1. Introduction
Industry 4.0 is a transformative force in industrial manufacturing, revolutionizing operations to be more efficient and productive. This revolution is powered by the seamless integration of emerging technologies such as the Internet of Things (IoT), Artificial Intelligence (AI), and cloud/edge computing [1]. These technologies indicate a new era in which the manufacturing process is continuously monitored, controlled, and optimized for unprecedented efficiency [2]. A key pillar of this revolution is predictive maintenance, a proactive approach to the maintenance of manufacturing equipment using data analytics and AI. This involves continuous monitoring of equipment conditions, AI-driven analysis, and the prediction of equipment failures for timely preventive maintenance. This significantly reduces unplanned downtime, reduces costs, and prolongs the lifetime of machines such as ventilation systems, thus increasing overall process efficiency [3].
To enable predictive maintenance, wireless industrial Internet of Things (IIoT) sensors are integrated directly into the equipment to monitor and gather the parameters that affect the condition of the equipment. Although simple, it comes with significant challenges, such as monitoring multiple parameters and providing a longer battery life for the sensor platform. A multiple-sensor approach seems reasonable and improves the quality of AI models to predict different features from different sensor data [4]. Recent trends in the industrial Internet of Things (IIoT) have led to the development of multi-sensor platforms that integrate various wireless communication technologies. The combination of LoRa (long-range) and 5G connectivity represents a promising approach, leveraging the strengths of both technologies. LoRa offers low-power, long-range capabilities suitable for sensor networks, while 5G provides the high-speed, low-latency communication essential for real-time data analytics. However, the integration of these technologies also presents challenges, including increased power consumption and complexity in network management.
This paper presents an innovative solution: a hybrid LoRa-5G multi-sensor platform explicitly designed for predictive maintenance of industrial equipment, particularly ventilation systems. Our research contributes to the field by integrating low-power LoRa communication with high-speed 5G capabilities, enabling real-time monitoring and efficient data transmission. This combination addresses the challenges of energy efficiency and data reliability while offering a versatile solution tailored to industrial applications. Although LoRa excels in low-power scenarios, it may not provide the bandwidth necessary for high-frequency data transmission. In contrast, 5G offers improved bandwidth and reliability, but at the cost of higher energy consumption [5].
Understanding these trade-offs is essential for optimizing predictive maintenance strategies. The organization of this paper is as follows: Section 2 details the methodology employed in our study, including the environment under consideration and the specific problems addressed. Section 3 discusses the architecture of the proposed platform, while Section 4 presents the validation methodology and results. Finally, we discuss the implications of our findings and conclude with future directions for research. In summary, this hybrid approach decisively balances energy efficiency, communication reliability, and performance, establishing itself as a formidable and effective solution for predictive maintenance in industrial environments.
2. Materials and Methods
This section outlines the methodology employed in the development of the hybrid LoRa-5G multi-sensor platform for predictive maintenance of ventilation systems. It details the hardware and software components, describes the environment under consideration, and articulates the specific problems addressed by this study.
The proposed system is designed for industrial environments where predictive maintenance is critical for ensuring the efficiency and reliability of machinery, particularly ventilation systems used in automotive production. These systems are integral to maintaining air quality and optimal operational conditions, significantly affecting product quality and worker safety. The challenges faced in this environment include the need for continuous monitoring of multiple variables, real-time data transmission, and efficient power management to extend battery life.
The hardware developed for this paper uses a commercially available microcontroller prototype called the Portenta H7 [6]. For sensors, it includes an industrial vibration sensor from Kemet [7], a typical PT1000 sensor [8] for temperature measurement, a current transformer for power measurement, and a humidity and air quality sensor for environmental monitoring. The software is written in an embedded C language, while data analysis and AI model training are carried out in Python.
From a research perspective, for hardware comparison, there are several wireless platforms based on LoRa, such as the WISE 2410, or based on Bluetooth, like the Bosch XDK. The proposed hardware draws inspiration from the sensor platform presented in [9]. The sensor platform of this research paper focuses on 5G aspects but not on applications that demand longer runtimes. By incorporating hybrid connectivity technology, the hardware becomes even more versatile and flexible for various applications.
The hardware used for the multi-sensor setup combines dual-core capabilities to switch between runtime requirements and also handle AI tasks. For performance evaluation, hardware capability is tested by measuring the computation time for inference, the time for data acquisition, and power consumption in both standalone and hybrid modes, along with any performance drop.
3. Literature Review
This study presents a hybrid LoRa-5G multi-sensor platform designed for predictive maintenance in industrial ventilation systems. Although recent research has explored similar themes, the work distinguishes itself by integrating edge processing and low-power and high-speed communication technologies to improve data transmission efficiency and operational flexibility.
Romanssini [10] explored the predictive maintenance of motor equipment using vibration sensors and LoRa/WiFi/BLE communication and cloud-based analytics. While the focus was on the possibility of predictive maintenance, it did not explore real-time detection as the processing was carried out in the cloud. In addition, power consumption was not considered, which is significant for deployment in real-world applications. Our platform considers power consumption using both LoRa and 5G while bringing the necessary performance and AI analytics possibilities.
Another study by Josh Lentz [11] also focused on predictive maintenance using LoRa-based communication, developed for Kellogg. This methodology explored LoRa as low-power and low-cost deployment, but did not consider the real-time computational requirements and edge processing needed to provide real-time alerts in case of anomalies since it only used LoRa. Our platform incorporates both LoRa and 5G to obtain real-time data transmission for alerts and data for the AI model.
Chen et al. [12] explored a LoRa mesh–5G integrated network for track-side smart weather monitoring, highlighting the potential of hybrid networks for environmental applications. However, their focus was primarily on weather monitoring and cannot be transferred to ventilation systems with different requirements. Using AI-driven analytics, our platform proactively addresses equipment failures, extending the operational life cycle and improving efficiency in industrial settings.
Busacca et al. [13] introduced “FedLoRa: IoT Spectrum Sensing through Fast and Energy-Efficient Federated Learning in LoRa Networks”. This research focused on federated learning for optimizing spectrum sensing in LoRa networks. Although the energy efficiency aspect aligns with battery-operated sensor platforms, our approach integrates real-time data processing and analysis capabilities, combining the strengths of LoRa for low-power communication and 5G for high-bandwidth data transfers. This hybrid methodology provides a more versatile solution for predictive maintenance applications, which was not the primary focus of Busacca et al.
4. Predictive Maintenance for Industrial Production
4.1. Use Case Description
Equipment maintenance is crucial for industry. It ensures that the manufacturing process runs smoothly and that the product’s quality is sufficient. Maintenance helps manage and reduce unanticipated downtime of machinery and equipment, thereby extending their useful life [14]. Traditionally, this process is performed periodically (planned maintenance) or when the equipment breaks down or does not perform well (reactive maintenance). These maintenance procedures have some disadvantages. Scheduled maintenance can lead to over-maintenance of the equipment, incurring significant costs, and a drastically reduced lifespan of the equipment. Reactive maintenance leads to significant downtime and cost to maintain.
A proactive and efficient strategy is essential for addressing this problem. Various methods, including predictive maintenance and condition-based maintenance, involve analyzing the condition of the equipment to predict failures and schedule maintenance accordingly. These approaches are highly advantageous for industries, as they can significantly minimize unexpected downtime and costs while extending the life of the equipment [15]. Predictive maintenance and condition-based maintenance involve integrating sensors and IoT platforms to gather the parameters that affect the operating condition of the equipment. Additionally, data analytics such as feature extraction, algorithms to understand the condition, and AI models assess the system’s condition and provide alerts to the operator.
In the context of automotive production, the maintenance of the ventilation system is critical, particularly in the assembly section where paint is applied to finished car frames. The effectiveness of the ventilation system directly influences the quality of the paint finish. It is responsible for maintaining proper airflow into the painting chamber, where air is mixed with the appropriate amount of paint solution to achieve the desired aesthetic results. In addition, the system plays a pivotal role in ensuring worker safety by filtering out toxic fumes generated at the plant. If the ventilation system malfunctions or fails, the paint may mix with contaminated air, leading to substandard finishes that require reworking [16].
Several factors can contribute to the deterioration of the airflow of the system, such as bearing wear, corrosion of the system due to high humidity or temperature, accumulation of dust in the ventilation ducts, and the need to replace filters. To maintain sufficient airflow, maintenance is necessary. One of the most common problems is bearing degradation or breakage, which is hard to analyze. To overcome these, periodic maintenance is performed with a frequency of 2–3 months. Since bearing condition cannot be accessed immediately, there is a higher chance of over-maintenance, resulting in significant maintenance costs and downtime.
Cases like this can benefit from predictive maintenance that uses smart sensor technology, data analysis, and AI models. Sensors are integrated into the ventilation system to collect vibration data and environmental parameters that affect the machine’s function. These data are then transmitted to a cloud or remote system, where they are analyzed and certain features are extracted. Based on these features, an AI model is generated to monitor and evaluate the machine’s condition. This allows us to predict when the ventilation system will fail and schedule maintenance work accordingly. This significantly reduces costs by reducing the frequency of maintenance. It also extends the life of the equipment, increasing overall efficiency and reducing costs [17].
4.2. Challenges
Although predictive maintenance provides substantial advantages, various factors impede its widespread adoption in the industry. An overview of hardware implementation challenges, data quality assurance, and security measures is given in the following subsection.
4.2.1. Hardware Implementation Challenges
Designing a platform for predictive maintenance systems for industrial applications involves several hardware-related challenges that need to be addressed to ensure performance, scalability, and reliability.
Sensor Selection and Integration
Choosing the right sensor to monitor the condition of the equipment is important. Since various parameters affect the condition of the machine, various sensors must be deployed to measure parameters such as temperature, vibration, humidity, and pressure. Integrating these into the hardware is challenging, as some sensors need to be in close proximity to the equipment and exposed to environmental conditions while having a smaller size. Ensuring that sensors operate reliably in these harsh conditions can be challenging.
Data Acquisition
Continuous data acquisition from sensors is important for predictive maintenance. It is challenging to develop a hardware system that can handle high data throughput from multiple sensors and simultaneously process and store data. Especially in environments where power resources are limited, a balance between processing power and energy efficiency must be found.
Power Management
Many industrial applications are difficult to access and rely on battery-powered sensor platforms for predictive maintenance and data access. Efficient power management is critical to prolong the operational life and reduce the maintenance costs of sensor hardware. Designing a sensor platform to minimize power consumption while maintaining performance is challenging and requires sophisticated software to ensure the operational lifetime.
Communication Limitations
Effective communication between sensors and the central system is fundamental for real-time data analysis and decision-making systems. However, the choice of communication technology imposes limitations on range, bandwidth, and power consumption. The selection of an appropriate communication method to balance the above is crucial.
Additional challenges, such as interoperability with existing systems for seamless integration and platform resilience to harsh industrial environments, also impede the design of sensor platforms.
4.2.2. Data Quality Assurance
Data quality is crucial in AI-based predictive maintenance applications, as it directly impacts the performance of AI algorithms and their prediction results. Accurate and high-quality data enable effective anomaly detection and informed decision-making regarding maintenance schedules. Poor data quality can lead to incorrect predictions, resulting in unnecessary maintenance or unexpected equipment failures. Several aspects must be taken into account throughout the process to ensure better data quality.
Data Acquisition
The system requires a range of sensors to monitor critical parameters that continuously affect the condition of the equipment. These sensors must be calibrated regularly and maintained to ensure accurate data collection and reduce errors.
Data Validation
The acquired data must be validated to minimize errors and ensure higher quality before analysis. This involves integration techniques such as anomaly detection, filters to reduce noise, etc. The validated data need to be timestamped for further processing.
Data Processing and Analysis
The conditioned data must undergo preprocessing before being analyzed. The preprocessing involves feature extraction such as Fourier transformation and sensor fusion. This enhances the dataset and provides additional features for more accurate data analysis. AI algorithms can then analyze the data to identify patterns and inform maintenance decisions. In addition, these algorithms must be retrained periodically to secure and improve the accuracy of prediction. These computations are resource intensive and require suitable hardware.
Continuous Monitoring
Processed data must be promptly communicated to the end application to facilitate effective decision-making. Given the unpredictable nature of condition monitoring, real-time data analysis is essential. In the event of anomalies, the system should alert users to enable detailed analysis and timely decision-making.
Remote Access
To enhance versatility and robustness, the system should provide remote access for the reconfiguration of parameters such as sampling rates and sensor periodicity and system updates (firmware and security). This capability allows the system to adapt to evolving application demands.
4.2.3. Security Measures
Data security is critical in predictive industrial maintenance as it involves sensitive operational data transmitted to a central system over a wireless communication medium. Wireless transmission exposes this system to various threats, such as interception, manipulation, and unauthorized access. To manage these risks, several strategies are implemented to mitigate possible impacts and improve overall safety.
Data Encryption
By encrypting the data before transmission, the risk associated with interception can be minimized. The encryption needs to be lightweight to maintain real-time performance, reducing overhead without compromising security. Physical layer encryption would be even more beneficial, as it provides additional security without creating additional overhead.
Secure Communication Protocols
Implementing protocols such as Transport Layer Security (TLS) can provide additional security for data transmission and ensure data integrity.
Security Updates
Periodic OTA updates are vital for deploying security patches and enhancements. This practice helps the system remain resilient against emerging threats.
5. Architecture Overview
To address the challenges that arise from the sensor system, a new hybrid IIoT sensor platform is presented in Figure 1 that offers a comprehensive solution for predictive maintenance in industrial settings. This new platform leverages the potential of IIoT technology with multiple sensor interfaces, LoRa, and 5G network capabilities to provide real-time monitoring and low power consumption.
The platform leverages LoRa technology for reliable long-range, low-power communication, making it ideal for remote or challenging industrial environments. In addition, the integration of 5G technology ensures high-speed data transfer and low latency, enabling real-time monitoring and analysis of equipment performance. Together, these technologies provide seamless connectivity and robust data transmission.
5.1. Hardware Components
The architecture of the hybrid LoRa and 5G multi-sensor platform is described in detail in Figure 2. The sensor platform comprises a core microcontroller, based on Portenta H7, on-board sensors for environmental monitoring and diagnostics, and an external vibration and temperature sensor that is connected to the equipment or machine. For communication, the board has a LoRa module and an Ethernet interface to which a 5G router is connected. The whole board is powered by a battery.
5.1.1. Microcontroller
The core of the multi-sensor platform is the microcontroller. It is responsible for data acquisition from multiple sensors and processing and transmission to the remote central location. Usually, a low-power microcontroller such as the TI MSP430 or STM32L series is preferred for applications that require a long battery runtime. For predictive maintenance applications that require more computational resources, powerful microcontrollers such as the NXP IMX series or single-board computers such as the BeagleBone Black, Jetson Nano, and Raspberry PI are preferred. However, these are power hungry and cannot rely only on a battery. A microcontroller must be chosen to handle the processing demands optimally while increasing battery runtime with low power consumption. Microcontrollers such as STM32F4 and Atmel SAMD are suitable options. The new Arduino pro-family microcontroller, the Portenta H7, is a good option based on the STM32H7 dual-core microcontroller. This effectively provides the resources needed for the acquisition of real-time data from multiple sensors and processing, and it can be programmed to consume little power when idle.
5.1.2. Sensor Selection
Predictive maintenance requires the integration and analysis of a diverse array of sensors. The interfacing of multiple sensors complicates the hardware design, as each sensor requires specific conditioning components, including signal filters, amplifiers, and analog-to-digital converters (ADC). Therefore, careful consideration and research are essential for selecting the appropriate sensors for industrial applications. For the ventilation system, critical variables that influence bearing conditions include induced vibration and stator temperature. In addition, monitoring parameters such as current consumption and airflow rate are important. These measurements are taken near the equipment, so external sensors are utilized and connected to the hardware via wired interfaces due to the size restrictions of the system and the hardware developed. For vibration monitoring, the resonant frequency of typical ventilation systems falls within the frequency range of 7–10 kHz. The Kemet vibration sensor is selected for this application, as it can detect frequencies ranging from 10 Hz to 15 kHz. The sensor is based on the piezoelectric principle and features an integrated amplifier, which simplifies the design by reducing the number of external components needed. The sensor outputs an analog signal that is digitized by the microcontroller-integrated ADC. An industrial-grade PT1000 sensor is employed for temperature sensing. The K-type thermoelectric transducer is chosen for its sensitivity and robustness, which make it suitable for monitoring high temperatures in challenging industrial environments. Additional interfaces such as serial Universal Asynchronous Receiver Transmitter (UART) communication and a Serial Peripheral Interface (SPI) are implemented to measure current and airflow rates, allowing for direct connection of the necessary sensor modules. In addition to the sensors mentioned above, several on-board sensors are incorporated to address other monitoring needs. The hardware is designed to accommodate additional sensors for applications like environmental monitoring. Parameters such as humidity and air quality are essential to track. Proper air quality ensures that the paint adheres correctly and reduces defects, while humidity levels can affect drying times and the quality of the paint application. Monitoring these parameters helps maintain optimal conditions and prevent problems. The SGD30 sensor, which measures CO2 and humidity, is interfaced with the microcontroller via the inter-integrated circuit (I2C) interface to facilitate this. The board also includes an on-board BME680 sensor, which measures humidity and temperature. This sensor provides critical diagnostics for the hardware, such as by monitoring board temperature and detecting potential water intrusion.
5.1.3. Connectivity Modules
The predictive maintenance application requires dynamic communication with the ability to transmit data over long distances while consuming low levels of power and providing real-time communication with a higher bandwidth. To achieve this, the hybrid multi-sensor platform integrates two key connectivity technologies: LoRa and 5G. WiFi is a well-known alternative, but, for remote deployment, the need for additional access points to cover the device range and connection to the central network increases the cost and network complexity. Fourth-generation cellular technology provides the needed convergence but lacks low latency and real-time data transfer with high throughput. In addition, this hardware technology module consumes more power and is typically not preferred for wireless IoT sensors.
5G Connectivity
To complement LoRa, 5G technology is integrated into the platform to enhance its capabilities with high-speed, low-latency communication. 5G supports significantly higher data rates compared to conventional cellular networks, making it suitable for applications that require real-time data analytics and immediate response actions. The high bandwidth of 5G facilitates the simultaneous transmission of data from multiple sensors, allowing effective management of the complex datasets generated by predictive maintenance systems. However, these advantages come with the trade-off of higher power consumption, which is slightly lower than that of 4G technology. The hardware 5G modules are typically available as M.2 interface modules or Land Grid Array (LGA) modules, requiring additional hardware designs to expose the necessary data communication lines. A common approach is to use a universal serial bus (USB) interface, which is relatively easy to implement in hardware. However, this requires that the host system has the appropriate drivers to establish the 5G connection. Additionally, the USB protocol introduces significant overhead that can restrict real-time performance, especially in scenarios requiring low-latency interrupts. An alternative approach is to connect the 5G modules via Ethernet. The Ethernet stack is considerably lighter than a USB, and many microcontrollers are equipped with a small Ethernet stack that can be utilized to connect these modules efficiently. This method minimizes protocol overhead, allowing for improved real-time performance and responsiveness in data transmission. For the ventilation unit monitoring application, the Quectel 5G M.2 RM520N module [18] is used, along with M.2-to-Ethernet converter hardware, to act as the 5G modem. The microcontroller has an Ethernet interface through which the 5G modem is connected.
LoRa Connectivity
LoRa technology is used for its low-power, long-range communication capabilities, making it ideal for industrial settings where sensors may be scattered over large areas. Operating in the 868–915 MHz frequency bands, LoRa enables devices to communicate over distances of several kilometers while minimizing energy consumption. The LoRaWAN (LoRa Wide-Area Network) protocol facilitates communication between sensor nodes and the central gateway and enables bidirectional data transmission. The LoRa E5 module from Seed Studio is used for the hardware module. It is based on the Semtech SX1272 and STM32 microcontroller, and the host Portenta H7 communicates with it via the UART interface.
5.1.4. Battery Management System
A Battery Management System (BMS) is essential and complex for hardware that incorporates multiple sensors and connectivity modules. The BMS ensures that the battery can provide the power output required by the microcontroller, sensors, and communication modules. It also safeguards against over-discharge and under-voltage conditions, which can pose fire hazards to the hardware. These battery systems must also be charged periodically during maintenance to ensure seamless operation. Given hardware runtime requirements, it is crucial to monitor both power consumption and battery levels. The control system can utilize this information to optimize data transmission methods over 5G or LoRa, select active sensors, and deactivate unnecessary sensors. For hardware implementation, the MP2723 BMS Integrated Circuit (IC) regulates a single-cell lithium-ion battery, providing a 5 V output at 2.4 A while ensuring compliance with the necessary battery safety protocols. An INA219 sensor measures the power draw from the battery to monitor power consumption. In addition, an MAX17048 fuel gauge IC is used to track the state charge percentage of the lithium-ion battery, facilitating informed decision-making about battery usage and maintenance.
The developed hardware is shown in Figure 3 below.
5.2. Software Components
Data acquisition from multiple sensors on a hybrid LoRa 5G platform must accommodate varying sampling frequencies, depending on the application. Sensors like vibration monitors may require high-frequency sampling, while others, such as temperature sensors, can function with lower sampling frequencies. Preprocessing steps such as noise reduction and frequency analysis ensure data quality and reduce the number of transmitted data. For condition monitoring, real-time processing is critical, necessitating immediate transmission to the cloud for analysis. In contrast, predictive maintenance allows data to be sent periodically, with or without preprocessing, depending on the system requirements.
With advanced microcontrollers like the Cortex M7 and dual-core processors, on-device data analysis becomes feasible. This edge computing approach minimizes transmission costs, improves energy efficiency, and reduces reliance on cloud infrastructure.
However, transmission power consumption is a crucial factor. Although 5G technology offers high data throughput and low latency, it consumes significantly more power than LoRaWAN or BLE. The platform must balance these trade-offs, leveraging LoRa for low-priority, low-frequency data and 5G for applications requiring high-speed, low-latency communication. This hybrid approach enables a versatile and energy-efficient system that adapts to diverse application demands while optimizing performance and resource usage.
Hence, robust reconfigurable software is needed to provide the application firmware for data acquisition, processing, and analysis and manage hybrid connectivity to balance performance and energy savings. The software, equipped with a real-time operating system, plays a crucial role in handling various sensor data acquisition tasks and buffering them for further analysis or transmission. An IoT lightweight machine-to-machine protocol (LWM2M) and an over-the-air update are integrated to manage the software application. This enables the remote user to customize the application according to the requirements. The software then triggers the data transmission to the end cloud application through the hybrid connectivity modules based on the data type, such as analyzed/condition data and raw data. Simple conditional and analytical data are transmitted by the software over LoRa, whereas the preprocessed/raw data are accumulated and sent as a burst over 5G.
For predictive maintenance of ventilation systems, data from environmental sensors such as humidity, barometric pressure, temperature, air quality, and CO2 sensors are sent over LoRaWAN with a cyclic periodicity of 10 min.
5.3. Software Architecture
Due to the complex application involved in predictive maintenance with multiple sensors and multiple connectivities and running a battery-powered sensor platform, the firmware should be aware of energy and provide the necessary performance for aspects such as data acquisition, processing, and transmission of processed data and updates to the end application. The software architecture for such an application is shown in Figure 4.
5.3.1. Data Acquisition and Processing
This module plays a vital role in the application by facilitating the acquisition of multiple data streams. This includes a series of steps: setting up the sensors and their configuration and preparing interfaces and data buffers to acquire and store the data.
Sensor Preparation
This step involves using the microcontroller to enable and set up various sensors on the hardware according to the application. For predictive maintenance of the ventilation system, the vibration sensor is set up to sample at a rate of 40 kHz. Other sensors, such as temperature, humidity, and pressure sensors, are sampled every 10 min. In addition to core sensors, additional diagnostic sensors, such as the IMU on board and temperature, humidity, fuel gauge, and current sensors, are sampled at a lower frequency.
Data Buffer and Acquisition
To acquire the data, the necessary interfaces and data buffers are configured to enable seamless data acquisition. The microcontroller uses Direct Memory Access (DMA) to receive data without loading the CPU, allowing it to be used for data processing and analysis. To enable real-time edge processing, the Cortex M4 of the microcontroller is used for data acquisition, while the Cortex M7 is used for data analysis and transmission.
Data Processing
To provide more insights, certain sensor data, such as vibration data, undergo preprocessing using feature extraction techniques such as Fast Fourier Transform (FFT) and statistical analysis. These preprocessing techniques reduce the data throughput while providing critical parameters that can determine the condition of the equipment.
5.3.2. Data Analysis
The processed sensor data are then analyzed on board before storage. This is performed to reduce the amount of data transmission required while providing real-time edge analytics. Typically, the analysis is performed on the basis of AI algorithms or on a rule-based basis, depending on the sensor type. Since the microcontroller used is resource constrained, the AI models need to be converted to be able to be deployed on the hardware. Techniques such as TinyML can be used to convert the model and deploy it on the controller. The model is deployed as a thread on the Cortex M7 core. For predictive maintenance, the time and frequency data from the vibration sensor are used as input for the model, where the model is pre-trained with the dataset from the same sensor. For other sensors, simple threshold filters are used to analyze the data and store them along with their timestamps.
5.3.3. Energy-Aware Computing
Due to the complex application involved in predictive maintenance with multiple sensors and connectivity options, running a battery-powered sensor platform requires energy-aware firmware that provides the required performance. For better energy optimization, the microcontroller is programmed to utilize the power-mode feature of the STM32. The Cortex M4 is activated only during data acquisition and then transitioned to sleep mode. The Cortex M7 is also configured to analyze data with a configurable periodicity, after which it also enters sleep mode. This approach effectively reduces the overall power consumption of the controller. The use of Direct Memory Access (DMA) further minimizes the wake-up time of the microcontroller and enables real-time data acquisition. The connectivity modules are programmed to enter sleep mode during idle times and are activated before communication. The LoRa module is employed to send periodic updates of the data analysis at the edge as it consumes less power for transmission and the payload data are small. The 5G module is kept in sleep mode and is awakened either upon data reception from the central monitoring system or when the microcontroller transmits data. The 5G module provides real-time data streaming from sensors with low latency and high reliability when needed, or it can enable burst transmission once a week, taking advantage of its large bandwidth. This optimized usage reduces the power consumption of the battery, thereby extending the runtime. Data transmission is performed using the UDP over a dedicated port as it allows more payload, and the use of 5G enables reliable communication with low latency. The choice between real-time transmission and burst transmission depends on the remote monitoring system application. By having a burst transmission, the accumulated data can be transmitted to the cloud for remote, permanent data storage. In addition, these data can be retrained in the cloud for better performance of the AI model. Real-time streaming is enabled when the remote monitoring system operator requires real-time condition monitoring to analyze critical failure or the current state of the equipment. This allows real-time data analysis over the cloud. Additionally, when an anomaly is detected and read back over LoRa, the 5G can be enabled for the real-time streaming of sensor data for further analysis, making the hardware more versatile and adaptable.
On-board diagnostics, such as Battery Management System (BMS) sensors and humidity and temperature sensors, are also analyzed periodically, providing the necessary diagnostic information such as battery health and status to the central monitoring system. Operators can use this information to estimate runtime and configure the platform by enabling or disabling sensors and data analysis to save energy or provide real-time analytics when needed for cloud-based detailed analysis.
5.3.4. Remote Configuration and OTA
For a hybrid LoRa 5G multi-sensor platform, the ability to reconfigure and perform OTA updates is crucial to maintain system performance, enhance features, and ensure security. The following aspects outline the mechanisms and benefits of reconfiguration and OTA updates within this platform.
Reconfiguration
Reconfiguration involves the ability to modify the operational parameters of the sensors and communication modules according to changing environmental conditions or user requirements. This can include adjusting data sampling rates, changing communication protocols, or enabling/disabling specific sensors. The microcontroller supports dynamic reconfiguration through a protocol such as MQTT over a dedicated port. The configuration parameters are set as topics to which the microcontroller is subscribed. The central core monitoring system can publish it over the 5G network.
OTA Updates
OTA updates enable the deployment of new firmware or software enhancements without the need for physical access to the devices. This feature is particularly valuable in remote or hard-to-reach locations where manual updates would be impractical. OTA enables firmware updates that are crucial to the performance of the hardware for maintenance. This could be a new software module for data acquisition, feature extraction, an update of the AI model on the hardware, or security patches. The update starts with the firmware receiving an update request over 5G, upon which it enters into bootloader mode, receives the data over 5G, updates the flash memory module, and boots up. The firmware number is then sent over to the core system via LoRA or UDP. The NXP SE050C2 and ATECC608, on-board security ICs, enable OTA updating with secure boot and TLS encryption for firmware update requests.
5.3.5. Secure Data Transmission
Since security is mandatory when transmitting sensitive industrial data, the processed data are encrypted first. Although there are software-based solutions, the Portenta H7 is equipped with ICs of secured elements on board, namely, the NXP SE050C2 and the ATECC608, which provide real-time encryption without overhead. Together, these components provide essential security features, such as secure boot and TLS authentication over protocols such as UDP, TCP, and MQTT, as well as encryption of OTA updates. The ATECC608 offers digital signatures that ensure data integrity and facilitate device authentication, securely manages cryptographic keys, and can generate signatures to verify the authenticity of the data transmitted. Meanwhile, the NXP SE050C2 is responsible for the generation of private keys and the generation of random numbers, which can be utilized by the core microcontroller to encrypt the data. The keys generated by the SE050C2 are stored in secure memory, which can be remotely accessed via secure TLS over MQTT, allowing the central monitoring system to secure key management and retrieval. The encrypted data are transmitted over UDP and LoRa, minimizing the risk of man-in-the-middle and eavesdropping attacks. The central monitoring system can then access and decrypt the data safely, ensuring that sensitive information remains protected throughout the transmission process.
6. Validation
6.1. Validation Methodology
The validation methodology used in this study focuses on assessing the performance of the LoRa 5G hybrid multi-sensor platform through a comprehensive analysis of various performance metrics. The following metrics are referenced from previous studies, ensuring that our findings are contextualized within the existing literature and promoting the benefits of the platform. Some of the metrics are measured and presented.
To evaluate the effectiveness of the platform, we utilize key performance metrics defined in prior research, which include the following:
Data Accuracy: This measures the accuracy of the acquired data from the sensor platform against the existing solution or off-the-shelf solution. For predictive maintenance, in this particular use case, the most important sensor is the vibration sensor. The other sensors are also necessary for embedding the vibration data in the current environmental conditions. These sensors are digitalized according to the sampling requirement. The performance accuracy of these sensors is already specified in their respective datasheet [19], which meets the industrial requirements for predictive maintenance based on vibrations [20].
Reliability: The reliability of data transmission is assessed by monitoring the uptime and success rates. The reliability of data transmission depends on the communication technology. When evaluated, 5G provides more reliability [21], enabling reliable burst data transmission. The reliability of LoRa depends on the spread spectrum factor used, as explained in [22]. The firmware is programmed to use the higher spread spectrum factor and is expected to achieve the same accuracy.
Anomaly Detection Rate: The detection of anomalies largely depends on the training model and the dataset used. Since the hardware for the application is developed and deployed for data acquisition, the actual dataset cannot be used, and, hence, the Tiny ML AI inference of the system cannot be evaluated on the actual dataset. The hardware is expected to perform at the same level as discussed in the other literature where the board is evaluated [23,24].
Energy Consumption: The amount of energy consumed during operation is crucial for evaluating the efficiency of the platform, particularly for battery-powered devices. Continuous monitoring of energy use is performed during both normal and fault conditions with data collected from the current sensor (INA219) and the battery state of the charge monitor (MAX17048). This metric provides insight into the operational sustainability of the platform. This is compared and presented in the following sections.
6.2. Experimental Design
The platform is evaluated against existing standard solutions and the configurations of sensor and network systems commonly used in industrial applications to evaluate its performance. The primary focus of this evaluation is to compare the performance with respect to current wireless solution such as Narrow-Band IoT (NB-IoT), WiFi, LoRa, and 5G-based sensor platforms. Typical current solutions in industries include LoRa-based systems such as the WISE 2410, based on LoRaWAN [25], Dragino’s NB-IoT-based hardware NBSN95 [26], which can be used to combine with a low-power sensor, and the Bosch NextGenXDK [27] multi-sensor platform, which uses Bluetooth or WiFi for data transmission. Devices from other vendors include the Beanair 2.4 GHz [28] wireless multi-sensor, which is based on low-rate WPAN, IEEE 802.15. In contrast, 5G-based solutions consist mainly of prototype research hardware, such as the 5G MSP [9]. Since each commercial platform comes with its own sensor configuration and limited flexibility, empirical analysis cannot be performed, and hence a mathematical analysis based on the datasheet of the sensor hardware is performed. For the prototype hardware, the measurement of power draw is performed using experimental hardware consisting of a simple RP2040 [29]. Microcontroller, a current sensor based on INA219, a fuel gauge IC based on LC7092, and a battery management module (MP2723) to safely output 5 Volt from a 2000 mAh Li-Ion battery. The setup is shown in Figure 5. The hardware to be evacuated is connected to the USB output.
The following hardware components are evaluated:
The 5G MSP board developed by the authors [9];
The hybrid hardware solution.
Power consumption is calculated by multiplying the current measured by the voltage. The estimated time is based on the capacity of the battery used, which, in this case, is a 2000 mAh Li-Ion battery.
The Portenta H7 hybrid platform is programmed to run in run mode during data processing and analytics and is then moved to stop mode. The M4 core of the cortex is used for data acquisition. The LoRa module used is an SX1272 Semtech module, and, for 5G data communication, the Quectel RM520N module is used with an M.2-to-Ethernet breakout shield. The module can be power controlled from the Portenta, to reduce power consumption. Since it is difficult to measure the power consumed for the setup with the configuration of sending data over LoRa and 5G for burst transmission, only the Portenta H7 module with Ethernet and LoRA enabled is used to measure the power. The 5G power measurement is carried out separately and then combined to provide the estimated runtime.
Impact of Propagation Path on Power Consumption
One thing to consider before calculating the energy consumption is the impact of the propagation path on the communication system. Unlike a wired system, power consumption in wireless IIoT platforms, such as LoRa and 5G, is influenced by the propagation path length due to the need for stronger signal strength and the efficiency of the link parameters.
For 5G, the modulation and coding scheme (MCS) has a similar impact on power usage. Higher-order modulation, such as 64-QAM, is more power efficient for short distances but requires lower-order modulation, such as QPSK, over longer distances to overcome weaker signals. This change increases the number of transmissions, which, in turn, increases the power consumption. Furthermore, transmission power control in 5G allows devices to adjust power levels based on signal strength, increasing transmission power for weak signals, thus enhancing the signal quality, but at a higher energy cost. Carrier aggregation, which uses multiple frequency channels for higher throughput, also increases power consumption, particularly in areas with poor network coverage [30]. While the choice of modulation depends on the hardware module, the aggregation and transmission power of the carrier depends on the network and the quality of service. For the used 5G module RM520, a rel 16 device, it supports both modulations. The carrier aggregation is not used and only the band n78 is configured. Based on this, the measured power consumption of the RM520N Quectel module is approximately 250 mA at the 5 V supply, resulting in approximately 1 W. The throughput measured using iperf is around 90–100 MBits/s.
In the case of LoRa, the propagation path directly affects power consumption. As the distance between the devices increases, signal attenuation occurs, requiring LoRa devices to boost their transmission power to maintain a reliable link. However, this compensation comes at the expense of battery life. Additionally, the Spreading Factor (SF) plays a significant role; higher SF values extend the communication range but increase transmission time, leading to higher energy consumption due to longer airtime. In addition, bandwidth and coding rate settings are crucial. Although a narrower bandwidth and higher coding rates improve the range and reliability of the system, they also extend airtime and increase power usage, making them a balancing act for energy efficiency in long-range communications [31]. For the hybrid platform, an SF of value 8 and a bandwidth of 125 KHz are used to provide better reliability and a transmission power of 17 dBm to cope with the long distance considering the field of deployment. The configuration provides an optimum current consumption of approximately 50 mA.
To model the power consumption, the timing diagram of the firmware provides an idea on the flow of information between sensors, microcontrollers, and the communication modules involved. The Figure 6 provides the flow of data between the modules. Based on this, the power consumption is calculated.
Figure 7 represents the measured power consumption of the Portenta H7 in the run and stop modes. The average runtime power is around 220–270 mA with a programmed clock frequency of 480 MHz. LoRA transmission, which typically consumes around 50 mA, is included in the baseline power consumption and is represented as spikes. The data point without spikes shows the power consumed when the Portenta H7 is in the active data acquisition, processing, and analytical stages. In stop or sleep mode, the measured power drops to 15–20 mA. In this stage, all peripherals are disabled, except DMA and interrupts, and are programmed to automatically start in a set period.
Assuming a vibration sensor sampled at 40 kHz with a data array size of 1024 16-bit integer values, the amount of data stored in the controller storage is around 7 GB (40,000 samples/sec X 2 Bytes/sample X 86400 (sec in a day)/8). Using 5G, this would take around 560 s to transmit as a UDP packet. The power consumed for this transmission is calculated as follows:
This is the power consumption of the communication burst transmission. Now, the sensor also acquires data with DMA and is analyzed once every 15 min. Suppose processing and anomaly detection takes 15 min and uses LoRa to transmit the status and go back to stop mode. If data processing and analytics are assumed to take 10 s, the effective runtime power in an hour is estimated to be
With a burst transmission occurring once a day, the approximated runtime with a 2000 mAh battery is around
With a 2000 mAh battery, the expected runtime is around 59 h.
The comparable hardware mentioned requires mathematical analysis.
Equations for Energy Consumption Analysis:
For each piece of hardware, the energy consumption per hour is calculated using the following equation: , where
is energy consumption per hour (mAh);
is current draw (mA);
is the number of transmissions per hour.
The estimated runtime is calculated as follows: , where
is estimated runtime in hours;
is battery capacity (mAh);
is energy consumption per hour (mAh).
Based on this, the following considerations are made. The battery capacity is set to 2000 mAh. The maximum current draw is set on the basis of the hardware datasheet, including the communication module. The values include sleep time and active time (data acquisition and transmission). Since the power consumption of the communication module varies according to signal strength and modulation, the maximum power consumed is considered. Based on the equation, the following energy consumption analysis (Table 1) is calculated. Since each piece of hardware has its own transmission interval, to keep it simplified, all of them are said to have the same transmission frequency and application exec. In addition, the execution time is kept constant as they all use similar low-power microcontrollers.
According to the table, the energy consumption varies significantly according to the communication module and the transmission frequency. For the 5G-MSP, as expected, the hardware consumes significantly more current due to the 5G module and has a reduced runtime. While other hardware, such as WISE, Beanair, and NextGenXDK, performs better in terms of energy efficiency, it lacks bandwidth, latency, and scalability aspects. Additionally, these commercial solutions do not provide the edge computational requirement needed for application.
6.3. Cost Efficiency
Cost efficiency is a crucial factor in the implementation of IIoT platforms, balancing hardware performance, power consumption, and financial viability. It is calculated on the basis of the hardware cost, the data transmission cost, power efficiency, or the runtime of the battery. Often LoRa or WiFi or BLE is more cost efficient, whereas cellular connection is not cost efficient due to the need for cellular connectivity, simcards, and data packs, which add to the cost in a scalability scenario.
-
WISE-2410: High cost efficiency driven by its long battery life, support for vibration monitoring, and the well-established LoRaWAN protocol.
-
Beanair BND-2.4 GHZ: High cost efficiency with predictable energy consumption and consistent performance in data streaming applications.
-
Bosch XDK110: Offers moderate cost efficiency due to its wide range of supported sensors and flexible communication options and use of a noncellular network, but its higher power consumption and shorter battery life impact long-term operational costs.
-
NBSN95: Moderately cost efficient with its long-range communication and relatively low data rates, but the need for a cellular connection and data pack may quickly add to the cost, and, if the application requires frequent data transmission, it becomes less efficient.
-
5G-MSP: Lower cost efficiency due to high power consumption, shorter battery life, and the need for a cellular network, although it provides unparalleled data rates and real-time capabilities.
-
Hybrid LoRa–5G Platform: Moderate cost efficiency due to the balance of low-power LoRa and high-speed 5G, offering robust performance but increased complexity and cost due to 5G and cellular connections.
7. Discussion and Limitations
To simplify the validation, the following observation (Table 2) is charted based on the available data pertaining to performance aspects, cost, and energy efficiency.
From the measurements, the hybrid solution demonstrates an optimal approach that takes advantage of the 5G and LoRa technologies efficiently. Although the hybrid system cannot match the longevity of the LoRa-only system, it offers significantly improved power performance and versatility. The dual-core microcontroller employed in the hybrid setup is capable of performing critical tasks such as data analysis and feature extraction while maintaining low power consumption. This efficiency is further enhanced by using only one core for data acquisition, resulting in energy savings. The periodic burst transmission facilitated by the 5G network drastically reduces power consumption during data transmission, a capability that the LoRa-based system lacks. Furthermore, the LoRa system does not meet the requirements for reliable data transmission and does not perform the advanced analytics that the hybrid solution offers. A low-power controller, such as that used in LoRa systems, cannot effectively handle data analytics without creating bottlenecks in data acquisition and transmission processes. By integrating data analytics, including feature extraction and AI analysis, directly onto the hardware, the system minimizes the volume of transmitted data, enabling real-time processing. In the event of failure detection, the sensor platform can promptly transmit warning signals via the 5G network, allowing operators to respond immediately. This flexibility effectively compensates for the shorter runtime of the sensor platform, emphasizing its practical benefits in predictive maintenance applications. Additionally, the sensor platform’s capability for remote reconfiguration, utilizing OTA updates and lightweight M2M (LwM2M) protocols, enhances its versatility, positioning it as a viable solution for Industry 4.0 applications.
The hybrid approach is not without its disadvantages. One of the primary drawbacks is the increased complexity of the system architecture. The combination of two different communication technologies complicates the design and creates issues of interoperability. Moreover, despite the advantages of improved data transmission speed and reliability, the higher power consumption of the 5G module can detract from the overall energy efficiency of the system. Although LoRa is designed for low-power applications, the hybrid system’s reliance on 5G for real-time data bursts introduces the risk of rapid battery depletion, particularly in scenarios where frequent data transmission is required. This balance between performance and battery life must be carefully managed to optimize the system’s operational sustainability. The need for efficient power management strategies is paramount to ensure that the system can maintain extended operational periods while effectively utilizing both communication channels.
However, it is important to note that the hardware could not be tested with the actual AI model due to the absence of a training dataset. The performance of the AI model will be evaluated once the necessary data are acquired, and the results are expected to be consistent with the findings reported in the existing literature. Although the current firmware provides essential features, its runtime is significantly shorter than that of the standalone LoRa counterpart.
8. Conclusions
This study presents a groundbreaking hybrid platform that integrates LoRa and 5G technologies specifically designed for predictive maintenance in industrial ventilation systems. By harnessing LoRa’s low-power, long-range communication alongside the high-speed, low-latency capabilities of 5G, this platform ensures efficient data collection, transmission, and real-time analysis.
The system features a diverse range of sensors for monitoring vibration, temperature, and environmental conditions combined with powerful on-board AI for edge analytics. This robust design guarantees exceptional adaptability to the demands of industrial applications.
The hybrid platform outperforms the standalone LoRa and 5G solutions by offering superior operational flexibility. It prioritizes energy efficiency through intelligent power management, significantly extending operational runtime while facilitating real-time monitoring and periodic data bursts. Additionally, its capacity for dynamic adaptation through OTA updates and remote reconfiguration positions it as a highly scalable solution that meets the evolving requirements of Industry 4.0.
In summary, this hybrid approach decisively balances energy efficiency, communication reliability, and performance, establishing itself as a formidable and effective solution for predictive maintenance in industrial environments.
Conceptualization, P.M.; software, P.M.; validation, P.M.; formal analysis, P.M.; data curation, P.M.; writing—original draft preparation, P.M.; writing—review and editing, P.M.; visualization, P.M.; supervision, R.H.S. All authors have read and agreed to the published version of the manuscript.
Not applicable.
More inforamation regarding the project is available in the website:
The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.
The following abbreviations are used in this manuscript:
| LoRa | Long-range |
| 5G | Fifth-generation cellular technology |
| IoT | Internet of Things |
| IIoT | Industrial Internet of Things |
| NB-IoT | Narrow-Band Internet of Things |
| AI | Artificial Intelligence |
| MSP | Multi-sensor platform |
| BMS | Battery Management System |
| USB | Universal serial bus |
| IC | Integrated Circuit |
Footnotes
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Enlarge this image.Figure 1. Hybrid LoRa and 5G predictive maintenance system of a ventilation unit.
Energy consumption of commercial modules.
| Hardware | Current Draw @5 V | Energy Consumption per Hour (mAh) |
|---|---|---|
| XDK110 | 30, 200 | 41.33 |
| WISE-2410 | 10, 80 | 14.67 |
| Beanair BND | 20, 100 | 25.33 |
| NBSN95 | 20, 180 | 30.67 |
| 5G MSP | 200, 450 | 216 |
| Hybrid Prototype | - | 32.44 |
Wireless sensor comparison.
| Feature | XDK110 | WISE 2410 | 5G-MSP | NBSN95 | Beanair BND 2.4 GHz | Hybrid Prototype |
|---|---|---|---|---|---|---|
| Mode of Comm. | BLE, Wifi | LoRaWAN | 5G | NB-IoT | Prop. 2.4 GHz | LoRaWAN, 5G |
| Power Consumption | moderate | low | high | moderate | moderate | moderate |
| Battery Life | low | high | very low | low | medium | moderate |
| Range | short (limited) | medium | long | long | short | long |
| Data TX Periodicity | medium | low | high | low | low | medium–high |
| Edge Processing | limited | limited | moderate | limited | limited | advanced |
| Cost Efficiency | moderate | high | low | high | moderate | moderate |
References
1. Bekmurzaeva, R.; Kovalev, G.S. Industry 4.0: The Fourth Industrial Revolution. SHS Web Conf.; 2023; 172, 02011. [DOI: https://dx.doi.org/10.1051/shsconf/202317202011]
2. Xu, L.D.; Xu, E.L.; Li, L. Industry 4.0: State of the art and future trends. Int. J. Prod. Res.; 2018; 56, pp. 2941-2962. [DOI: https://dx.doi.org/10.1080/00207543.2018.1444806]
3. Carvalho, T.P.; Soares, F.A.A.M.N.; Vita, R.; Francisco, R.P.; Basto, J.P.; Alcalá, S.G. A systematic literature review of machine learning methods applied to predictive maintenance. Comput. Ind. Eng.; 2019; 137, 106024. [DOI: https://dx.doi.org/10.1016/j.cie.2019.106024]
4. Herrera-Granados, G.; Misaka, T.; Herwan, J.; Komoto, H.; Furukawa, Y. An experimental study of multi-sensor tool wear monitoring and its application to predictive maintenance. Int. J. Adv. Manuf. Technol.; 2024; 133, pp. 3415-3433. [DOI: https://dx.doi.org/10.1007/s00170-024-13959-0]
5. Xu, D.; Zhou, A.; Zhang, X.; Wang, G.; Liu, X.; An, C.; Shi, Y.; Liu, L.; Ma, H. Understanding Operational 5G: A First Measurement Study on Its Coverage, Performance and Energy Consumption. Proceedings of the Annual Conference of the ACM Special Interest Group on Data Communication on the Applications, Technologies, Architectures, and Protocols for Computer Communication; Online, 10–14 August 2020.
6. Arduino. Portenta H7. Available online: https://www.arduino.cc/pro/hardware/product/portenta-h7 (accessed on 25 January 2025).
7. KEMET Electronics Corporation. VS Vibration Sensors. Datasheet. Available online: https://content.kemet.com/datasheets/KEM_SE0209_VS.pdf (accessed on 25 January 2025).
8. Omega Engineering. (n.d.). PT1000 RTD Temperature Sensor. Available online: https://www.omega.com (accessed on 1 March 2025).
9. Mohanram, P.; Passarella, A.; Zattoni, E.; Padovani, R.; König, N.; Schmitt, R.H. 5G-Based Multi-Sensor Platform for Monitoring of Workpieces and Machines: Prototype Hardware Design and Firmware. Electronics; 2022; 11, 1619. [DOI: https://dx.doi.org/10.3390/electronics11101619]
10. Romanssini, M.; de Aguirre, P.C.C.; Compassi-Severo, L.; Girardi, A.G. A Review on Vibration Monitoring Techniques for Predictive Maintenance of Rotating Machinery. Eng; 2023; 4, pp. 1797-1817. [DOI: https://dx.doi.org/10.3390/eng4030102]
11. Lentz, J.; Hill, S.; Schott, B.; Bal, M.; Abrishambaf, R. Industrial Monitoring and Troubleshooting Based on LoRa Communication Technology. Proceedings of the IECON 2018—44th Annual Conference of the IEEE Industrial Electronics Society; Washington, DC, USA, 21–23 October 2018; pp. 3852-3857. [DOI: https://dx.doi.org/10.1109/IECON.2018.8591510]
12. Chen, Y.; Shi, G.; Al-Quraan, M.; Sambo, Y.; Onireti, O.; Imran, M. LoRa Mesh-5G Integrated Network for Trackside Smart Weather Monitoring. IEEE Trans. Veh. Technol.; 2024; 73, pp. 8903-8914. [DOI: https://dx.doi.org/10.1109/TVT.2024.3361160]
13. Busacca, F.; Mangione, S.; Neglia, G.; Tinnirello, I.; Palazzo, S.; Restuccia, F. FedLoRa: IoT Spectrum Sensing Through Fast and Energy-Efficient Federated Learning in LoRa Networks. Proceedings of the 2024 IEEE 21st International Conference on Mobile Ad-Hoc and Smart Systems (MASS); Seoul, Republic of Korea, 25 September 2024; pp. 295-303. [DOI: https://dx.doi.org/10.1109/MASS62177.2024.00047]
14. Zonta, T.; da Costa, C.A.; Righi, R.d.R.; de Lima, M.J.; Trindade, E.S.d.; Li, G.P. Predictive maintenance in the Industry 4.0: A systematic literature review. Comput. Ind. Eng.; 2020; 150, 106889. [DOI: https://dx.doi.org/10.1016/j.cie.2020.106889]
15. de Jonge, B.; Teunter, R.; Tinga, T. The influence of practical factors on the benefits of condition-based maintenance over time-based maintenance. Reliab. Eng. Syst. Saf.; 2017; 158, pp. 21-30. [DOI: https://dx.doi.org/10.1016/j.ress.2016.10.002]
16. Wan Hamzah, W.A.; Sakinah, S.; Alias, J.; Zawawi, N. Automotive Paint Production Line Blistering Improvement and Approach. J. Appl. Eng. Des. Simul.; 2022; 2, pp. 19-28. [DOI: https://dx.doi.org/10.24191/jaeds.v2i1.48]
17. Pech, M.; Vrchota, J.; Bednář, J. Predictive Maintenance and Intelligent Sensors in Smart Factory: Review. Sensors; 2021; 21, 1470. [DOI: https://dx.doi.org/10.3390/s21041470] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/33672479]
18. Quectel. 5G RM520N Series. Quectel. Available online: https://www.quectel.com/product/5g-rm520n-series/ (accessed on 26 January 2025).
19. Kemet Corporation. VS-JV10A Vibration Sensor. Available online: https://www.kemet.com/us/en/solutions/technologies/sensors/vibration-sensors/vs-jv10a.html (accessed on 28 February 2025).
20. Muñiz, R.; Nuño, F.; Díaz, J.; González, M.; Prieto, M.J.; Menéndez, Ó. Real-time monitoring solution with vibration analysis for industry 4.0 ventilation systems. J. Supercomput.; 2023; 79, pp. 6203-6227. [DOI: https://dx.doi.org/10.1007/s11227-022-04897-3]
21. Sefati, S.S.; Halunga, S. Ultra-reliability and low-latency communications on the internet of things based on 5G network: Literature review, classification, and future research view. Trans. Emerg. Telecommun. Technol.; 2023; 34, E4770. [DOI: https://dx.doi.org/10.1002/ett.4770]
22. Cattani, M.; Boano, C.A.; Römer, K. An experimental evaluation of the reliability of lora long-range low-power wireless communication. J. Sens. Actuator Netw.; 2017; 6, 7. [DOI: https://dx.doi.org/10.3390/jsan6020007]
23. Fusco, P.; Rimoli, G.P.; Ficco, M. TinyIDS—An IoT Intrusion Detection System by Tiny Machine Learning. Computational Science and Its Applications—ICCSA 2024 Workshops. ICCSA 2024; Lecture Notes in Computer Science Gervasi, O.; Murgante, B.; Garau, C.; Taniar, D.; C. Rocha, A.M.A.; Faginas Lago, M.N. Springer: Cham, Switzerland, 2024; Volume 14816, [DOI: https://dx.doi.org/10.1007/978-3-031-65223-3_5]
24. Lin, T.-H.; Chang, C.T.; Putranto, A. Tiny machine learning empowers climbing inspection robots for real-time multiobject bolt-defect detection. Eng. Appl. Artif. Intell.; 2024; 133, 108618. [DOI: https://dx.doi.org/10.1016/j.engappai.2024.108618]
25. Advantech. WISE-2410: LoRaWAN Smart Vibration Sensor. Available online: https://www.advantech.com/en/products/b7e2306f-d561-4ca9-b0e3-33f7057e185f/wise-2410/mod_25018dc7-355c-40b4-bf9b-c93f6c73f1a0 (accessed on 24 February 2025).
26. Dragino. NBSN95—NB-IoT Outdoor Sensor Node. Available online: https://www.dragino.com/products/nb-iot/item/163-nbsn95.html (accessed on 24 February 2025).
27. Bosch Connected Devices and Solutions. LEGIC XDK110 IIoT Sensor Module Flyer. PDF File. Available online: https://www.xdk.io/fileadmin/user_upload/LEGIC_XDK110_IIoT_Sensor_Module_Flyer_EN.pdf (accessed on 28 February 2025).
28. Beanair. 2.4 GHz Wireless IoT Accelerometer Sensors. Available online: https://www.beanair.com/2.4ghz-wireless-iot-accelerometer-sensors.html (accessed on 24 February 2025).
29. Raspberry Pi Foundation. RP2040 Microcontroller. Available online: https://www.raspberrypi.com/products/rp2040 (accessed on 1 March 2025).
30. Cebecioglu, B.B.; Mo, Y.K.; Dinh-Van, S.; Evans, A.; Mi, D.; Higgins, M.D.; Abozariba, R.; Aneiba, A. Experimental Analysis of 5G NR for Indoor Industrial Environments. IEEE Access; 2024; 12, pp. 89310-89321. [DOI: https://dx.doi.org/10.1109/ACCESS.2024.3419011]
31. Llap, A.; Nunez, M. Evaluating Energy Consumption and Maximum Communication Distance for SX1280 LoRa Transceiver at 2.4 GHz towards Adaptive Networks. Proceedings of the 2023 19th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob); Montreal, QC, Canada, 21 June 2023; pp. 393-398. [DOI: https://dx.doi.org/10.1109/WiMob58348.2023.10187857]
© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.