the desired dominant second-order dynamics defined by ζ and ωn are complemented with an additional pole p3. Therefore, the desired characteristic polynomial is defined as:Expanding this expression gives:Matching the coefficients of Δ(s) and Δd(s) yields:andTherefore, the PID gains are obtained as:andFor the second-order PID design, the additional pole was selected as p3=a1+50=522.6205s−1, corresponding to a closed-loop pole located at s=−522.6205. This pole was intentionally placed far to the left of the desired dominant complex-conjugate poles so that the associated dynamics are considerably faster and have a negligible influence on the dominant transient response.

C(s)=Kp+Kis+Kds.(6)

Δ(s)=s3+(a1+b0Kd)s2+(a0+b0Kp)s+b0Ki.

Δd(s)=(s2+2ζωns+ωn2)(s+p3).

Δd(s)=s3+(2ζωn+p3)s2+(ωn2+2ζωnp3)s+ωn2p3.

a1+b0Kd=2ζωn+p3,

a0+b0Kp=ωn2+2ζωnp3,

b0Ki=ωn2p3.

Kd=2ζωn+p3−a1b0,

Kp=ωn2+2ζωnp3−a0b0,

Ki=ωn2p3b0.

After obtaining the controller gains, learners can evaluate the expected closed-loop behavior through simulation and subsequently implement the controller on the FrED platform. Comparing simulated and experimental responses provides an opportunity to examine the effects of modeling assumptions, parameter-estimation accuracy, and practical implementation constraints.

Figure 5 shows a simulated response obtained with the designed PID controller. The simulation stage allows learners to verify that the controller satisfies the desired performance specifications and provides a reference for subsequent comparison with the experimental response. Through this process, learners can analyze the relationship between model predictions and real-system behavior.

Figure 6 presents an experimental response obtained after implementing the designed PID controller on the FrED platform. This stage of the framework allows learners to transition from simulation to experimentation and evaluate how closely the physical system follows the predicted behavior. The comparison between simulated and experimental responses provides an opportunity to analyze model accuracy, controller performance, and real-world implementation effects.

For the second-order model, the PI controller is defined as:Considering the normalized second-order plant in Equation 2, the closed-loop characteristic polynomial under unity feedback is:As in the PID design, the desired dominant second-order dynamics are defined by the damping ratio ζ and natural frequency ωn, while the third-order characteristic polynomial requires an additional pole p3. Therefore, the desired characteristic polynomial is expressed as:which expands to:Matching the coefficients of Δ(s) and Δd(s) yields:andThus, the additional pole is determined as:and the PI gains are obtained as:andThe resulting controller can then be evaluated through simulation and experimental implementation. This activity allows learners to compare the performance of different controller structures and examine the relationship between design specifications and observed system behavior.

C(s)=Kp+Kis.(7)

Δ(s)=s3+a1s2+(a0+b0Kp)s+b0Ki.

Δd(s)=(s2+2ζωns+ωn2)(s+p3),

Δd(s)=s3+(2ζωn+p3)s2+(ωn2+2ζωnp3)s+ωn2p3.

a1=2ζωn+p3,

a0+b0Kp=ωn2+2ζωnp3,

b0Ki=ωn2p3.

p3=a1−2ζωn,

Kp=ωn2+2ζωnp3−a0b0,

Ki=ωn2p3b0.

Figure 7 presents a simulated response obtained using the PI controller designed from the identified second-order model. This stage allows learners to verify whether the selected controller gains satisfy the desired performance specifications before implementation on the physical system. The simulation also provides an opportunity to compare the behavior of different controller structures and to analyze the influence of controller design choices on the predicted closed-loop response.

After validating the controller in simulation, learners implement the PI controller on the FrED platform and assess its performance using real experimental data. Figure 8 shows an experimental response obtained from the subsystem. Comparing the experimental and simulated responses provides learners with an opportunity to examine the effects of model simplifications, parameter-estimation errors, and practical implementation constraints that influence real-world control-system performance.

3.3.3.2 Controller design using the first-order model

The first-order model provides a simpler representation of the system and allows learners to investigate how model complexity influences controller tuning and closed-loop performance. Using the same performance specifications, the controller gains are obtained by matching the closed-loop characteristic polynomial with the desired second-order characteristic polynomial. The PID and PI gains are computed as follows.

For the first-order model, the PID controller is defined as in Equation 6. Considering the identified first-order model in Equation 1, the closed-loop characteristic polynomial under unity feedback is given by:The controller gains are obtained through characteristic-polynomial matching. The desired second-order characteristic polynomial is defined according to the specified damping ratio ζ and natural frequency ωn as:Matching the coefficients of Δ(s) and Δd(s) yields:andTherefore, the proportional and integral gains are obtained as:andSince the first-order plant combined with a PID controller results in a second-order closed-loop characteristic polynomial, coefficient matching provides two independent equations for the three controller gains. Therefore, Kd constitutes a free design parameter. For the representative case study, Kd=0.03 was selected, and the remaining gains were calculated from the expressions above.

Δ(s)=(τ+kmKd)s2+(1+kmKp)s+kmKi.

Δd(s)=(τ+kmKd)(s2+2ζωns+ωn2).

1+kmKp=2ζωn(τ+kmKd),

kmKi=ωn2(τ+kmKd).

Kp=2ζωn(τ+kmKd)−1km,

Ki=ωn2(τ+kmKd)km.

After tuning the controller, learners can compare simulation and experimental results to evaluate how the simplified model affects controller performance and prediction accuracy.

Figure 9 shows a simulated response obtained with the PID controller designed from the identified first-order model. By analyzing this response, learners can evaluate the expected system performance and compare the predictions of simplified and higher-order models. This activity provides an opportunity to examine the relationship between model complexity, controller tuning, and closed-loop behavior prior to experimental implementation.

Figure 10 shows an experimental response obtained after implementing the PID controller designed from the identified first-order model. The comparison between experimental and simulated responses allows learners to assess the predictive capability of the simplified model and to contrast its performance with that obtained using the second-order representation.

Through this activity, learners can explore the relationship between model complexity, controller tuning, and real-system behavior.

For the first-order model, the PI controller is defined as in Equation 7. The closed-loop characteristic polynomial under unity feedback is:The desired second-order characteristic polynomial is selected according to the specified damping ratio ζ and natural frequency ωn as:Matching the coefficients of Δ(s) and Δd(s) gives:andTherefore, the PI gains are obtained as:andAs with the second-order model, learners can implement the controller on the FrED platform and compare the resulting behavior with simulation predictions. This comparison highlights the trade-off between model simplicity and control performance and provides an opportunity to examine the importance of model selection in engineering practice.

Δ(s)=τs2+(1+kmKp)s+kmKi.

Δd(s)=τ(s2+2ζωns+ωn2).

1+kmKp=2ζωnτ,

kmKi=τωn2.

Kp=2ζωnτ−1km,

Ki=τωn2km.

Figure 11 shows a representative simulated response obtained with the PI controller designed from the identified first-order model. By analyzing this response, learners can assess the expected system performance and compare the predictions obtained from first-order and second-order representations. This activity helps illustrate the relationship between model complexity, controller tuning, and the resulting closed-loop behavior.

Figure 12 shows a representative experimental response obtained after implementing the PI controller designed from the identified first-order model. By comparing the experimental results with both the simulation predictions and the responses obtained using the second-order model, learners can investigate the effects of model complexity on controller performance and examine the trade-offs involved in control-system design.

Tables 2, 3 summarize the controller gains and performance metrics obtained for the PI and PID controllers designed from the identified first-order and second-order models. Rather than focusing exclusively on controller performance, these results serve as a basis for educational analysis and discussion. Learners can compare how model complexity influences controller tuning, evaluate the differences between simulation and experimental results, and investigate the trade-offs between tracking accuracy and control effort. In the presented case study, the PID controllers consistently achieved lower tracking errors than the PI controllers in both simulation and experimental implementation. The accumulated control effort remained of a comparable magnitude for the two controller structures, with small differences depending on the identified model and experimental implementation. From an educational perspective, this comparison provides learners with an opportunity to examine key control-design trade-offs and to interpret how different controller structures influence real-world process behavior. The analysis therefore connects numerical performance indices with their engineering interpretation within the proposed learning activity.

Table 2. Controller gains obtained for the first- and second-order models.

Table 3. Performance comparison of PI and PID controllers in simulation and experimental implementation.

As an optional extension activity, learners may analyze the open-loop and closed-loop pole locations to further explore the relationship between controller tuning and system dynamics. Such analyses can complement the proposed workflow by providing additional insight into stability and transient-response characteristics. However, since the primary objective of this work is to illustrate the integration of system identification, controller design, and experimental validation, detailed pole-location analyses are outside the scope of the present study.

The IAE is defined as:where r(t) and y(t) are expressed in rad/s; therefore, the resulting IAE is expressed in radians.

IAE=∫0T|r(t)−y(t)|dt,

The accumulated control effort is evaluated in terms of the PWM duty cycle as:where T=10 s is the common evaluation interval. Therefore, JPWM is expressed in %⋅s. For the experimental implementation, the recorded PWM duty cycle is used directly. For both simulation and experimental implementation, the control effort is expressed in terms of the equivalent PWM duty cycle to enable a consistent comparison.

JPWM=∫0T|PWM(t)|dt,

Regarding the accumulated control effort, the experimental values were consistently higher than their corresponding simulation values. For the first-order model, the experimental control effort increased from 149.55 to 188.27%⋅s for the PI controller and from 149.62 to 176.47%⋅s for the PID controller, corresponding to increases of approximately 25.9% and 17.9%, respectively. For the second-order model, the corresponding increases were approximately 25.5% for the PI controller and 27.1% for the PID controller. Thus, across the four evaluated controllers, the experimental accumulated control effort was approximately 18%–27% higher than in simulation. These differences are consistent with the additional control action required in the experimental system in the presence of effects not fully represented by the identified models, such as friction, mechanical backlash, measurement noise, and other unmodeled dynamics.

Although characteristic-polynomial matching is used in this work as the controller-design method, the proposed framework is not restricted to a particular tuning approach. Alternative design techniques such as frequency-response methods, optimal control, state-space methods, robust control, or data-driven approaches may be incorporated depending on the learning objectives and course level. Consequently, the framework can support learning activities ranging from introductory PID and PI tuning exercises to more advanced control-design topics. The key contribution of the framework is the integration of system identification, controller design, simulation, experimental implementation, and performance evaluation within a unified learning workflow.

Closed-loop stability can also be incorporated as an additional learning objective within the proposed framework. Beyond evaluating the desired performance specifications, learners can verify that the designed controllers produce stable system behavior before experimental implementation. This activity provides an opportunity to examine the relationship between controller parameters, system dynamics, and closed-loop performance while highlighting the importance of stability analysis in practical engineering applications.

3.3.4 Educational implementation and assessment

The proposed framework is intended for undergraduate and graduate courses in control, automation, mechatronics, and related engineering disciplines, as well as for workforce training involving technicians and engineers in industrial environments. Through the FrED platform, learners engage in activities that mirror professional control-engineering workflows, including system identification, controller design, simulation, experimental implementation, and performance evaluation.

The framework also provides multiple opportunities for educational assessment. Depending on the learning objectives, instructors may employ pre-concept and post-concept tests to evaluate knowledge gains related to system identification, modeling, and controller design. Practical competencies may be assessed through the analysis of performance metrics obtained during simulation and experimental activities, such as tracking error, settling time, or control effort. In addition, rubric-based evaluations can be used to assess higher-level outcomes, including model interpretation, justification of design decisions, and reflection on discrepancies between simulated and experimental responses.

By combining theoretical analysis with experimental validation, the framework provides opportunities to address both conceptual knowledge and practical engineering activities. Although a formal educational evaluation is outside the scope of the present work, the proposed assessment strategy establishes the basis for future studies involving academic and industrial learners. Future work will focus on implementing the framework in classroom and workforce-training environments to quantitatively evaluate learning gains, practical skill development, and learner perceptions using the assessment instruments outlined in this work.

3.3.4.1 Example instructional implementation and assessment

A typical academic implementation can be distributed across four to six laboratory sessions of approximately 2–3 h each. Each session addresses one or more stages of the V-model while using the outputs of the previous session as inputs to the next activity. This modular structure allows instructors to adapt the depth of the activities to the learners’ prior knowledge, available laboratory time, and course objectives. A representative five-session implementation focused specifically on the spooler subsystem case study is described in Table 4 to illustrate one possible way in which the proposed framework can be translated into a concrete instructional sequence. This implementation is provided as an example.

Table 4. Representative instructional implementation of the FrED-based V-model learning framework.

As an example of the level of instructional guidance provided within an individual 2–3 h laboratory period, the system-identification session may be organized as follows. During the first 15–20 min, the instructor reviews the objective of the experiment, the relationship between the spooler input and measured motor speed and the selected sampling interval. Learners then inspect the open-loop experiment to determine the input excitation required to obtain sufficiently informative data.

During the following 30–40 min, teams acquire motor input-output data from the FrED spooler using step, ramp, or combined excitation signals. Learners inspect the raw encoder-derived speed measurements and identify effects such as noise, quantization, or irregular samples. When necessary, they apply basic preprocessing and compare the raw and processed signals before proceeding with parameter estimation.

Approximately 40–50 min are then dedicated to guided model fitting. Learners construct the regression matrices associated with the first-order and second-order ARX structures, estimate the unknown parameters using least squares, and obtain the corresponding continuous-time models using the conversion procedure described in Section 3.3.2. They simulate the identified models using the same input signal employed experimentally and compare the predicted and measured motor responses.

During the next 20–30 min, learners validate the identified models using an independent dataset that was not used during parameter estimation. They compare the first-order and second-order models using response plots and quantitative error measures and discuss whether the increased model complexity produces a meaningful improvement in prediction accuracy.

The final 15–20 min are used for interpretation and reflection. Each team summarizes the selected model, reports the estimated parameters, identifies the main sources of discrepancy between model and experiment, and explains how the quality of the identified model may affect subsequent controller design. The identified model and validation results then become the input to the following controller-design laboratory session.

The data-fitting activity is guided through a sequence of tasks rather than through a fully automated identification routine. Learners are asked to: An example assessment rubric for the representative FrED laboratory implementation is presented in Table 5.

• Inspect the acquired input and motor-speed signals and identify regions containing measurement noise or other undesirable effects.
• Preprocess the data when required while documenting any filtering or data selection applied.
• Construct the regressor matrix and output vector corresponding to the first-order ARX structure and estimate its parameter vector using the least-squares formulation.
• Repeat the estimation procedure using the second-order ARX structure.
• Simulate both identified models using the experimental input and compare their predicted outputs with the measured motor response.
• Validate both models using an independent experimental dataset and compare their prediction errors.
• Justify which model provides the most appropriate balance between model simplicity and prediction accuracy for the subsequent controller-design activity.

Table 5. Example assessment rubric for the representative FrED laboratory implementation.

A short pre-laboratory activity can be used to verify that learners understand the concepts required to perform the experiment. Representative questions include: Representative post-laboratory questions can be used to assess interpretation of the experimental results rather than only completion of the mathematical procedure:

• What are the manipulated input and measured output of the FrED spooler subsystem used in this activity?
• Why is an excitation signal required for system identification, and what characteristics should the input signal have to produce informative experimental data?
• What is the difference between a first-order and a second-order dynamic model, and what additional behavior may be represented by the second-order model?
• How does the sampling interval affect the information contained in the acquired motor-speed signal?
• What effects can encoder quantization and measurement noise have on parameter estimation?
• Why should model validation be performed using data that were not employed to estimate the model parameters?

• Which identified model provided the best agreement with the validation dataset, and what evidence supports this conclusion?
• Did the second-order model provide a meaningful improvement over the first-order model? Explain whether the additional model complexity is justified by the experimental results.
• What differences were observed between the measured and predicted motor responses, and what physical or measurement-related effects could explain these differences?
• How did filtering or preprocessing affect the identified parameters and model validation results?
• How could an inaccurate identified model affect the subsequent PI or PID controller design?
• If the experiment were repeated, what change would you make to the input excitation, sampling procedure, or preprocessing method to improve the identification results?

3.3.5 Open resources and code availability

To facilitate reproducibility and adoption of the proposed framework, the Python and MATLAB resources developed for system identification, controller design, simulation, and experimental implementation are publicly available through a GitHub repository. These materials can be used by instructors, students, technicians, and engineers to replicate or extend the learning activities presented in this work. The repository is continuously updated as new experiments, learning activities, and software improvements are developed. The repository is available at: https://github.com/FreadteamCCM/FREDMexico.

4 Discussion, implications, and limitations

The results presented in this work demonstrate the feasibility of implementing the proposed V-model-based learning framework on the FrED platform through a complete control-engineering workflow. The experimental case study shows that learners can work within the same physical manufacturing-oriented platform to perform process analysis, acquire experimental data, identify and validate mathematical models, design and simulate controllers, implement them on hardware, and evaluate their performance.

From an educational perspective, the proposed framework provides a structure for organizing control-engineering activities around the development cycle of a physical system rather than treating modeling, identification, simulation, controller design, and implementation as isolated exercises. This organization can expose learners to the dependencies among the different stages of control engineering. For example, the identified model is not only an outcome of an identification exercise, but becomes an input to controller design and simulation, while the subsequent physical implementation allows the simulated and experimental responses to be compared. Consequently, discrepancies caused by modeling assumptions, measurement noise, actuator limitations, sampling, and other practical effects can become part of the learning process. Although the present study does not measure learning gains, the framework establishes an experimental environment in which these relationships can be explicitly explored and subsequently evaluated through formal educational studies.

The framework also has implications for industrial workforce training. FrED represents a manufacturing process in which sensing, actuation, modeling, control, and product-related variables interact within the same physical system. Organizing training activities according to the V-model therefore provides a way to connect control-engineering concepts with a structured engineering-development workflow. The same sequence of requirements analysis, modeling, identification, design, implementation, and evaluation can be used to formulate training activities for students, technicians, and engineers at different levels of complexity. This may facilitate the adaptation of the framework to academic laboratories and professional training environments without requiring a different methodological structure for each learner population. However, the effectiveness of this approach for industrial workforce development remains to be evaluated with industrial participants.

An additional implication concerns the extensibility of the proposed framework. Although the spooler DC motor is used solely as the representative case study in this work, the learning framework is defined at the platform level rather than for a specific subsystem. Its structure therefore allows the same V-model-based workflow to be formulated for other FrED subsystems, such as the thermal, feeding, and fiber-diameter dynamics, which present different modeling and control challenges. The general stages of the proposed framework may also support its adaptation to other physical platforms in which learners can perform the control-engineering development cycle. Experimental evaluation of these extensions constitutes an important direction for future work.

The present study nevertheless has limitations. The complete workflow is experimentally demonstrated using the FrED spooler subsystem as a representative case study. Although the framework is designed at the platform level to support its application to other FrED subsystems, these extensions have not yet been experimentally evaluated. In addition, the study focuses on the design and experimental demonstration of the learning framework and does not yet include a formal educational intervention with learners. Consequently, no conclusions can presently be drawn regarding learning gains or the educational effectiveness of the framework across different learner populations, including students, technicians, and engineers. Although the framework is structured around general stages of the control-engineering workflow and may therefore be adaptable to comparable physical platforms, its application beyond FrED remains to be investigated.

These limitations define different directions for future research. Formal educational studies should evaluate the framework with student, technician, and engineering populations using appropriate learning and performance measures. Future implementations should also extend the workflow to additional FrED subsystems and investigate its application to other experimental platforms and manufacturing processes. Such studies would provide evidence regarding the educational effectiveness, transferability, and generalizability of the proposed framework beyond the experimental demonstration presented in this work.

5 Conclusions

This work proposed a V-model-based learning framework implemented on the MIT FrED platform to integrate the complete control-engineering workflow within a single manufacturing-oriented environment for control engineering education and industrial workforce training. The framework addresses the gap identified in the literature between educational approaches that successfully incorporate individual or combined activities in modeling, system identification, controller design, simulation, and hardware experimentation, and the need for a structured learning approach that connects these activities throughout a complete engineering-development workflow.

The proposed framework addresses the central research question by organizing process analysis and requirements specification, mathematical modeling, experimental system identification and model validation, controller design, simulation, physical implementation, and performance evaluation according to the V-model within the same physical platform. The FrED spooler subsystem was used solely as a representative experimental case study to demonstrate the execution of this workflow. Experimental data from the physical system were used for model identification and validation, the identified models supported controller design and simulation, and the resulting controllers were implemented on the physical platform for comparison between simulated and experimental responses. These results demonstrate the feasibility of implementing the proposed framework as an integrated control-engineering workflow rather than as a collection of isolated educational activities.

The main contribution of this work is therefore the structured learning framework itself and its experimental demonstration on a manufacturing-oriented physical platform. By connecting the different stages of control-system development, the framework provides a common structure in which theoretical models, experimental data, simulation, and physical implementation are directly related. Its organization at the platform level also allows learning activities of different complexity to be formulated within the same general workflow, supporting its intended use in both academic control education and industrial workforce training.

The present study also has limitations. The complete workflow has been experimentally demonstrated using the FrED spooler subsystem as a representative case study, while its implementation on other FrED subsystems has not yet been experimentally demonstrated. In addition, the study does not include a formal educational intervention with learners. Consequently, no claims are made regarding learning gains or the educational effectiveness of the framework across students, technicians, or engineers. Furthermore, its application to physical platforms beyond FrED remains to be investigated.

Future work will therefore focus on evaluating the framework through formal educational studies involving different learner populations, extending its implementation to additional FrED subsystems, and investigating its adaptation to other physical platforms and manufacturing processes. These studies will allow the educational effectiveness, applicability, and generalizability of the proposed learning framework to be assessed beyond the experimental demonstration presented in this work.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Hariton Polatoglou, Aristotle University of Thessaloniki, Greece

Reviewed by: Miguel Sellitto, University of the Rio dos Sinos Valley, Brazil

Rakshith Badarinath, Independent Researcher, San Jose, United States