AIoT Monitoring Technology for Optimal Fill Dam Installation and Operation

Abstract

Fill dam structures are generally considered safe, but frequent heavy rainfall in recent years due to climate change has increased their risk of collapse. Technologies for the monitoring and safety management of these structures have attracted considerable attention, and methods to utilize technologies such as artificial intelligence (AI) and Internet of Things (IoT) for maintenance have been investigated. However, the measurement and communication processes of sensors used in the IoT technology are prone to measurement errors. Moreover, the communication environment and frequency range used by unlicensed operators vary within a country. Technologies for accurate interpretation of measurement results and optimal communication are required to address these issues. In this study, a technology for dam safety management and communication environment optimization was developed using AIoT (AI+IoT), and its field applicability was verified.

1. Introduction

In recent years, heavy rainfall has become frequent because of the unusual weather due to climate change. This has led to the collapse of agricultural dams, which are generally considered safe structures. In Korea, the agricultural dam measurement system for disaster prevention was revised in 2015 to respond to climate change. However, the measuring instruments for this system depend only on the ground displacement and water level. Unfortunately, agricultural and fill dams have collapsed despite measurement management being performed during construction. An investigation of the causes of fill dam failure by the dam laboratory of UNSW University revealed that more than 91% of the 136 cases collapsed in the operational phase after the completion of construction. The primary causes of the failure were overflow (flooding), piping, slope activities, and earthquakes [1,2]. The series of accidents that occurred indicates the need for a professional real-time measurement system, as old agricultural dams are highly vulnerable to the effects of climate change, such as heavy rainfall and earthquakes. Moreover, the current disaster prevention measurement system cannot easily predict the risk of facility collapse in advance. Figure 1 shows the collapse of agricultural dams in Korea. Figure 1a is a reservoir in Yeongcheon-si, Korea in 2014, and Figure 1b is a dam in a reservoir in Boseong-gun, Korea in 2018. Both dams received a “Good” safety diagnosis, but the embankment and spillway collapsed under heavy rainfall. Moreover, 91 and 81 agricultural dams collapsed in 2003 and 2004, respectively, under heavy rainfall due to typhoons. Agricultural dams in Korea are designed based on a relatively low level of maximum precipitation with a frequency of 30 to 50 years, and most of them are in deteriorated condition. Their maintenance and measurement are based on visual inspection, ground displacement, groundwater level, and pore pressure. Major agricultural dam disasters due to heavy rainfall are caused by the activities of downstream slopes, which cannot be easily detected using the current measurement method; consequently, disasters cannot be predicted in advance and prevented. Systematic measurement technologies are required to respond to recent unusual weather, such as heavy rainfall. Disasters caused by the collapse of dams have also occurred in Vietnam (Figure 2). The failure of the Ha Tinh dam located in central Vietnam (KE2/20 REC dam) occurred in the middle of the closed conduit within less than a year of operation, resulting in erosion with a length of approximately 8.5 m and depth of 3.5 m [3]. Moreover, disasters caused by the collapse of dams have occurred repeatedly. These include the collapse of the Dakrong 3 dam in 2013, damage to the Son La dam drain in 2017, and the accident in the Rao Trang 3 hydroelectric power plant construction project in 2020. Vietnam has approximately 7000 dams, and only approximately 1150 dams are managed by monitoring systems. Several old dams were constructed before 1990 (Word Bank, 2015). Furthermore, dams managed by monitoring systems are not subjected to integrated management at the national level. They are managed by different systems depending on the installation company, resulting in low efficiency. A management system that is not efficient can lead to the following problems: (a) Dam operations generate vast amounts of data through numerous sensors and devices. Effectively handling and analyzing this large volume of data can be challenging. (b) Ensuring the integrity and accuracy of dam measurement data is crucial. Incorrect or low-quality data can have negative effects on dam safety and stability. (c) Large volumes of critical dam data may be susceptible to security issues, such as hacking or illegal access. Protecting the dam data from unauthorized access is of paramount importance. (d) Integrating and ensuring the compatibility of data obtained from various systems can be challenging. Harmonizing the data generated in different formats and protocols into a consistent form is a complex task. (e) Some dams require real-time data processing. Effectively processing and responding to large volumes of data at high speeds presents technical challenges. (f) Managing and processing large-scale dam measurement data can be resource-intensive and costly. Effectively managing and optimizing these resources is a significant challenge. Therefore, the efficient operation of a monitoring system is essential.

In this study, to manage agricultural dam disasters, which occur within a short period, an AIoT (AI+IoT) monitoring system was developed. In Section 2, a literature review on measurement and monitoring systems was conducted. In Section 3, the distinction between existing monitoring systems and the newly developed AIoT monitoring system was discussed. The AIoT monitoring system includes features such as autonomous operation, malfunction prevention, optimal communication frequency search, communication data authentication, and hacking prevention. In Section 4, the application to the field for performance validation was explained. The study was concluded through Section 5.

2. Literature Review

Approximately 616 dam facilities are present in Korea, of which 45% are more than 30 years old. Among them, 62% are expected to deteriorate in 10 years [4]. Deteriorated agricultural fill dams are vulnerable to collapse. Moreover, the safety of agricultural fill dams in excellent condition can rapidly deteriorate due to heavy rainfall exceeding the design frequency because of climate change. Although safety management of fill dam facilities is increasingly being considered, it continues to be dependent on conventional approaches based on simple measurements or visual inspection. Damage continues to occur annually in countries where technologies for the safety management of dams have not been established.

Dam safety management is essential for the following reasons. (a) Dam failures or safety issues can lead to potential loss of life. Therefore, ensuring the safe operation and proper management of dams is crucial to minimize such risks and guarantee human safety. (b) Dam failures can result in significant damage to the surrounding areas, including homes, farmlands, and infrastructure. Maintaining the structural integrity of dams through safe operations contributes to the protection of nearby properties. (c) Dams primarily serve the purposes of storing water and controlling water resources. Safe dam operations aid in efficient water resource management and help in preparing for natural disasters, such as droughts or floods. (d) Some dams contribute to energy generation through hydropower. Dam safety is essential to ensure continuous and reliable energy supply to meet the power demand of a nation or region. (e) Dam operations can affect the surrounding environment. Safeguarding dams through proper safety management contributes to the protection and preservation of the local ecosystems. (f) Safe dams contribute to the stability and development of local communities. Ensuring a secure water supply and protecting the properties of local residents contributes to the economic and social development of the region. Therefore, research on dam safety is continuously being published.

Jang et al. [5] used GPS to monitor the deformation of an old fill dam structure (length: 680 m and width: 5 m) constructed in 1949. A static GPS observation technique was applied for the monitoring. The coordinate values were obtained by repeatedly observing the deformation of the fill dam structure at a certain time interval. A comparison of these values showed that large deformations occurred at specific monitoring points.

Han et al. [6] developed an automatic visual monitoring system that can acquire real-time digital photographic images, create three-dimensional (3D) coordinates for the surface displacement of a dam, and automatically interpret the vector extraction process. This system can be used in situations where real-time measurements cannot be performed. It can especially address the challenge of applying softcopy photogrammetry, a real-time measurement method that uses photographic images; the method cannot be easily applied owing to the time-consuming interpretation process. The field applicability of the developed system was examined through outdoor testing, and the system showed relatively high precision.

In recent years, various techniques, such as advanced photogrammetry, digital twin, GNSS, deep learning algorithms, and numerical methods [7,8,9,10], have been reported. Lee [11] constructed a 3D model for the Malpasset Dam located in southern France by introducing short-range photogrammetry and observed the dam failure geometry using the model. The 3D model was created by obtaining 158 images from the site and was used for collapse simulation analyses and geological surveys.

Ryu et al. [12] developed an HMI solution to improve dam safety monitoring, thus enabling the transmission of the collected measurement data. Furthermore, they established a communication network as a control network to ensure secure data management, protecting against external hacking and intrusion. However, in the collected data, normal data cannot be distinguished from abnormal data; the presence of abnormal data can cause malfunctioning, resulting in economic losses.

Chao et al. [13] proposed an automated framework based on data microservices for accurate automated analysis of dam monitoring data. The framework comprised structural components, monitoring sensors, finite element models, geometric models, mathematical models, and digital virtual models integrated with deep learning algorithms. Long short-term memory was used for accurate predictions. The framework was tested on a dam safety monitoring site. The results showed that it improved the prediction accuracy of the monitored data with an error of 0.35 mm.

In contrast to the above-mentioned studies, this study implements a “malfunction prevention” feature to enhance the accuracy. This malfunction prevention feature improves the accuracy of the monitored data by compensating for measurement errors or communication errors. In this study, we employed a statistics-based data filtering technique to achieve this feature, thus minimizing false alarms in anomaly detection.

Yoon et al. [14] constructed a real-time displacement monitoring system based on GNSS in 37 dams in Korea. The system consists of a base, rover, and server. The base was installed at a point that can represent the ground around a dam facility, and the rover was installed at the major points of the dam structure to monitor the external displacement. The base, which can identify accurate coordinate data, was constructed to correct the errors generated during the signal transmission process from the satellite to the GNSS receiver. The base offsets the errors by transmitting the corrected data to the rover with the same errors through the satellite. The displacement data thus obtained were compared with the data measured using an electronic distance measurement device. The results showed that stable measurement was performed, as the distribution range of the GNSS-based displacement monitoring system data was within ±5 mm.

In contrast to the previous studies, our study offsets errors through data validation by using “sensor nodes” and “gateways”. The measurement data obtained from the sensor are transmitted to the sensor node to perform a cyclic redundancy check (CRC), and the generated CRC data are selected through data verification at the gateway. The unselected data are excluded by performing re-measurements, and only the selected data are sent to the server to ensure data quality.

Peng et al. [15] established a prediction model using an artificial neural network to monitor the deformation of a dam based on its subsidence measurement data. The proposed model was proven to be effective and highly practical in predicting dam deformation.

Mohod [16] developed IoT applications in Dam Safety and water management. The IoT applications can monitor the entire dam and the main pipeline for 24 h, every day through various sensors. These wireless sensor nodes are connected to each other and transmit the data to a gateway. Common storage space ‘CLOUD’ stores provide online information to the observer. This has contributed to dam safety and water conservation.

Liu et al. [17] proposed a hybrid prediction model based on Bayesian Optimization and Random Forest to predict dam deformation. Initially, the monitoring data were pre-processed, and the parameters for the Random Forest model were configured. After determining the optimal parameters through Bayesian Optimization based on Gaussian processes, the relationship between the dam deformation and influencing factors was explained using the Gini coefficient. The developed prediction model demonstrated high prediction accuracy in tests by showing only a small discrepancy between the predicted values and actual measurements.

Yihong and Afzal [18] designed a computerized system that enables the use of IoT for the safety monitoring of dams and mega hydropower projects. When security parameters depart beyond an anticipated level, this system promptly warns the relevant authority to take necessary action. By controlling dam security through this system, the danger of a significant breakdown of the dam is decreased.

Gong et al. [19] proposed a dam safety monitoring and early warning method based on a graph database and aggregation model. This system solved the problem of traditional methods having difficulty handling complex relationships between dam monitoring points. In addition, various monitoring system research is contributing to the safety and enhancement of dams [20,21].

In this study, the limitations of the existing fill dam monitoring methods were identified, and solutions were proposed. The existing IoT system-based monitoring method cannot respond to abnormal data measurements, and it degrades the reliability of the measurement data because data with errors are used. In addition, it measures and analyzes only the current data; consequently, future emergencies cannot be rapidly addressed. Moreover, the system has security vulnerabilities, resulting in a risk of hacking. Therefore, it is essential to collect real-time monitoring data efficiently and, through this, promote a proactive approach for monitoring systems to plan dam maintenance and repair operations effectively. This can lead to cost savings and enhanced maintenance efficiency. Additionally, leveraging accurate data to predict the future performance of dams and identify potential issues is crucial for enhancing safety. To address these issues, an AIoT (AI+IoT) monitoring technology for dam facilities was developed in this study.

3. AIoT Dam Monitoring Technology

3.1. Overview

To monitor social infrastructure, it is necessary to go beyond simple measurement and information communication. Instead, it is necessary to autonomously adjust the sensor measurement cycle by recognizing the communication between the sensor and device when hazardous factors such as heavy rain, typhoons, and earthquakes are detected.

In this study, we improved the existing automated measurement system to realize an AIoT-based monitoring system by enhancing the system’s ability to detect sensor measurement malfunctions, which is a problem inherent to conventional automation technology. In addition, the monitoring system has been enhanced to maintain communication efficiency and enable system maintenance based on low power consumption, even in outdoor environments. Furthermore, the system has been improved to surpass a simple monitoring technology, allowing for immediate evaluation of the safety condition of facilities by utilizing sensor measurements and simultaneously predicting the time to reach the limit state, thus maximizing the AIoT functionality of the monitoring system.

The improved AIoT system exhibited the following functions: autonomous operation, malfunction prevention, optimal communication environment construction, machine learning prediction, and data authentication and security. Based on these characteristics, we attempted to secure faster and safer dam safety management technology by analyzing the fill dam measurement data in real time.

Table 1. Problems in the current technology and advantages of the developed AIoT technology.

Category	Problems in the Current Technology	AIoT Technology
(1) Malfunction prevention technology	Sensor measurement error due to unstable voltage and sensing noise ⇒ Decreased reliability of measurement data ⇒ Frequent alarm notifications	Application of a statistics-based data filtering technique ⇒ Measurement data error removal and reliable alarm notification
(2) Communication performance improvement	Use of the open-frequency area of shared Wi-Fi and LoRa communications ⇒ Communication delay	Frequency channel setting feature (0–15 channels) ⇒ Improved communication performance by using a frequency with high communication sensitivity
(3) Autonomous operation	Limited to periodic measurements (once per hour, week, day) and simple communication capabilities ⇒ Impossible to perform precision measurement when a risk occurs	Autonomous switching from normal measurement (once per hour) to precision measurement (once per minute) in the event of a risk ⇒ Precision measurement/bi-directional communication is possible
(4) Data authentication	Use of open IoT communications (LoRa and Wi-Fi) ⇒ Problem in collecting error data (other communication data)	CRC checking when collecting measurement data (received data authentication) ⇒ Verification of data quality and reliability
(5) Prediction technology using machine learning	Simple measurement data monitoring ⇒ The state of safety of the facility cannot be recognized ⇒ Expert judgment is required	Assessment of safety status based on real-time data collected ⇒ Use of normal and anomaly detection information to predict safety status
(6) Security features	Vulnerable to external connections and hacking on open communication frequencies ⇒ Decreased security	Data transmission after pattern encryption ⇒ Prevention of hacking and enhancement of security features ⇒ Smooth control of devices/systems

summarizes the problems in the current technology and the advantages of the AIoT technology developed in this study.

3.2. Autonomous Operation

Existing IoT sensors have only measurement and simple communication functions. Consequently, the function of precise measurements for abnormality detection or disaster prediction cannot be expected. However, the developed intelligent IoT sensor enables autonomous switching between normal measurement (once per hour) in daily life and precision measurement (once per minute) for abnormality detection and disaster prediction. Figure 3 shows the results of the autonomous operation performance test of the intelligent IoT system. Data were measured from 07:00 to 14:00. The results showed that normal measurement was replaced by precision measurement for one hour, i.e., from 08:00 to 09:00, because a load that exceeded the design load (14 Tonf) was detected. Figure 4 shows the results of the autonomous operation performance test for the pore pressure. Operation without errors in the pore pressure was also confirmed.

3.3. Malfunction Prevention

IoT systems linked to forecast and warning systems may exhibit measurement errors or communication errors because of communication interference. This may cause the malfunctioning of the warning systems, resulting in delayed prediction and increased cost. Therefore, an intelligent IoT sensor was developed to address the malfunctioning problem. A malfunction prevention function was developed, which employed a data filtering technique based on statistical analysis of the measurement data. Errors that could occur in the sensor measurement were removed using the function. Figure 5 is an example of measurement error. The process of IoT sensors for malfunction prevention is as follows: (1) Receive the current value from the measurement sensor through the interface. (2) Save the data. (3) Generate cumulative distribution data using saved values. (4) Convert the cumulative distribution data to normal distribution data for each sensor. (5) If the measured value is within the filtering confidence interval, it is transmitted to the administrator server. (6) If the measured value is not within the filtering confidence interval, the current measured value entered is discarded.

3.4. Optimal Communication Environment

IoT systems that use unlicensed communication frequency ranges may exhibit communication performance degradation and communication errors because of communication interference, thereby lowering the reliability of the measurement data. Hence, a function that automatically searches for channels in the open-frequency ranges was added to the intelligent IoT sensor. The communication performance was improved by determining the optimal communication frequency in the field using this function (Figure 6).

3.5. Data Authentication and Security

Existing IoT sensors are vulnerable to data security as they use an open communication environment, thereby lowering the reliability of the field measurement data. To address this issue, data quality was ensured through a CRC during the collection of measurement data, and data security was strengthened by transmitting the measurement data to the platform after encryption (Figure 7). Figure 8 shows the conceptual diagram of the security function that transmits the data patterns through encryption for IoT system control. Figure 9 shows the sensor nodes and gateways developed using the conceptual diagram.

Table 2. Specifications of sensor node.

Category	Specifications
Sensor measurement	Electrical resistance type, electric type, RS485, and RS232 simultaneous measurements
Measurement cycle	1 min to 60 min or higher (measurement cycle control feature included)
MCU	ARM Cortex M3 32-bit 120 MHz
Internal memory	Minimum 128 KB (Flash Memory)
Power supply	Always-on, primary battery, secondary battery + solar power
Applied voltage	3.3–12 V
Node communication method	LoRa, Wi-Fi, LTE
Operating temperature	−20~60 °C, humidity 80% or lower
Durability	IP 67 or higher
Quality assurance	Minimum 2 years

and

Table 3. Specifications of gateway.

Category	Specifications
Number of node connections	20
MCU	ARM Cortex M3 32-bit 120 MHz
Internal memory	Minimum 128 KB (Flash Memory)
Power supply	Always-on, primary battery, secondary battery + solar power
Applied voltage	3.3–12 V
Node communication method	LoRa, Wi-Fi, LTE
Server communication method	Router communication (3G, 4G, 5G)
Operating temperature	−20~60 °C, humidity 80% or lower
Durability	IP 67 or higher
Quality assurance	Minimum 2 years

show the specifications of the sensor node and gateway.

4. Field Application of AIoT Monitoring Technology

4.1. Field Test in Korea

For field application, the developed AIoT technology was tested in a reservoir in Korea. Figure 10 shows the reservoir, and Figure 11 shows the standard cross-section of the test bed. In the target structure, a groundwater level meter and an inclinometer were installed for general measurements, and volumetric water content gauges and piezometers were installed for wireless sensor network measurement. The normal operation of each of the installed measuring instruments could be examined through the system shown in Figure 12a. Notably, the communication conditions were excellent (Figure 12b). Figure 13 shows the performance test results for the malfunction prevention function. Figure 13a shows the results before the application of the malfunction prevention function, and Figure 13b shows those after the application of the function. Notably, the error values (marked in red circles) occurred before the application of the function. Accurate data could be obtained, as the error values were removed through data filtering. The field test confirmed the normal operation of each system.

4.2. Future Field Test Plan in Vietnam

The normal operation of the developed AIoT system was confirmed via the field test in Korea. To verify its applicability to diverse sites, a field test will be conducted in Vietnam. A fill dam structure with a length of 220 m and a height of 33 m near the Hoa Binh province has been selected as the test bed through a preliminary on-site survey. Figure 14 shows the fill dam structure of the test bed in Vietnam. In the target structure, nine inclinometers and one piezometer will be installed to examine the displacement of the dam slope and water leakage, as shown in Figure 15. In addition, as the temperature in Vietnam is different from that in Korea, additional parameters will be considered. The ambient temperature will be sensed using the sensor and system installed at the site, and the normal temperature range will be set. If measurements are performed at temperatures beyond this range, the error values will be eliminated using data filtering.

5. Conclusions

In this study, the convenience and safety management of dam monitoring were improved using the AIoT technology for fill dams, where damage is frequently caused by heavy rainfall and monitoring is performed using conventional safety management technologies, which have several challenges.

Therefore, we developed an AIoT monitoring system with the following functions: an autonomous operation function that detects abnormal conditions and autonomously adjusts the measurement frequency; a malfunction prevention function that utilizes a statistical analysis-based data filtering technique; an optimal communication environment construction technology that automatically searches for channels and determines the optimal frequency in the frequency range; a machine learning prediction function; and a data authentication and security function. The autonomous operation of the load sensor and pore pressure gauge has been tested. Through this function, continuous monitoring (once per hour) transitions to precision measurement (once per minute) due to the occurrence of abnormal conditions. The data filtering system confirmed the generation of cumulative distribution and normal distribution data based on measurement values. Also, a filtering reliability of 90% or higher is drawn. Establishing an optimal communication environment and enhancing data authentication functions are worked seamlessly to the Korean testing without any error. However, this is configured with sensors suitable for the Korean environment. Subsequently, this system is scheduled to be tested in Vietnam.

The performance of the developed AIoT monitoring system was verified by applying it to a test bed in Korea. The normal operation of each function was confirmed, and the reliability of the data was improved by removing the error values using a data filtering technique.

In the future, additional data will be obtained, and predictions will be performed using machine learning with the test bed in Korea. In addition, a system suitable for Vietnam will be constructed by developing a measurement data correction technology based on the on-site temperature. This system will be applied to a site in Vietnam to test its application to diverse sites.