A hierarchical architecture for brownfield digitization: design, analytical insights, and validation of an ultra-low-cost Wi-Fi/MQTT IIoT framework

The digitization of brownfield manufacturing environments, particularly within Small and Medium-sized Enterprises (SMEs), is fundamentally constrained by the lack of robust, scalable, and cost-effective communication infrastructure capable of retrofitting legacy equipment. While Industry 4.0 applications demand reliable, low-latency data acquisition, existing solutions often fail to balance cost-effectiveness with the performance demands of industrial settings. This paper introduces the design, optimization, and empirical evaluation of a highly cost-effective Industrial Internet of Things (IIoT) communication framework tailored for challenging industrial environments. Utilizing a conventional hierarchical edge-to-cloud architecture, this study leverages existing Wi-Fi infrastructure and the MQTT protocol to enable decentralized, parallel data acquisition while optimizing localized sensor density through aggregator nodes. The hardware design utilizes custom ESP32-based nodes, achieving a unit cost of approximately $74.00. Rather than proposing a fundamentally novel topology, the core scientific contribution of this work is the provision of an applied deployment blueprint and the extraction of transferable analytical insights for accessible MQTT/Wi-Fi deployment. Performance was quantified through a facility-level pilot deployment of 19 nodes in an operational industrial bakery characterized by significant electromagnetic interference and physical obstructions. The results demonstrate stable operation, achieving a system uptime of 99.92% over a 30-day period. Network performance metrics confirm the framework’s suitability for high-fidelity monitoring, recording an average end-to-end latency of 185 ms and a low jitter of 86 ms. Furthermore, the scalability of the proposed architecture was explored through Discrete-Event Simulation (DES), suggesting acceptable monitoring performance for approximately 200 concurrent nodes under stated assumptions. Security trade-offs were theoretically analyzed, justifying a defense-in-depth strategy that prioritizes system availability and latency budgets over device-level encryption in resource-constrained environments. This study provides an applied deployment blueprint for establishing the foundational communication layer required for localized Industry 4.0 applications in brownfield settings.

The “brownfield” challenge and the need for data

The realization of Industry 4.0 concepts, particularly the development of Digital Twins (DT), relies heavily on the continuous acquisition of high-fidelity data from the shop floor (Cavalieri and Salafia, 2020; Mendonca et al., 2025a; Kiangala and Wang, 2025). A Digital Twin requires a robust communication channel between the physical asset and its virtual counterpart. However, the manufacturing landscape, especially within Small and Medium-sized Enterprises (SMEs), is dominated by “brownfield” facilities. These facilities utilize legacy equipment that often predates modern connectivity standards, lacking the sensors and communication protocols necessary for integration into digital frameworks (Atzeni et al., 2023; Korodi et al., 2024; Mendonca et al., 2025b).

This “brownfield dilemma” presents a significant barrier to digitalization. While the replacement of functional legacy machinery is economically unfeasible, traditional retrofitting approaches often involve expensive Programmable Logic Controller (PLC) upgrades, proprietary IIoT gateways, or extensive rewiring, which are disruptive and capital-intensive (Vuković et al., 2022; Kaiser et al., 2023; Ulewicz et al., 2019). Consequently, many SMEs operate with critical data blind spots, unable to optimize energy consumption, monitor asset health in real-time, or implement predictive maintenance (PdM) strategies.

Research gap in IIoT retrofitting

Despite significant advancements in Industrial Internet of Things (IIoT) technologies, the digitization of brownfield manufacturing environments remains challenging, primarily due to economic constraints (Vuković et al., 2022; Kaiser et al., 2023; Ulewicz et al., 2019). The literature consistently highlights that the high cost of implementation and the significant infrastructure investments required for Industry 4.0 solutions render them inaccessible, particularly for Small and Medium-sized Enterprises (SMEs). While commercial solutions and established industrial standards (e.g., utilizing OPC UA or specialized gateways like the Siemens Simatic IoT 2040) offer robust performance, they are often financially prohibitive (Korodi et al., 2024; Vuković et al., 2022; Rusu et al., 2025). The high cost associated with integrating or replacing functional legacy equipment necessitates the development of economically effective sensing systems that reduce expenditure without compromising efficiency. Furthermore, the financial viability of retrofitting is often difficult to ascertain, as studies frequently lack access to the necessary data to calculate a clear Return on Investment (ROI).

Beyond cost, significant technological and scalability challenges persist. The IoT and edge computing domains suffer from a lack of standardization in both infrastructure and software, resulting in heterogeneous environments. This heterogeneity creates substantial difficulties in integrating diverse communication protocols and managing the complexity of the Operational Technology (OT) layer (Franke et al., 2024; Szűcs, 2023). Many existing frameworks fail to simultaneously provide technology-independent functionality while effectively abstracting this underlying OT complexity. Architectural choices also present limitations; for instance, Low Power Wide Area Networks (LPWANs) such as LoRaWAN and Sigfox, while suitable for low-energy applications, are constrained by low data transmission rates, limiting their use in data-intensive monitoring (Nurelmadina et al., 2021; Sámano-Robles et al., 2021). Conversely, legacy industrial protocols like Modbus/TCP carry inherent vulnerabilities, lacking robust authentication mechanisms and raising concerns about the resilience of OT systems when adapted to modern TCP/IP stacks (Federici et al., 2023; Lupascu et al., 2020; Tidrea et al., 2019; Tarjan et al., 2020; Kotsiopoulos et al., 2025). Consequently, scalability remains a persistent obstacle for many proposed integration solutions.

Finally, there is a critical gap concerning the validation and generalization of proposed IIoT frameworks in real-world industrial settings. A substantial portion of the current research relies on data generated through simulations, which may not accurately reflect the complexities and variabilities of actual manufacturing operations (Kapoor et al., 2025; Heusel et al., 2024; Callegati et al., 2025). Many studies present prototypes or proofs-of-concept with limited scope, focusing on specific components (e.g., bearings) or single machine modules, rather than addressing facility-wide integration. Furthermore, validation efforts are often constrained by short data collection periods, which may be insufficient to capture the full operational spectrum or potential equipment failures. This reliance on simulated data and limited case studies restricts the generalizability of these solutions, highlighting a need for robustly validated frameworks deployed at scale in authentic brownfield environments.

Our contribution

This study addresses these gaps by proposing, evaluating, and extracting analytical insights from a highly cost-effective IIoT deployment blueprint utilizing a standard hierarchical Wi-Fi/MQTT architecture designed for brownfield environments. The main contributions of this work are:

An Applied Blueprint for Brownfield IIoT: We formalize a flexible, hierarchical deployment strategy that balances the scalability of direct Wi-Fi/MQTT connections with the localized efficiency of dedicated aggregator nodes. This standard architecture is optimized specifically to eliminate the complexities of traditional fieldbuses while adhering to strict SME cost boundaries.

Cost-Effective Node Design: We demonstrate a hardware design based on the ESP32 microcontroller and COTS components, achieving a total cost of approximately $74.00 per node, significantly lowering the barrier to entry for digitization.

Real-World Facility-Level Pilot and Transferable Insights: We evaluate the framework through a pilot deployment of 19 nodes in an operational industrial bakery. Crucially, we abstract this engineering case study into generalizable analytical guidelines regarding spectrum selection, buffer management, and proxy metrics for network health, empowering future researchers and engineers to optimize deployments in high-EMI environments.

System overview

The proposed IIoT framework is designed based on a modular, four-layer architecture that facilitates scalable data acquisition and processing in brownfield environments (Figure 1). This architecture delineates the flow of information from the physical assets on the shop floor to the visualization interfaces, ensuring separation of concerns between data generation, transmission, storage, and application: Perception Layer: This foundational layer comprises the physical assets and the integrated sensing hardware responsible for raw data generation. It includes industrial machinery instrumented with certified energy meters (e.g., Sayman, Orman) and environmental sensors. This layer forms the interface between the physical domain and the cyber domain.

Network Layer: This layer is responsible for data acquisition, protocol translation, and reliable data transport. Distributed ESP32 microcontrollers act as edge acquisition nodes, interfacing with the sensors in the Perception Layer via localized industrial protocols (e.g., Modbus RTU over RS-485). These nodes preprocess the data and transmit it over the existing Wi-Fi infrastructure (IEEE 802.11) using the lightweight MQTT protocol. An on-premise gateway (Raspberry Pi 5) hosts the MQTT Broker (Eclipse Mosquitto), managing the message routing and ensuring Quality of Service (QoS).

Data Layer: This layer handles the storage and management of the high-velocity data streams. The MQTT Broker forwards validated payloads to InfluxDB, a time-series database optimized for the high-write throughput and efficient querying of time-stamped IoT data.

Application Layer: The final layer provides data visualization, analysis, and user interaction. It utilizes Grafana for real-time operational dashboards and alerting, enabling monitoring and supervisory functions.

Distributed IoT node design

The hardware architecture of the individual IIoT node prioritizes cost-effectiveness, reliability in industrial conditions, and ease of replication by utilizing Commercial Off-The-Shelf (COTS) components. The design facilitates non-invasive retrofitting, allowing the node to be integrated into existing electrical cabinets. The complete assembly is housed within an IP-rated enclosure (IP51 or higher, depending on the deployment zone) to protect against contaminants such as dust and humidity prevalent in manufacturing environments (Figure 2). The node comprises the following key subsystems: Processing Unit (ESP32-WROOM-32): The core of the node is the ESP32 System-on-Chip (SoC). Selected for its dual-core architecture, integrated Wi-Fi (IEEE 802.11 b/g/n) connectivity, and low cost, the ESP32 manages all node operations, including sensor interfacing (via UART), local data buffering, protocol translation (Modbus RTU to MQTT), and communication with the central broker. To mitigate the risk of RAM overflow and maintain data integrity during network saturation phases (e.g., tail latencies > 700 ms), the ESP32 firmware implements a local circular buffer (ring buffer) strategy. In scenarios where MQTT QoS 1 ACK packets are severely delayed, the buffer temporarily stores outgoing payloads. If the memory limit is approached, the oldest state data is systematically overwritten by the freshest readings, prioritizing real-time operational context over historical completeness.

Metrology Subsystem (Energy Meter): Accurate measurement of electrical parameters (voltage, current, active/reactive power, power factor, and phase imbalance) is performed by standardized, certified three-phase or single-phase energy meters (e.g., Sayman/Orman). These devices provide the necessary accuracy class (e.g., Class 1.0) required for industrial monitoring.

Communication Interface (MAX485): To ensure robust communication between the Processing Unit and the Metrology Subsystem in an electrically noisy environment, a MAX485 transceiver module is employed. This module performs signal conditioning, converting the ESP32’s 3.3 V TTL UART signals to the differential signals required by the RS-485 standard. The Modbus RTU protocol is utilized over this physical layer. Critically, the RS-485 bus is strictly localized within the enclosure and is not used for primary network communication.

Power Management Unit (HLK-PM01): The node is powered directly from the monitored electrical line using a compact, isolated AC-DC converter (HLK-PM01). This module efficiently steps down the 220 V AC input to a stable 5 V DC rail required by the digital electronics, eliminating the need for external power adapters and simplifying deployment.

Network and server architecture

The Network and Application layers utilize robust, open-source technologies deployed on-premise to ensure data sovereignty and minimize recurring costs.

Network Layer: The framework leverages the existing Wi-Fi infrastructure (IEEE 802.11n/g). MQTT (Message Queuing Telemetry Transport) is the chosen communication protocol due to its lightweight nature, publish/subscribe model, and ability to handle intermittent connectivity, which is crucial in potentially congested industrial Wi-Fi environments (Kiangala and Wang, 2025). We utilize MQTT QoS 1 (at least once delivery) to ensure data reliability.

Server Architecture: The centralized server is hosted on a Raspberry Pi 5, providing a cost-effective yet powerful computing platform suitable for the scale of this deployment. The software stack includes: MQTT Broker (Eclipse Mosquitto): Handles the routing of messages between the nodes and the database.

Time-Series Database (InfluxDB): Optimized for storing and querying the high volume of time-stamped IoT data generated by the nodes.

Visualization and Alerting (Grafana): Provides real-time dashboards, operational analysis, and an alerting engine.

System Configuration and Replicability Parameters To ensure the replicability of this deployment, the core configuration parameters, message schemas, and broker settings utilized in the host facility are explicitly detailed below. MQTT Broker Settings (Eclipse Mosquitto) The central broker was configured to prioritize stability and connection retention in a high-EMI environment. Key mosquitto.conf parameters included:

allow_anonymous false (Enforcing basic credential-based access)

max_connections −1 (Unlimited connections to prevent artificial bottlenecks)

max_queued_messages 1,000 (Buffer for handling traffic bursts)

keepalive_interval 60 (Seconds)

Node Communication Parameters (ESP32)

Protocol: MQTT v3.1.1 over TCP/IP (IPv4)

Quality of Service (QoS): QoS 1 (At least once delivery) for all telemetry data to prevent loss during transient Wi-Fi disconnects.

Publish Interval: 10,000 ms (10 s) per metrology payload.

Wi-Fi Standard: IEEE 802.11n (2.4 GHz only, leveraging Channels 6 and 11).

Standardized Message Schema (JSON Payload) To standardize data ingestion into InfluxDB, a lightweight JSON schema was utilized for all ESP32 published messages. A representative payload for an energy meter node is structured as follows:

Inline FX Screenshot of a JSON data structure displaying device information, including device ID ESP32_Dala_Meter_002006, timestamp 1700215200, voltage and current readings for three phases, total active power 8540.5, frequency 49.98, signal strength RSSI negative 68, and sequence number 14502.

Cost-benefit analysis

A critical objective of this framework is to minimize the financial barrier to digitization (Vuković et al., 2022; Kaiser et al., 2023). Table 1 details the Bill of Materials (BoM) for a single monitoring node.

The total cost of $74.00 per node is significantly lower than commercially available IIoT solutions or industrial gateways, which often range from $500 upwards. This ultra-low-cost approach enables SMEs to deploy monitoring systems at scale. For the 19-node deployment presented in this study, the total hardware investment (including the Raspberry Pi 5 server) was under $1500 USD.

Performance validation methodology

To systematically characterize the communication reliability and network performance of the proposed Wi-Fi/MQTT framework within the industrial environment, a systematic methodology was employed for quantifying key performance indicators (KPIs): End-to-End Latency, Network Jitter, and Packet Loss Rate (PLR). The validation spanned 30 continuous days of operation.

Time synchronization protocol

Accurate measurement of one-way latency requires precise clock synchronization between the distributed ESP32 nodes and the central on-premise server. The Network Time Protocol (NTP) was implemented across all devices, synchronizing with a localized NTP stratum server to minimize the impact of external network variability. The synchronization accuracy was empirically verified and maintained with a clock offset within ±10 ms, providing the necessary temporal precision for latency assessment.

Latency and jitter measurement

End-to-end latency (L e 2 e) is defined as the total time elapsed from the acquisition of a sensor reading at the node to the successful processing and insertion of the corresponding data packet into the time-series database. To measure this, the firmware on each ESP32 node was instrumented to append a high-resolution timestamp (T n o d e) to the MQTT payload immediately following the sensor acquisition (e.g., Modbus RTU transaction).

Upon reception at the server, a dedicated subscriber script recorded the timestamp of the database write operation (T d b). The latency for each individual packet was calculated using Equation 1:

Network Jitter, defined as the Packet Delay Variation (PDV), was subsequently calculated as the standard deviation of the measured L e 2 e values collected over the 30-day period, providing a quantitative measure of the communication channel’s temporal consistency.

Packet loss rate (PLR) measurement

To accurately determine the Packet Loss Rate (PLR) at the application layer, independent of the transport layer mechanisms, a sequence number mechanism was implemented. Each MQTT message published by a node included a monotonically increasing integer identifier embedded within the payload.

On the server side, the subscriber script analyzed the sequence of incoming messages for each specific node. Gaps detected in the sequence numbers indicated lost packets. While MQTT QoS 1 (at least once delivery) mitigates transient network errors, the measured PLR reflects sustained connectivity issues or node failures. The PLR was calculated based on the total number of messages expected (N s e n t, inferred from the sequence numbers) and the total number of messages successfully received (N r e c e i v e d), as shown in Equation 2:

Scalability simulation methodology

While the empirical validation was conducted on a physical testbed of 19 nodes, industrial deployments often require scaling to hundreds of sensor points. To explore potential architectural performance under high-density conditions without incurring the prohibitive cost of physical deployment, an exploratory Discrete-Event Simulation (DES) approach was employed.

The simulation environment was constructed using the Python SimPy library. While sophisticated network simulators such as ns-3 offer high-fidelity physical (PHY) layer and intricate electromagnetic channel modeling, our analytical focus was strategically directed toward application-layer behavior—specifically, MQTT broker queuing dynamics (modeled as an M/M/1 queue), CSMA/CA backoff penalties, and payload delivery times under extreme congestion. Therefore, a stochastic DES model using SimPy was deemed optimally suited to isolate and evaluate these application-layer saturation thresholds without the computational overhead of deep radio propagation simulation.

To improve methodological transparency and replicability, the core simulation parameters are summarized in Table 2.

Host facility description and deployment

The framework was implemented in a medium-sized industrial bakery in Almaty, Kazakhstan. This facility represents a typical brownfield environment, characterized by a mix of modern and legacy equipment, high electrical loads, fluctuating temperatures, and airborne particulates (flour dust).

The critical equipment includes large industrial ovens (e.g., Wachtel Columbus, Wachtel Compact, Figures 3a,b), various dough mixers, and supporting machinery such as ice machines (Figure 3c).

Prior to implementation, the facility had no granular energy monitoring. A total of 19 distributed IoT nodes were deployed across these critical assets, utilizing the existing Wi-Fi network. The installation was minimally invasive. Figure 4 illustrates the facility layout (simulation model) and the placement of the monitoring nodes across the production flow.

The deployment utilized the facility’s existing Wi-Fi infrastructure, operating exclusively in the 2.4 GHz ISM band (IEEE 802.11n), primarily utilizing Channels 6 and 11. The deliberate architectural choice to utilize legacy Wi-Fi 4 at 2.4 GHz, rather than adopting modern 5 GHz standards (Wi-Fi 6/7), was driven by two fundamental constraints: physical penetration and hardware cost. First, the 2.4 GHz wavelength provides vastly superior signal penetration through the dense physical obstructions heavily prevalent in brownfield environments, such as structural steel and thick metal industrial ovens, where 5 GHz suffers rapid attenuation. Second, standard Wi-Fi 4 hardware perfectly aligns with the extreme cost constraints of SME retrofitting, while its available bandwidth effortlessly accommodates the ultra-low data requirements of MQTT telemetry (256-byte payloads).

The Radio Frequency (RF) environment presented severe physical challenges, characterized by high levels of electromagnetic interference (EMI) from industrial machinery and Variable Frequency Drives (VFDs). Furthermore, numerous massive metal obstructions (e.g., multi-deck ovens, structural steel, racking) caused significant signal attenuation and multipath propagation effects. Despite this highly challenging physical environment, an RF site survey conducted during peak operation quantified a remarkably resilient link quality across the deployment zones. Through strategic node placement and the inherent propagation characteristics of the 2.4 GHz band, the average Received Signal Strength Indicator (RSSI) across all 19 nodes was maintained at −68 dBm (ranging from −55 dBm to −82 dBm). Overcoming the environmental obstacles, the deployment successfully yielded a Signal-to-Noise Ratio (SNR) that was consistently maintained above 20 dB. An SNR of 20 dB is indicative of an excellent, high-quality channel, guaranteeing highly reliable MQTT communication despite the harsh surroundings.

In dense 2.4 GHz industrial environments, co-channel interference and packet collisions are dominant impairments. Ideally, these are quantified using the Signal-to-Interference-Plus-Noise Ratio (SINR). However, a significant constraint of deploying ultra-low-cost hardware is that the standard API of the ESP32 (Xtensa LX6) network stack natively exposes RSSI, but lacks the hardware-level hooks required to extract raw SINR data. To overcome this limitation and accurately assess true channel health, our methodology relies on application-layer proxies, specifically Packet Loss Rate (PLR) and Network Jitter, which effectively capture the systemic impact of collisions that RSSI alone cannot.

Real-time monitoring results

The system successfully established real-time data acquisition from all 19 nodes. Data was transmitted via MQTT to InfluxDB and visualized using Grafana.

Figure 5 demonstrates the real-time acquisition of electrical parameters for a primary production machine (Asset ID: ESP32_Dala_Meter_002006). The visualization presents synchronized time-series data illustrating the voltage profiles across the three phases (top panel), the corresponding current draw which clearly delineates the machine’s pulsed operational cycles (middle panel), and the total active power consumption (bottom panel). This granular data stream validates the framework’s capability to capture high-fidelity metrics, revealing insights such as distinct phase asymmetry in the voltage profiles and peak power consumption during active cycles.

In addition to energy monitoring, the framework’s flexibility was demonstrated by deploying nodes equipped with environmental sensors to monitor temperature and humidity in critical production areas (Figure 6), utilizing the same Wi-Fi/MQTT architecture.

Operational state monitoring and event alerting

Beyond raw data collection, the framework provides the foundation for deriving actionable insights, a key requirement for Digital Twin implementation. By analyzing energy consumption profiles (as seen in Figure 5, middle panel), the operational state of the machinery (e.g., ON, OFF, Idle, Active) can be accurately determined (Atzeni et al., 2023).

The system also demonstrated capabilities in threshold-based alerting and event detection. Figure 7 illustrates an illustrative instance where the visualization engine successfully captured and automatically annotated a critical Grid Frequency Instability event. The detection of such deviations from the nominal frequency (e.g., 50 Hz) is crucial for protecting sensitive legacy equipment, particularly asynchronous motors and VFDs, from operational irregularities and potential damage caused by grid stress, thereby ensuring rapid operational response.

Framework performance validation

A central premise of this research is the viability of using a hierarchical topology over existing Wi-Fi for localized IIoT deployment. To assess the reliability and performance in this industrial setting, we analyzed network performance metrics over a continuous 30-day monitoring period. During this period, with all 19 nodes configured to a publish interval of 10 s, the central MQTT broker processed approximately 4.92 million telemetry messages.

System Uptime was calculated as the ratio of successfully received heartbeat messages (transmitted every 60 s independent of sensor payloads) to the theoretically expected number of heartbeats over the 720-h period. The empirical results, including a comprehensive breakdown of the latency distribution, are summarized in Table 3.

The results demonstrate excellent performance despite relying on the existing Wi-Fi infrastructure in a challenging RF environment. The median and average latencies (165 ms and 185 ms, respectively) are well within the requirements for near-real-time monitoring. The low packet loss rate (0.8%) and high system uptime (99.92%) support the reliability of the architecture within this single-site empirical context.

Latency distribution analysis

To move beyond aggregate average metrics and descriptively evaluate the reliability of the communication framework, the empirical latency distribution of the network was analyzed. Figure 8 presents the Cumulative Distribution Function (CDF) of the end-to-end latency. The CDF reveals the “tail latency” characteristics, a critical metric for industrial control systems where worst-case performance dictates reliability. The analysis confirms that 99% of all data packets arrived within 453 ms (P (X ≤ 453 ms) = 0.99). This descriptive bound indicates that the system aligns with typical soft real-time monitoring tolerances, ensuring that outliers caused by temporary Wi-Fi retransmissions do not impact the overall observability of the Digital Twin.

Furthermore, the latency distribution was visualized using the histogram shown in Figure 9. The observed latency data can be approximated by a Gaussian profile, with a fitted standard deviation of σ = 86 ms. This narrow distribution quantitatively supports the previously observed low jitter metric, illustrating relatively consistent performance despite the presence of significant Electromagnetic Interference (EMI) in the brownfield environment.

Scalability simulation results

The theoretical scalability of the architecture was evaluated by sweeping the network size from N = 19 to N = 300 nodes within the discrete-event simulation model. The results, illustrated in Figure 10, quantify the simulated relationship between node density and network latency.

The blue solid line represents the average latency, while the red dashed line depicts the 99th percentile (P99) worst-case latency. The simulation reveals two distinct operational phases. In the linear phase (N ≤ 200), the average latency increases marginally, remaining well below the critical monitoring threshold of 500 ms. Specifically, at N = 200, the average latency is recorded at 492 ms. However, beyond N = 250, the system enters an exponential degradation phase. The P99 latency rises sharply, exceeding 700 ms at N = 300. This inflection point is attributed to channel saturation and the exponential increase in packet collisions and CSMA/CA backoff times.

Consequently, rather than serving as a definitive proof of scalability, this exploratory simulation suggests that the proposed hierarchical architecture has the theoretical capacity to maintain acceptable monitoring performance for medium-scale deployments of up to approximately 200 nodes, under the stated modeling assumptions. Beyond this point, network segmentation (e.g., additional access points or frequency division) becomes a requisite.

It is critically important to note a quantitative discrepancy at the baseline (N = 19), where the empirical latency is 185 ms, while the simulated baseline approaches 220 ms. This variance is an intentional consequence of our modeling parameters. The SimPy broker processing parameters and CSMA/CA contention windows were deliberately configured with highly conservative, worst-case overhead values. Therefore, rather than claiming this model absolutely ‘validates’ the architecture for 200 nodes, we reframe the simulation as establishing a conservative theoretical upper bound for network latency degradation. By deliberately employing this pessimistic calibration, the model bounds the risk for future large-scale deployments, suggesting that real-world performance will likely exceed these simulated thresholds until true physical saturation is reached.

Interpretation of results

The successful implementation of the 19-node system demonstrates the practical viability of the proposed hierarchical topology framework. We have observed that reliable and granular data acquisition is feasible in a brownfield environment without expensive proprietary solutions or infrastructure upgrades.

The framework effectively “switched on the lights” in a previously “blind” facility. The ultra-low-cost nature ($74.00/node) allowed for comprehensive coverage of assets, which would have been economically unfeasible using traditional methods. The performance metrics (Table 3) empirically challenge the assumption that industrial environments require specialized wired or high-cost wireless solutions for reliable monitoring. The combination of existing Wi-Fi infrastructure, the lightweight MQTT protocol, and robust low-cost hardware proved highly effective.

Comparison with alternative solutions

The proposed hierarchical framework offers a distinct balance of cost, performance, and scalability compared to alternative digitization solutions identified in the literature (Table 4).

When compared to Low Power Wide Area Network (LPWAN) technologies such as LoRaWAN, Zigbee, and NB-IoT, our framework demonstrates superior performance for near-real-time applications. LPWAN solutions are optimized for long-range communication (up to 10–20 km) and exceptional energy efficiency, making them ideal for scenarios requiring infrequent transmission of small data payloads. However, they are inherently constrained by low data rates and high latency (Nurelmadina et al., 2021; Sámano-Robles et al., 2021). In contrast, our Wi-Fi/MQTT-based framework achieved an average end-to-end latency of 185 ms, enabling high-frequency data acquisition and rapid operational feedback essential for effective process monitoring and Digital Twin synchronization.

Compared to architectures centered on industrial interoperability standards such as OPC UA, DDS (Data Distribution Service) (Ho et al., 2022), and the Asset Administration Shell (AAS) (Cavalieri and Salafia, 2020), our approach significantly reduces complexity and cost. These standards are designed to achieve high levels of semantic harmonization and robust security in complex, multi-vendor environments, facilitating deep vertical and horizontal integration. However, implementing these standards in brownfield settings often necessitates expensive industrial gateways (e.g., Siemens Simatic IoT 2050), complex wrappers (e.g., converting OPC DA to OPC UA), and specialized expertise (Korodi et al., 2024; Vuković et al., 2022; Rusu et al., 2025). The reliance on such hardware increases the per-node cost substantially. Our framework, utilizing a lightweight MQTT protocol and COTS hardware, achieves a per-node cost of approximately $74, offering a pragmatic and economically feasible alternative for SMEs prioritizing rapid data acquisition over complex semantic modeling.

Our solution aligns with the trend of utilizing open-source, low-cost hardware stacks, such as those based on Raspberry Pi, Arduino, or other ESP-based modules (e.g., M2Cloud), often integrated via Node-RED or similar platforms (Kiangala and Wang, 2025; Alexopoulos et al., 2022). However, while many studies in this category remain at the proof-of-concept stage (Tarjan et al., 2020; Callegati et al., 2025), our work provides empirical validation of the robustness (99.9% uptime) and scalability of a facility-wide deployment. When contrasted with high-security and Big Data architectures—such as those employing Distributed Ledger Technology (DLT)/Blockchain (Lupascu et al., 2020) or Zero Trust Architectures (ZTA) (Federici et al., 2023)—our framework is significantly less complex and avoids reliance on costly, centralized commercial cloud platforms (e.g., Microsoft Azure, AWS). While DLT and ZTA address critical needs for data immutability and advanced security, they demand substantial investment and technical overhead, contrasting sharply with the accessibility and low-investment profile of our proposed solution.

Security constraints and trade-offs in low-cost architectures

In resource-constrained brownfield retrofitting, prioritizing Availability (Real-Time Data Freshness) over device-level Confidentiality is often a necessary architectural compromise. While the absence of TLS/SSL encryption at the application layer is a recognized limitation of this framework, a theoretical overhead analysis validates this decision for the ESP32 (Xtensa LX6) hardware under strict latency budgets. Based on hardware specifications and existing literature—rather than direct empirical measurement in this study—a standard TLS 1.2 handshake involving an RSA-2048 exchange typically requires approximately 1.5–2.0 s of processing time on this specific SoC. In an industrial environment characterized by high EMI and mobile assets, occasional Wi-Fi reconnections are inevitable. If TLS were enabled, every reconnection event would trigger a re-handshake penalty, persistently violating the system’s < 500 ms latency budget and creating operational data blind spots.

To mitigate the lack of transport-layer encryption, the real-world deployment in the host facility strictly implemented a “Defense-in-Depth” strategy centered on Network Isolation. Specifically, the IIoT nodes were deployed on a dedicated Virtual Local Area Network (VLAN), logically air-gapped from the corporate IT infrastructure and the public internet. Furthermore, access to the central MQTT broker was strictly controlled via MAC address filtering and firewall whitelisting at the on-premise gateway level. This network-level isolation ensures that the basic integrity and availability of the data streams are maintained against remote external threats without compromising the high-frequency data acquisition required for Digital Twin synchronization.

However, a rigorous security analysis requires acknowledging the residual risks inherent in this low-cost architecture. Because the MQTT payloads are transmitted in plaintext within the VLAN, the system lacks cryptographic integrity verification and robust replay protection. Consequently, an insider threat with access to the local network could theoretically intercept, spoof, or replay sensor readings. Furthermore, while VLAN isolation mitigates remote network-level intrusion, the system remains highly vulnerable to physical device compromise. In its current ultra-low-cost iteration, the framework does not implement ESP32 hardware-level Flash Encryption. If an edge node is physically stolen or tampered with on the shop floor, the plaintext Wi-Fi and MQTT credentials could be extracted via physical interface connections. Evaluating the latency overhead of enabling Flash Encryption, alongside lightweight transport security alternatives (e.g., DTLS, or symmetric payload encryption such as AES-128-GCM), remains a critical direction for future iterations of this architecture.

Limitations

While the proposed framework demonstrates significant advantages, several limitations must be acknowledged: Dependence on Wi-Fi Infrastructure: The performance is contingent upon the stability and coverage of the existing Wi-Fi network. Facilities with extremely poor Wi-Fi infrastructure or severe electromagnetic interference may require network upgrades (e.g., additional access points).

Security: This implementation utilized basic MQTT security (authentication). For deployment in highly sensitive environments, further security measures, such as TLS/SSL encryption, network segmentation (VLANs), or the implementation of a Zero-Trust Architecture (Federici et al., 2023; Tidrea et al., 2019), are necessary.

Data Analysis Complexity: The framework successfully acquires foundational data. However, translating this data into advanced insights, such as predictive maintenance models, requires the application of Machine Learning (ML) algorithms and specialized data science expertise (Kapoor et al., 2025; Afolaranmi et al., 2023; Anwar et al., 2025).

Latency for Control: The observed latency makes this framework suitable for monitoring and supervisory control, but not for real-time closed-loop control applications requiring deterministic behavior.

Transferable insights for industrial Wi‐Fi/MQTT deployments

While this paper presents an engineering case study of a specific facility, the empirical data yields several generalizable analytical insights that can serve as a blueprint for researchers and engineers implementing low-cost IIoT networks in brownfield settings: Spectrum Selection Prioritizes Penetration over Bandwidth: The default industry trend is to upgrade to Wi-Fi 6/7 (5 GHz) for modernization. However, this study demonstrates that for decentralized telemetry frameworks utilizing lightweight protocols (like MQTT), physical signal penetration is exponentially more valuable than peak channel bandwidth. Engineers retrofitting legacy facilities laden with heavy metals should deliberately target the 2.4 GHz band, which demonstrates vastly superior propagation through dense industrial obstructions compared to 5 GHz.

Proxying Channel Health in Resource-Constrained Hardware: When deploying ultra-low-cost System-on-Chips (such as the ESP32-WROOM-32), the standard APIs often lack support for reporting raw, hardware-level Signal-to-Interference-plus-Noise Ratio (SINR). To accurately evaluate co-channel interference and collision probabilities without escalating hardware costs, engineers must rely on application-layer proxies. Analyzing standard RSSI in strict conjunction with application-layer Packet Loss Rate (PLR) and timestamp-derived Network Jitter provides a highly reliable, empirical assessment of true channel health.

Localized Buffer Management for Tail Latency Mitigation: Industrial machinery generates severe, transient EMI spikes (e.g., upon asynchronous motor startup), which routinely cause brief Wi-Fi disconnections or extreme tail latencies (our probabilistic analysis recorded P99 > 450 ms). To prevent permanent data loss during these events, edge nodes must be designed with dynamic circular buffering (ring buffers) in RAM. During network saturation, firmware must intelligently overwrite historical data with fresh state data, ensuring that the central Digital Twin prioritizes real-time operational context over delayed, complete historical logs.

QoS Parameter Tuning in High-EMI: Operating in an environment with VFDs induces frequent localized packet collisions. While MQTT Quality of Service 1 (at least once delivery) is mandatory to prevent data loss, engineers must tune the MQTT client’s acknowledgment (ACK) timeout parameters to align with the worst-case tail latency of the facility. Setting the ACK timeout too aggressively (e.g., <200 ms) will trigger recursive publication loops, flooding the 2.4 GHz spectrum and exacerbating network congestion.

This paper presented a highly cost-effective IIoT deployment blueprint based on a standard hierarchical architecture, designed for the digitization of brownfield manufacturing environments. By leveraging existing Wi-Fi infrastructure, the MQTT protocol, and ESP32 microcontrollers, we achieved a unit cost of $74.00 per node.

The framework was evaluated through a real-world pilot deployment of 19 nodes in an industrial bakery, demonstrating stable operation (99.92% uptime) and sufficient performance for monitoring applications (185 ms average latency). The results indicate that this hierarchical topology, utilizing parallel MQTT connections over standard Wi-Fi augmented by localized aggregator nodes, is a practical solution for localized data acquisition. Exploratory simulation results suggest that the current architecture can support acceptable monitoring performance for up to 200 nodes under the stated assumptions before channel saturation significantly impacts QoS. This approach provides an economically feasible pathway for SMEs to overcome the “brownfield dilemma” and establish the foundational data layer required for implementing advanced Industry 4.0 applications.

Future work will focus on: Predictive Maintenance (PdM): Applying Machine Learning algorithms to the collected dataset to develop models for predicting equipment failures based on energy consumption patterns and power quality data.

Enhanced Security: Implementing robust security measures, including TLS encryption and certificate-based authentication, and evaluating their impact on the resource-constrained ESP32 hardware.

Digital Twin Development: Utilizing the collected real-time data to develop a comprehensive, dynamic Digital Twin of the bakery’s production processes for simulation and optimization.