Sensor System Design and Deployment for Environmental and Visitor Monitoring in Cultural Heritage Sites

1. Introduction

The preservation of cultural heritage sites, ranging from archaeological ruins to historic buildings and underground structures, requires accurate and continuous monitoring of environmental and structural conditions. In recent decades, wireless monitoring systems have emerged as a key enabling technology for conservation and preventive maintenance. These systems, typically based on Wireless Sensor Networks (WSNs) and Internet of Things (IoT) architectures, allow long-term data acquisition without the need for intrusive cabling or permanent alterations to historically sensitive structures.

WSNs consist of spatially distributed sensor nodes equipped with sensing elements, microcontrollers, and wireless communication modules. These nodes autonomously collect and transmit data on physical variables such as temperature, relative humidity, vibrations, and material deformation. Their flexibility, low power consumption, and minimal visual impact make them particularly suitable for heritage environments, where non-invasive monitoring is essential [1,2,3].

Traditionally, monitoring in cultural heritage relied on manual inspections and wired instrumentation systems. While effective in controlled environments, these approaches are labor-intensive, costly to maintain, and limited in spatial scalability. In contrast, wireless systems enable rapid deployment across complex geometries and hard-to-access areas, significantly reducing installation costs and physical impact on heritage assets.

Early applications of WSNs in cultural heritage primarily focused on architectural and structural monitoring. Mecocci and Abrardo implemented a WSN on the “Rognosa” tower in San Gimignano (Italy), integrating multiple sensors to monitor environmental and structural parameters such as temperature, humidity, light intensity, masonry fissures, and rainfall [2]. Similarly, Ceriotti et al. deployed a WSN in the Torre Aquila in Trento, successfully collecting vibration and deformation data over long periods [3].

Subsequent research expanded system architectures and applications. Aparicio et al. analyzed routing topologies (tree, mesh) for monitoring historic buildings, highlighting communication challenges in obstructed environments [1]. Rodríguez-Sánchez et al. proposed modular WSN systems with flexible gateways for geographically dispersed heritage assets [4]. Klein et al. demonstrated large-scale deployments in museums with over 200 sensor nodes for microclimate and visitor impact monitoring [5].

More recent work shows the transition toward IoT-enabled and intelligent heritage monitoring systems. Lombardo et al. demonstrated integrated wireless sensing for indoor and outdoor atmospheric monitoring in cultural heritage contexts [6], while Palomeque-Gonzalez proposed modular IoT systems combining low-cost sensors and cloud connectivity for scalable deployment [7].

Recent advances (2023–2026) further extend these systems through Artificial Intelligence AI, digital twins, and predictive analytics. IoT and AI integration allows not only real-time monitoring but also early detection of deterioration patterns and structural risks [8]. Casillo et al. highlight IoT-based frameworks combined with digital twin models that simulate the physical behavior of historical buildings, enabling predictive maintenance and scenario analysis [9].

Low-cost and scalable IoT architectures are also gaining importance, particularly for widespread adoption in resource-constrained heritage sites [10]. In parallel, machine learning approaches are being applied to specific degradation processes, such as corrosion risk estimation in stone and metal heritage materials using sensor-driven environmental data [11]. These models allow forecasting of material decay based on humidity, temperature, and pollutant exposure.

At a broader level, recent research proposes unified frameworks that integrate physics-based modeling, AI, and IoT sensing into hybrid systems for cultural heritage conservation. Valentino et al. introduce such an integrated architecture, combining physical simulation with data-driven AI models to improve prediction accuracy and conservation decision-making [12].

Additionally, bibliometric and systematic studies confirm the rapid expansion of this interdisciplinary field. Shehata et al. provide a comprehensive analysis of smart heritage preservation research, identifying IoT, AI, and digital monitoring systems as the dominant emerging paradigms [13]. Their findings highlight a clear shift toward intelligent, interconnected, and data-centric conservation ecosystems.

In summary, wireless monitoring systems for cultural heritage have evolved from early WSN prototypes into advanced IoT- and AI-driven platforms. These systems now support continuous, multi-parameter, and predictive monitoring of environmental, structural, and human-related factors. This evolution enables a transition from reactive conservation to proactive and intelligent heritage preservation strategies.

The European project ARGUS (Non-destructive, scalable, smart monitoring of remote cultural treasures) [14] builds on this technological trajectory by advancing the state of the art in wireless, autonomous monitoring specifically tailored to the needs of cultural heritage conservation.

In this paper the design, development, and validation of a set of low power, low-cost sensor systems intended for environmental monitoring and visitor tracking in the cultural heritage sites of ARGUS Project is shown. The preliminary results, from a preventive monitoring strategy applied to the underground wine cellars of Baltanás (Palencia, Spain), has been already published [15].

2. Background

ARGUS is a multidisciplinary Horizon Europe research and innovation project focused on advancing technologies for the preventive conservation and smart monitoring of built heritage assets across Europe [14].

ARGUS addresses the critical challenge of effective long-term remote monitoring of cultural heritage sites, especially those that are remote, difficult to access, or vulnerable to environmental, climatic, and human-induced threats. The project aims to:

• Develop a novel digital twin model of built heritage, capable of integrating multi-scale and multimodal data.
• Establish an advanced digitalization strategy to support the digital twin.
• Create portable measurement systems that use miniaturized sensors for non-destructive physical and chemical monitoring—including both ground and aerial components.
• Implement AI-enabled methods for modelling and identifying threat factors and impacts.
• Use AI for multimodal data fusion, combining remote sensing (e.g., climate, weather, pollution), disaster and institutional statistics, and on-site measurements.
• Deliver trustworthy AI decision support methods to enable preventive conservation strategies and informed policy-making.

In sum, ARGUS integrates cutting-edge sensing technologies, AI, digital twins, and preventive conservation frameworks to provide real-time monitoring, long-term analysis, and decision support for heritage managers and stakeholders.

The project consortium is coordinated by the Athena Research Centre and includes 13 partners from Greece, Cyprus, Germany, Italy, Spain, and Switzerland, spanning research organizations, universities and SMEs.

ARGUS is being tested and validated through diverse pilot sites across Europe, each representing different types of heritage and conservation challenges. These include:

• Island of Delos (Greece): A UNESCO World Heritage archaeological site, known for its monumental ruins and complex stratigraphy. The site presents challenges related to weathering, environmental exposure and large-area monitoring.
• Schenkenberg Castle (Switzerland): Medieval fortifications that require structural and environmental monitoring to assess decay processes and stability risks over time.
• Baltanás Underground Wine Cellars (Spain): Located in the province of Palencia, this extensive network of underground cellars is culturally significant and exhibits complex microclimatic conditions that affect preservation. ARGUS sensors and models help detect humidity gradients, material degradation and environmental threats in subterranean heritage.
• Monti Lucretili (Italy): Historic mountain area with heritage sites challenged by climatic variability, land movement, and accessibility limitations.
• Abbey of Sant’Antonio di Ranverso (Italy): A historic monastery where environmental and structural monitoring helps understand material behavior, moisture ingress, and conservation needs in historic ecclesiastical architecture.

Each pilot site contributes unique geographical, climatic and cultural contexts—testing ARGUS’s capacity to operate in remote and varied scenarios.

3. Materials and Methods

In this section the monitoring systems developed by the authors are introduced.

Most commercial monitoring systems used in structural and environmental monitoring of cultural heritage are generally based on proprietary architectures that limit interoperability and the integration of application-specific sensors, which is a critical requirement in heterogeneous heritage environments. Recent reviews highlight that traditional structural health monitoring (SHM) solutions are often fragmented and rely on closed or semi-closed systems, which reduces their adaptability and long-term sustainability in cultural heritage applications [16]. In addition, these systems are frequently associated with high deployment and maintenance costs, as well as scalability limitations when applied to large or complex heritage sites [8]. From an operational perspective, conventional monitoring infrastructures may also require significant energy and maintenance resources, which restricts their suitability for continuous long-term deployments in fragile or remote environments [17]. Consequently, recent research has increasingly focused on WSNs and IoT-based architectures that enable low-cost, energy-efficient, and modular monitoring systems, improving flexibility and enabling the integration of heterogeneous sensing technologies tailored to cultural heritage conservation needs [18].

To overcome these limitations, the authors designed custom LoRaWAN-based systems using commercially available electronics and sensors. These systems are tailored to the particular needs of the study, allowing for flexible sensor integration and low-power operation.

Data collected by these systems are transmitted wirelessly to a cloud platform provided by one of the project partners, enabling real-time access, monitoring, and download. This approach not only facilitates continuous data acquisition but also allows for remote management and analysis, improving the efficiency and reliability of environmental monitoring or experimental studies. By combining commercially available components with open communication protocols, the designed systems provide a cost-effective, adaptable, and scalable solution compared to conventional closed commercial platforms.

Several static and dynamic monitoring systems have been designed, developed and tested for the ARGUS project. Components were selected based on five criteria: (i) small physical footprint, (ii) low power consumption, (iii) low manufacturing cost, (iv) LoRaWAN compatibility, and (v) type specific performance specifications. In Table 1 the list of static and dynamic systems developed in the project is presented. These systems were based on commercial platforms such as Arduino and they use LoRaWAN communication protocols.

3.1. Static Systems

Static monitoring systems in cultural heritage environments refer to fixed or permanently installed sensing infrastructures designed to continuously collect environmental and usage-related data without mobility. These systems typically focus on long-term, in situ measurements of key parameters relevant to conservation and structural assessment. As summarized in the table, static deployments include movement detection and airflow monitoring, visitor and people counting systems (including both basic and enhanced visitor monitoring approaches), as well as environmental sensing such as internal air temperature and relative humidity measurements at different depths within the structure. They also encompass more specialized measurements such as volumetric water content (VWC) combined with internal thermal-hygrometric conditions, wind direction and velocity monitoring, and pollutant concentration assessment. In contrast to dynamic systems, which rely on mobile platforms such as unmanned aerial vehicles (UAVs) or quadruped robots for gas sensing in specific scenarios, static systems provide continuous, distributed, and long-term observations that are essential for understanding gradual environmental changes and their impact on cultural heritage assets.

3.1.1. Indoor Movement and Airflow Monitoring System

To measure both human entry and ventilation conditions in the case study, an integrated monitoring system was developed. The system employs a PIR HC-SR501 sensor (Utmel Electronic Ltd, Hong Kong, China) for motion detection and an Omron (model DF6V, Omron Corporation, Kyoto, Japan) Micro-Electro-Mechanical Systems (MEMS) airflow sensor, integrated with an Arduino WAN 1310 board (Arduino S.r.l., Ivrea, Italy). The system was deployed in two cellars in Baltanás. A custom structure coupled to a commercial enclosure was fabricated using a 3D printer to house the airflow sensor. The sensor was positioned near the main entrances of the cellars to monitor both entry events and airflow, as the primary concern in this case study was the high humidity levels.

Figure 1. A system for people detection and air flow measurement installed in one of the wineries in Baltanás.

Figure 1. A system for people detection and air flow measurement installed in one of the wineries in Baltanás.

3.1.2. Outdoor Visitor and Vehicle Monitoring System

The main objective of the visitor monitoring subsystem is to accurately count the number of people or objects passing through specific channels (such as entrances, exits, or corridors) and to precisely determine their direction of movement. Its hardware composition is relatively simple, consisting primarily of a central controller and an infrared sensor unit.

Most of the systems for that purpose are based on cameras and their consumption is very high or they are prepared for indoors.

Two different versions have been developed. From a very simple proof-of-concept prototype (V1), its limitations were analyzed, and it subsequently evolved into an improved version (V2) with enhanced functionality and a more comprehensive decision-making logic.

Prototype V1 consists of an Arduino MKR WAN 1310, an infrared photoelectric sensor (SICK VL180-2N41131, SICK AG, Waldkirch, Germany) and a reflector, and a LoRa antenna. The externally powered sensor detects the passage of objects by identifying interruptions in a reflected infrared beam, and the system attempts to classify pedestrians and vehicles based on the duration of the beam obstruction. Its operating principle is as follows: the emitter continuously transmits an infrared light beam, and the reflector installed on the opposite side accurately reflects the beam back to the receiver, which is integrated within the same sensor module. When an opaque object (such as a pedestrian or a vehicle) passes through and blocks the beam, the receiver no longer detects the reflected light, and the digital output signal changes state.

The software performs a preliminary classification of passing objects by measuring the total duration of the infrared beam interruption. For example, based on experimental data, a threshold is defined whereby events with a blockage duration of less than 1.5 seconds are classified as “pedestrians,” those between 1.5 and 6 seconds as “vehicles,” and those exceeding 10 seconds as abnormal occupancies, which trigger an alarm.

However, this first version presents several limitations. It cannot determine the direction of movement, as it only registers that an object has passed, and its accuracy is limited because the classification method—relying solely on the time the beam is blocked—is unreliable in real-world scenarios.

These shortcomings motivated the development of a more advanced V2 prototype. To evaluate the performance of V1, the system was tested for approximately one week in a 1.5-meter-wide corridor at the Institute of Physical and Information Technologies (ITEFI) of CSIC, with the sensor and reflector mounted on opposite sides at a height of 1 meter. Figure 2 shows the sensor and its reflector installed for testing.

To overcome the key functional limitations of Version V1, the V2 prototype underwent a significant hardware upgrade, most notably through the addition of a second reflective photoelectric sensor. The two sensors are placed side by side along the walking direction, separated by a small, fixed distance (typically 10–20 cm). The main components of the system include an Arduino MKR WAN 1310, two reflective photoelectric sensor modules the same model as in the first version with their corresponding reflectors, and a LoRa antenna. Because the physical appearance of these components was already shown in the V1 description, it is not repeated here.

Although the hardware modification appears simple, it provides the logical foundation required for determining direction. The principle is similar to industrial quadrature encoders: when an object passes, it activates the two sensors in sequence, and by analyzing which sensor is triggered first, the firmware can accurately determine the direction of movement. If sensor A is blocked first and sensor B second, the system registers an entry; if sensor B is triggered before sensor A, it records an exit.

To fundamentally address the aforementioned issues, firmware version V2 restructures the core logic using a Finite State Machine (FSM) approach. The FSM abstracts the system into a finite number of states and clearly defines state transitions and the actions to be executed when a specific event is triggered. In embedded systems, the FSM methodology enables complex and intertwined logic to be divided into maintainable and predictable units. Moreover, it compels developers to systematically consider all possible states and transitions prior to implementation, thereby reducing logical vulnerabilities.

In addition, the firmware incorporates an anti-bounce delay of approximately 20 ms when processing sensor inputs. A state change is only confirmed if it remains stable for this duration, effectively reducing misinterpretations caused by mechanical vibrations.

The FSM design in firmware version V2 is structured as follows:

States:

• IDLE: Inactive state in which the system waits for any sensor to be triggered. This is the starting point of all detection sequences.
• PASSAGE_IN_PROGRESS: Passage state indicating that a detection sequence has started and that the system is tracking the object’s transit process.

Events:

• Sensor state change: The signal level of either infrared sensor (ir1_state or ir2_state) changes from LOW to HIGH (blocked) or from HIGH to LOW (unblocked), after passing vibration filtering.
• Timeout event: The passage duration exceeds the maximum threshold (PASSAGE_TIMEOUT), or the second sensor is not triggered within the specified time (SEQUENCE_INITIATION_TIMEOUT).

Actions:

• Start/stop the timer, record the initial direction (IN or OUT), and update internal flags (bothSensorsWereBlocked, secondSensorTriggered).
• When secondSensorTriggered is TRUE:

  -

  If bothSensorsWereBlocked is TRUE, increment the vehicle counter by +1;

  -

  Otherwise, increment the pedestrian counter by +1.

• Set the dataNeedsSending flag to trigger subsequent LoRaWAN data transmission.

To formally and unambiguously describe this core logic, Table 2 presents a simplified version of the FSM state transition rules.

Compared to a simple if–else logic structure, this FSM-based design provides a clearer architecture and more rigorous logical framework. It effectively prevents misclassification caused by signal fluctuations or complex traffic situations (such as direction changes during passage). This approach represents a method for achieving reliable direction determination.

This enhancement transforms the system from merely counting objects to offering directional counting, greatly increasing its practical value.

To investigate the relationship between vehicle traffic and potential vibration events inside the cellars, Version 1 of the system was installed outside two cellars in Baltanás. Since the direction of vehicle movement was not required for this analysis, the system was configured solely to detect passing vehicles and correlate their occurrence with vibration measurements collected within the cellars.

3.1.3. System to Measure Internal Ambient TH at Different Depths

A system to monitor internal ambient TH was designed by the authors. The sensors measure the conditions inside the material’s chamber using an Arduino-based platform. They are digital sensors with adaptable probe lengths. The TH sensors employed are commercial Honeywell HDC302x-Q1 (Honeywell, Charlotte, United States of America) devices, characterized by very low power consumption (<1 mA). The system operates on a 3.7 V, 2.6 mAh battery.

Figure 3. Sensor developed to measure internal TH.

Figure 3. Sensor developed to measure internal TH.

A first version of this prototype was tested in Schenkenberg Castle.

3.1.4. System to Measure VWC and Internal Ambient TH

A custom-built system was developed to measure internal wall temperature, relative humidity, and VWC. The system is based on an Arduino MKR WAN 1310 and integrates the previously developed TH sensor together with the commercial SEN0193 to determine VWC. The SEN0193 is an analog capacitive sensor characterized by very low power consumption. The system is powered by a 3.6 V, 2.6 Ah AA battery.

Figure 4. Custom-built system developed to measure internal wall temperature, relative humidity, and VWC.

Figure 4. Custom-built system developed to measure internal wall temperature, relative humidity, and VWC.

Two systems were deployed at Ranverso abbey to measure internal TH inside 2 cracks and one system in Schenkenberg Castle.

3.1.5. Wind Speed and Direction Measurement Prototypes

Two prototype systems for measuring wind speed and direction have been developed using MEMS airflow sensors. The first prototype (V1) is designed for low wind speeds up to 10 km/h, while the second prototype (V2) extends the measurement range to 54 km/h. Both systems share a similar architecture based on a low-power microcontroller and multiple airflow sensors arranged in different orientations to estimate wind direction.

V1: Low-speed prototype (0–10 km/h)

This system is based on an Arduino MKR WAN 1310 microcontroller and incorporates four Omron D6F-V03A1 airflow sensors. Each sensor is oriented toward one of the cardinal directions (North, South, East, and West), enabling the estimation of both wind speed magnitude and direction. These MEMS airflow sensors are capable of measuring air velocities up to 10 km/h, making the system suitable for environments with low air movement.

Figure 5. Prototype V1 of 4-direction anemometer, velocities up to 10 km/h.

Figure 5. Prototype V1 of 4-direction anemometer, velocities up to 10 km/h.

V2: Extended-range prototype (0–54 km/h)

The second prototype also uses an Arduino MKR WAN 1310 microcontroller but employs four Renesas (FS3000-1015, Renesas Electronics, Tokyo, Japan) airflow sensors. These sensors are based on a MEMS thermopile sensing element and provide digital output with 12-bit resolution, allowing wind speed measurements of up to 54 km/h. As in the first prototype, the four sensors are arranged in different orientations to estimate wind direction.

In addition to airflow sensing, the system incorporates an (SHT35, Sensirion AG, Stäfa, Switzerland) sensor to measure TH with high accuracy, providing environmental context for the wind measurements.

Both prototypes are designed for low-power operation, enabling long-term autonomous deployments powered by batteries. The Arduino MKR WAN 1310 provides integrated LoRaWAN communication, allowing the collected data—wind speed, wind direction, temperature, and humidity—to be transmitted wirelessly over long distances. This design makes the system suitable for environmental monitoring, field experiments, and scientific studies, particularly in remote locations where power availability is limited.

Figure 6. a) FS3000-1015 air velocity sensor extracted from sparkfun.com, b) electronic board with 4 air velocity sensors, c) electronic board with the Arduino and the battery, d) first encapsulated prototype V2 developed with a 3D printer.

Figure 6. a) FS3000-1015 air velocity sensor extracted from sparkfun.com, b) electronic board with 4 air velocity sensors, c) electronic board with the Arduino and the battery, d) first encapsulated prototype V2 developed with a 3D printer.

A new version of the system was developed, equipped with a solar panel to allow it to operate without external power.

Figure 7. New version of the wind and velocity measurement system V2.

Figure 7. New version of the wind and velocity measurement system V2.

Two systems V2 were deployed in two case studies: one in Delos and another at Ranverso Abbey.

3.1.6. System to Measure Pollution

The authors have been testing various low-cost, low-power sensors to measure different gases, including CO₂, CO, NO₂, NH₃, SO₂, particulate matter (PM), and volatile organic compounds (VOCs). The sensors selected for the prototype are shown in Table 3.

A visual representation of the system´s components is shown in Figure 8.

A custom enclosure was designed using computer-aided design (CAD) software. Manufacturing was carried out through additive manufacturing (3D printing). The three-dimensional design, developed with NX software, considered essential criteria of functionality, protection, and non-invasive deployment. The final version of the prototype is shown in Figure 9.

3.2. Dynamic Systems

Dynamic monitoring systems in cultural heritage refer to approaches where the sensing and data acquisition tasks are performed by mobile platforms rather than fixed installations. Unlike static systems, these solutions rely on movable devices such as UAVs, ground robots, or portable sensing units that can be deployed on demand to inspect specific areas of interest. This mobility enables flexible, high-resolution, and targeted monitoring of inaccessible or hazardous zones, such as upper facades, roofs, or structurally complex regions. Typical applications include gas and pollutant detection, visual inspection, and environmental sampling in localized areas where static sensor deployment is not feasible or sufficient. However, dynamic systems are generally used for short-term or episodic measurements due to constraints in battery life, operational complexity, and limited continuous coverage. As a result, they are often considered complementary to static monitoring infrastructures, providing additional spatial detail and on-demand inspection capabilities within integrated cultural heritage monitoring strategies.

Among the gas-monitoring systems developed in the project, several configurations have been designed and tested. All the systems described so far are intended for static monitoring, meaning they are installed in a fixed location. To monitor large areas, hard-to-reach locations, or sporadically relevant zones, two additional systems were created for dynamic monitoring.

3.2.1. UAV-Mounted Ultralight System

An ultralight system was designed to be placed on an UAV. The system was composed of a light microcontroller the Adafruit Feather M0 WiFi - ATSAMD21 board. This microcontroller was chosen for its light weight, only 6.1 grams. The lightweight system (under 130 g including the battery) designed by the authors is available in two different chimney configurations, vertical and horizontal, see Figure 11. Both configurations include the same set of gas sensors, (MiCS-6814, Amphenol SGX Sensortech, Corcelles-Cormondrèche, Switzerland) sensor for CO, NH₃ and NO₂ and Sensirion SCD41 (Sensirion AG, Stäfa, Switzerland) for CO₂, temperature, and humidity.

Figure 10. Ultralight hardware used to design the system to be placed on an UAV.

Figure 10. Ultralight hardware used to design the system to be placed on an UAV.

A measurement was performed every 10 seconds and it was send using WiFi to the cloud.

The sensors are housed inside the drone’s mounting box, which incorporates a vertical chimney-like ventilation structure. In one configuration, the chimney faces forward; in the other, it faces upward. Figure 11 shows both systems. This system was tested on the Pilot Site of Schenkenberg in Switzerland. The first results are presented in [19].

Figure 11. UAV with two chimney configurations.

Figure 11. UAV with two chimney configurations.

3.2.2. Quadruped-Robot System

The proposed system uses the Unitree GO2 quadruped robot as its mobile platform, on which a low-power, modular multisensor node has been integrated. The system incorporates environmental sensors, including TH, particulate matter, and different gases. A movable component that rotates the sensor panel toward the wind direction has been designed. This rotation is controlled by a system based on an anemometer, which detects the wind direction and positions the panel accordingly. Connectivity with the central data platform has also been evaluated using two communication technologies: the long-range, low-power LoRaWAN protocol and WiFi communication, enabling remote data transmission, automatic synchronization, and real-time integration with the digital twin platform.

At the hardware level, component selection and integration have been carried out according to criteria of low energy consumption, lightweight design, and modularity. On the software side, embedded programming has been developed to enable synchronized acquisition from multiple sensors, edge preprocessing, and data packaging and transmission management. In addition, preliminary tests have been conducted in real-world environments to validate the feasibility and stability of the system in representative cultural heritage scenarios.

Figure 12 shows the gas-sensing system (SO₂, CO₂, CO, NH₃, NO₂, PM_2.5, temperature, and humidity) with direction detection mounted on the Unitree GO2 robot.

The mechanical design of the system is composed of 3 modules: base, central module and cover, see Figure 13. In the base is where the rotation system is placed, in the central module the monitor system and in the cover is where the anemometer is placed. When a gas measurement has to be performed the system measures the wind direction and it orientates the central module to this direction to measure the gas pollution.

For the rotation of the system and Arduino WAN 1310 has been used with a magnetic limit switch is incorporated into the Base, while in the Central Module the opposite limit is implemented using a screw with a magnet, see Figure 14 and Figure 15. This limit switch is essential in any mechanical system, as without it the angular position of a servomotor cannot be properly determined. The detected position is defined as the mechanical 0°, which allows the system to eliminate accumulated drift and ensure consistent and reliable positioning over time. Once the reference point is calibrated, the system performs the rotation toward the target angle based on the wind direction (θ) using the minimum path strategy.

In the central module is where the monitoring system is located and it is composed of an Arduino microcontroller with several gases sensors. This module is in charge of sending the gas measurements using WiFi or LoRaWAN to the cloud.

Figure 16. Monitoring system with the gas sensors used in this application.

Figure 16. Monitoring system with the gas sensors used in this application.

A first version of this prototype was tested in Baltanás. The preliminary results obtained with this system have been presented in [16].

4. Results

In this section, the preliminary results obtained with the different tailor-made systems are presented, demonstrating their capability to monitor key parameters and provide valuable information on the condition of cultural heritage sites, as well as to support decision-making for their preservation.

4.1. Indoor Movement and Airflow Monitoring System Preliminary Results

In Figure 17 the map of the cellar indicating where the system to measure PIR and airflow has been placed is marked. It should be noticed that there is an open window at the entrance and the door also have some holes to have a good ventilation, see Figure 18.

The preliminary results in one of the cellars are shown in Figure 19. In this figure the airflow measurement with the PIR sensor in the interior of the cellar close to the main door are compared with the exterior wind and direction velocity obtained from the Open-Meteo Historical Weather API, using the ERA5-Land reanalysis dataset provided by the European Centre for Medium-Range Weather Forecasts [21]. It can be seen that velocity measure with the airflow sensor at the interior follow the same trend as the wind velocity measured in the exterior in the ERA dataset. The values show in the figure are normalized but it should be noticed that the air flow values at the interior are always greater or equal to 0 m/s meaning that the ventilation is coming from the window or the main door. It can also be seen that when the wind direction is close to the north the velocity at the interior does not increase very much due to the orientation of the cellar. It was also noticed that when somebody enters to the cellar and the wind velocity outside is high a sudden increase can be noticed also in the interior but the main ventilation is occurring for the window close to the entrance.

In Figure 20, the three points in the cellar that have a direct connection to the outdoor environment are shown. The direction of airflow within the cellar is also inferred from the measurements performed. To validate this hypothesis, two additional airflow sensors should be installed at the other outdoor connection points.

In the other cellar that there is not that window the airflow sensor is measuring values lower than 0 m/s and this is because the ventilation goes in the other way around. The air is going from the interior to the exterior thought the main door.

Figure 21. Picture of the main entrance of the second cellar where the door can be seen.

Figure 21. Picture of the main entrance of the second cellar where the door can be seen.

Figure 22. Air flow and PIR detections measured inside the cellar and wind speed and wind direction measured outside obtained from the Open-Meteo Historical Weather API, using the ERA5-Land reanalysis dataset provided by the European Centre for Medium-Range Weather Forecasts [21].

Figure 22. Air flow and PIR detections measured inside the cellar and wind speed and wind direction measured outside obtained from the Open-Meteo Historical Weather API, using the ERA5-Land reanalysis dataset provided by the European Centre for Medium-Range Weather Forecasts [21].

To corroborate these results, four airflow sensors were installed at each connection to the exterior. The preliminary results were shown in [15]. The results show that air flows outward through two of the openings and inward through the other two, while at one additional opening the airflow direction varies, sometimes entering and sometimes exiting, see Figure 23.

4.2. Outdoor Visitor and Vehicle Monitoring System Preliminary Results

Before deploying these systems in the Baltanás cellars, they were tested at ITEFI. Figure 24 shows part of the data stream transmitted in real time via the LoRaWAN network to the ThingSpeak (MathWorks, Natick, United States of America) cloud platform during the deployment of the V1 visitor monitoring node. Since the test took place in an indoor corridor with frequent pedestrian traffic, all detected objects were real pedestrians rather than vehicles. The figure presents three data curves. Field 1 (person) represents the cumulative number of detected pedestrian crossings. The curve displays a clear upward trend during daytime working hours—such as between 9:00 and 11:00 and between 15:00 and 18:00, which corresponds to the busiest periods of daily activity at the institute (ITEFI). This behavior demonstrates the strong detection sensitivity and operational stability of the system in real-world use. Field 2 (vehicle) is intended to indicate the number of vehicle crossings. However, because no actual vehicle activity occurred in the test area, all recorded counts must be considered false positives. These are likely caused by pedestrians walking slowly through the detection zone or briefly lingering within it—for example, cleaning staff with carts—exceeding the 1.5-second threshold for vehicle recognition and consequently being misclassified as vehicles. Field 3 (alarm) represents alarm events triggered by abnormally long beam obstructions lasting more than 10 seconds. This curve remains generally stable, with only small peaks during specific periods. On-site observation confirms that these peaks correspond to cases where people stopped in front of the sensor to talk, temporarily blocking the infrared beam.

All three curves show a sudden drop or discontinuity at roughly the same point each day (around 13:00). This effect is not a system anomaly but rather the result of the programmed reset mechanism built into the firmware. To ensure long-term operational stability in unattended environments, the V1 firmware triggers an automatic reset every 24 hours, freeing system resources and refreshing communication states. This prevents issues associated with prolonged operation, such as memory leaks or communication module freezes. Additionally, if consecutive LoRaWAN uplink transmission failures exceed the preset threshold of five attempts, the system also activates the reset mechanism, initiating a forced self-recovery process. Each reset clears the counter values stored in the microcontroller’s memory, appearing as a step-like drop in the graph. This behavior is fully expected and intentionally implemented to enhance long-term reliability in unattended deployments.

The V2 monitoring system was also tested at the facilities of ITEFI, as shown in Figure 25. In the center of the figure, the access road to the parking area can be observed. On the left, a wooden panel holding the reflectors is installed, while on the right, two reflective photoelectric sensors are mounted on an aluminum support structure.

Figure 26 shows the data collected during a week for system V2. Field 1 corresponds to people entering, field 2 to people exiting, field 3 to cars entering, field 4 to cars exiting and field 5 to the alarm.

The information from Figure 26 can be further analyzed in detail in Table 4.

During the testing period, the system automatically recorded the daily number of vehicles and people entering and leaving the facility. Manual checks of the parking lot were also performed periodically to assess the accuracy of the system. Data for vehicle movements over one week were collected, showing daily entries, exits, and net flow.

The monitoring data clearly reveal cyclical patterns in parking occupancy frequency. Weekday traffic was significantly higher than weekend traffic, reflecting the real movement trends of staff. This demonstrates that the system can reliably capture dynamic changes in usage and maintain stable operation over time.

Counting accuracy was also high. Over the week, the system recorded 149 entries and 139 exits, with daily counts largely balanced. Minor discrepancies, such as 0 entries and 1 exit on Sunday, are typical in long-term parking management, for example when vehicles remain parked overnight.

The net flow error of 10 cases in one week was not random. It was primarily caused by cleaning staff moving large waste containers near the prototype, which blocked multiple infrared beams and were mistakenly classified as vehicles. Excluding these identifiable interference events could raise the vehicle counting accuracy above 96%.

In conclusion, a single hardware upgrade from V1 to V2, adding a second sensor, along with software refactoring that introduced a finite state machine, systematically addressed all critical issues from the previous version. The result is a robust, reliable visitor monitoring system that meets operational requirements.

As previously mentioned, since the direction of vehicle movement was not relevant for this analysis, two Version 1 systems were deployed in Baltanás. However, additional data are required to perform a comprehensive analysis and draw robust conclusions regarding the relationship between vehicle traffic and cellar vibrations.

4.3. TH and Soil Moisture Preliminary Results

Two systems (3D6617 and 26680F), each equipped with two sensors, one custom sensor developed by the authors to measure ambient temperature and humidity at two different depths, and one commercial sensor to measure VWC, were installed in two cracks in the Ranverso Abbey. The 26680F system presented a malfunction, as the custom-developed sensor consistently recorded a value of zero.

Figure 27. TH and soil moisture systems a) 3D6617 b) 26680F.

Figure 27. TH and soil moisture systems a) 3D6617 b) 26680F.

In Figure 28 the results obtained with these systems are shown. The temperature sensor Temp1 follows the same general trend as Temp2, including the significant drop observed in early January, indicating that both sensors are influenced by the same external conditions. However, Hum1 remains almost constant near 100%, which suggests possible sensor saturation. The sensor Hum2 behaves as a partially exposed microenvironment rather than a fully enclosed one. Hum2 remains consistently high (around 90–97%) but shows variability in response to temperature changes. This indicates interaction between ambient air and moisture stored in the wall. Overall, the crack is not completely isolated but experiences intermittent exchange with the exterior, resulting in a hybrid microclimate with elevated humidity and moderated temperature fluctuations.

The soil moisture sensors show a very stable behavior, with only gradual changes over time and a slight decrease during the cold period in January. This is consistent with moisture retained in the material, where variations occur slowly and are governed by long-term processes. Additionally, the soil moisture and humidity sensors of system 26680F respond to the temperature drop, whereas the soil moisture measurements from system 3D6617 do not clearly capture this variation. Instead, they begin to decrease around mid-January, indicating a delayed response that is likely due to differences in installation conditions.

Overall, both temperature sensors respond to the same external forcing, the crack environment maintains high but variable humidity, Hum1 is likely compromised, and the soil moisture reflects a stable subsurface moisture regime with limited short-term variability.

4.4. Anemometer Preliminary Results

The anemometer based on MEMS sensors designed by the authors was compared with a commercial calibrated station LoRaWAN (SenseCAP S2120, Seeed Studio, Shenzhen, China), see Figure 29. This weather station is an 8-in-1 measures air temperature, humidity, wind speed/direction, rainfall, light intensity, UV index, and barometric pressure. It features long-range connectivity and low power consumption (solar + batteries).

The results obtained for measuring the wind speed and direction are shown in Figure 30 and Figure 31 respectively.

The MEMS prototype demonstrates strong potential for wind monitoring, showing good agreement with the calibrated SenseCAP station in wind speed dynamics and general directional trends. Nevertheless, it exhibits higher noise levels and reduced stability in wind direction estimation. These results suggest that further signal processing (e.g., filtering or averaging) would significantly improve its performance and make it suitable for deployment as a low-cost alternative sensing solution.

4.5. Pollution System Preliminary Results

The system was tested for five days before being approved for deployment at Ranverso Abbey. Figure 32, field 1 corresponds to temperature, field 2 to humidity, field 3 to particulate matter 2.5 (PM_2.5) and field 4 to SO₂.

Figure 33 shows the sensor readings of CO, NO₂, CO₂ and volatile organic compounds (VOC). Field 5 corresponds to CO, field 6 to NO₂, field 7 to volatile organic compounds (VOCs) and field 8 to CO₂. CO and NO₂ are very noise and the values are quite constant. CO₂ measurements show the difference when there is a work day and when there is a free day, 19 and 20 July were Saturday and Sunday. Also, VOC is decreasing slowly during the weekend the windows were also close and start increasing on Monday.

In next figures the results obtained during almost 2 months in the Ranverso Abbey are shown. In shadow is marked when the abbey is open to the public from Wednesday to Sunday. In Figure 34 the temperature and humidity values in the interior of the abbey are shown. These values are compared with the meteorological data obtained from the Open-Meteo Historical Weather API, using the ERA5-Land reanalysis dataset provided by the European Centre for Medium-Range Weather Forecasts [21], Figure 35.

The comparison between indoor measurements and ERA5-Land reanalysis data highlights the strong hygrothermal buffering capacity of the abbey. While outdoor conditions exhibit pronounced diurnal variability and rapid responses to precipitation events, indoor temperature and relative humidity remain significantly more stable, with reduced amplitude and smoother temporal evolution. The absence of sharp humidity peaks indoors, even during rainfall events, suggests a delayed and attenuated moisture transfer through the building envelope. This behavior indicates the presence of high thermal inertia and moisture buffering capacity of the construction materials, which effectively decouple indoor microclimatic conditions from external atmospheric variability.

In Figure 36 the different gases measured at the interior of the abbey are shown. The SO₂ values are not shown since always were 0. The results show that the different pollutants analyzed are governed by clearly distinct emission sources and control mechanisms within the building.

First, CO₂ exhibits a pattern consistent with occupancy. Concentrations increase during periods when the building is open to the public and reach higher, sustained levels during weekend events. This confirms CO₂ as a robust proxy for human presence and ventilation effectiveness.

In contrast, volatile organic compounds (VOCs) display a highly episodic behavior, characterized by sharp, short-lived peaks. Although these peaks occur more frequently during occupied periods and events, they do not scale linearly with occupancy. This suggests that VOC levels are primarily driven by specific, short-term indoor sources, such as cleaning activities, use of products, materials, or particular event-related actions, rather than by occupancy alone.

Regarding particulate matter (PM₁, PM₂.₅, PM₄, and PM₁₀), all fractions show a smooth temporal evolution and are strongly correlated with each other, indicating a common and continuous source, see Figure 37. PM concentrations do not exhibit clear increases associated with opening periods or events and lack the abrupt peaks observed for VOCs. This pattern suggests that PM levels are mainly influenced by background conditions, likely of outdoor origin, and by ventilation and filtration processes, rather than by intermittent indoor activities. It should be noted that the abbey is very close to a highway that could affect the measurements.

Importantly, VOC peaks do not show a consistent pattern of co-occurrence with PM concentrations, indicating that the processes governing these pollutants are largely independent in the studied environment. While occasional simultaneous increases may occur, they are not systematic and do not support a direct causal relationship.

Overall, the findings indicate that indoor air quality in the building is controlled by three primary factors: (i) occupancy (reflected by CO₂), (ii) specific indoor activities (VOCs), and (iii) background environmental conditions and air handling systems (PM). These results highlight the need for pollutant-specific management strategies, rather than assuming coupled behavior among different air quality indicators.

VOC concentrations do not show a consistent relationship with rainfall events. However, a moderate dependence on environmental conditions is observed, with higher concentrations generally associated with increased temperature and, to a lesser extent, high humidity levels. This suggests that VOC variability is primarily driven by emission processes and atmospheric conditions rather than wet scavenging.

The results show a clear decrease in particulate matter concentrations during rainfall events, confirming the effect of wet scavenging. However, high PM episodes also occur independently of precipitation, indicating the influence of additional atmospheric or emission-related factors.

5. Conclusions

In this paper, the design, development, and validation of a set of low-power, low-cost sensor systems for environmental monitoring and visitor tracking in cultural heritage sites have been presented. Within the framework of the ARGUS project, several static and dynamic monitoring systems were designed, implemented, and tested under real deployment conditions.

The selection of system components was guided by five main criteria: (i) small physical footprint, (ii) low power consumption, (iii) low manufacturing cost, (iv) LoRaWAN compatibility, and (v) application-specific sensing performance. Table 1 summarizes the static and dynamic monitoring systems developed throughout the project. These systems were built using commercially available platforms such as Arduino and integrated with LoRaWAN communication protocols to ensure long-range, low-power wireless connectivity.

Unlike most commercial monitoring solutions, which are often proprietary, expensive, and limited in terms of sensor flexibility, the proposed systems are open and modular. Commercial alternatives frequently lack the specific sensing capabilities required for heritage applications and typically present high energy consumption, making them unsuitable for long-term autonomous deployments. To address these limitations, custom LoRaWAN-based solutions were developed using off-the-shelf electronic components and sensors, enabling flexible configuration, scalability, and energy-efficient operation tailored to the specific requirements of cultural heritage monitoring.

All collected data are transmitted wirelessly to a cloud-based platform, providing real-time access, visualization, and data retrieval. This architecture supports continuous monitoring, remote system management, and advanced data analysis, significantly improving the efficiency, accessibility, and reliability of environmental and experimental studies in heritage contexts.

The proposed systems have been successfully deployed across five pilot sites in four different European countries within the ARGUS project. Preliminary results demonstrate their effectiveness in capturing relevant environmental patterns and supporting the identification of factors influencing the conservation state of cultural heritage assets. Moreover, the systems enable near real-time risk detection, contributing to proactive preservation strategies.

In addition to static monitoring configurations, this work also addresses dynamic monitoring scenarios. Two additional mobile sensing systems were developed to support the observation of large areas, inaccessible locations, or temporally relevant zones. This combination of static and dynamic sensing approaches provides a more comprehensive and flexible monitoring strategy, enhancing the overall understanding of environmental and anthropogenic impacts on cultural heritage sites.

Overall, the results confirm that low-cost, LoRaWAN-based wireless sensing systems constitute a robust, scalable, and adaptable solution for long-term cultural heritage monitoring, bridging the gap between traditional conservation practices and modern IoT-enabled smart preservation frameworks.