Smart Toilet IoT Project

Smart Toilet IoT Project — Topics for IoT Students

SUBMIT TO IEEE INTERNET OF THINGS MAGAZINE 1

AIMNET: An IoT-Empowered Digital Twin for Continuous Gas

Emission Monitoring and Early Hazard Detection Zifan Zhou1 ,* , Xuan Wang1 ,* , Yang Yan2 , Lkhanaajav Mijiddorj2 , Yu Ding3 , Tyler Beringer2 , Parisa Masnadi Khiabani 4 , Wolfgang G. Jentner 4 , Xiao-Ming Hu3,7 , Chenghao Wang 3,6 , Bryan M. Carroll5 , Ming Xue3,5 , David Ebert 2 Fellow, IEEE, Bin Li1 Senior Member, IEEE, Binbin Weng 2,† Senior Member, IEEE

Abstract—A Digital Twin (DT) framework to enhance carbon- volatile organic compounds (VOCs), which are associated based gas plume monitoring is critical for supporting timely with serious health risks, including cancer [3]. From the and effective mitigation responses to environmental hazards such industrial aspect, CH4 is highly flammable, posing severe as industrial gas leaks, or wildfire outbreaks carrying large risks in mining, and oil and gas production sectors [4]. arXiv:2512.06148v1 [cs.NI] 5 Dec 2025

carbon emissions. We present AIMNET, a one-of-a-kind DT framework that integrates a built-in-house Internet of Things These challenges highlight the important needs for a scalable (IoT)-based continuous sensing network with a physics-based real-time and high-resolution gas leak monitoring systems to multi-scale weather-gas transport model, that enables high- facilitate timely and effective actions to mitigate environmental resolution and real-time simulation and detection of carbon gas harm, safeguard public health, prevent catastrophic accidents emissions.

AIMNET features a three-layer system architecture: (i) physical world: custom-built devices for continuous monitoring; and enhance economical robustness. (ii) bidirectional information feedback links: intelligent data Recent advances in the Internet of Things (IoT) are ushering transmission and reverse control; and (iii) digital twin world: AI- in a new era of emission monitoring, enabling real-time, driven analytics for prediction, anomaly detection, and dynamic automated data acquisition with cost-effective and remote weather-gas coupled molecule transport modeling.

Designed for deployment [5]. Industries have widely adopted these systems scalable, energy-efficient deployment in remote environments, AIMNET architecture is realized through a small-scale dis- for specific, critical tasks like gas monitoring on production tributed sensing network over an oil and gas production basin. sites, structural health monitoring of bridges, and water-level To demonstrate the high-resolution, fast-responding concept, monitoring for flood alerts.

However, the practical utility of an equivalent mobile-based emission monitoring network was current IoT-based monitoring systems has been constrained deployed around a wastewater treatment plant that constantly by a critical limitation: narrow geographical coverage. Most emits methane plumes. Our preliminary results through which, have successfully captured the methane emission events whose existing deployments are confined to specific facilities or urban dynamics have been further resolved by the tiered model simu- areas [6], limiting their ability to capture the broader spa- lations.

This work supports our position that AIMNET provides tiotemporal dynamics of gas transport and accumulation. This a promising DT framework for reliable, real-time monitoring limitation hinders the accurate attribution of emissions sources, and predictive risk assessment. In the end, we also discuss environmental impact assessment, and the development of key implementation challenges and outline future directions for advancing such a new DT framework for translation deployment. reliable forecasting models, thereby undermining large-scale regulation and mitigation strategies.

To address these limitations, researchers have begun devel- oping large-scale IoT platforms aimed at expanding spatial I. I NTRODUCTION coverage and enhancing monitoring capabilities across broader Carbon gases, including carbon dioxide (CO2 ) and methane regions, such as IoT-Mobair for air quality tracking [7] and the (CH4 ), are increasingly valued due to their wide-ranging Internet of Maritime Things for marine surveillance [8].

These impacts on climate, human health, and industrial safety man- systems leverage low-power wide-area networks (LPWANs) agement. Environmentally, carbon gases exacerbate global and heterogeneous sensor arrays to extend coverage across warming, making it a significant contributor to climate change urban, industrial, and remote environments. However, despite [1].

From a health standpoint, elevated CO2 concentrations in their promise, critical limitations remain. First, network relia- closed or poorly ventilated environments can displace oxygen, bility is a persistent challenge: LPWAN links often experience leading to symptoms such as dizziness, fatigue, and cognitive intermittent outages in rugged terrain or under extreme envi- impairment [2]. In addition, CH4 emissions from oil and ronmental conditions, leading to delayed or lost data.

Second, gas production processes are often accompanied by various spatial and temporal resolution is frequently insufficient, with *Zifan Zhou and Xuan Wang contributed equally to this work. many systems operating at coarse scales (e.g., ≥ 4.0 km or † Binbin Weng (binbinweng@ou.edu) is the corresponding author. ≥ 1.0 h [9]), failing to capture transient or localized emission 1 Department of Electrical Engineering, Pennsylvania State University, Uni- events.

Third, intelligence and responsiveness are limited, as versity Park, PA, USA. 2 School of Electrical and Computer Engineering, Uni- most platforms function primarily as passive data collectors versity of Oklahoma, Norman, OK, USA. 3 School of Meteorology, University of Oklahoma, Norman, OK, USA. 4 Data Institute for Societal Challenges, without integrated modeling, anomaly detection, or forecasting University of Oklahoma, Norman, USA. 5 National Weather Center, University capabilities.

Finally, data visualization is often rudimentary, of Oklahoma, University of Oklahoma, Norman, OK, USA. 6 Department lacking context-aware synthesis or decision-support interfaces of Geography and Environmental Sustainability, University of Oklahoma, Norman, OK, USA. 7 Center for Analysis and Prediction of Storms, University such as dynamic heatmaps or multi-parameter correlation of Oklahoma, Norman, OK, USA. views, which are crucial for rapid interpretation and response

SUBMIT TO IEEE INTERNET OF THINGS MAGAZINE 2

in multi-hazard scenarios. The main contributions of this paper are: Addressing these challenges requires a paradigm shift from • Built-In-House IoT Gas Sensing Instrument: We create passive, static data collection toward an intelligent environ- a non-dispersive infrared (NDIR) technology-based IoT gas mental monitoring paradigm. Unlike conventional IoT systems sensor that features a single digit-parts per million (ppm) that rely on fixed sensor placements, we can leverage mobility sensitivity towards CH4 , and a highly-integrated hardware and autonomy to enable dynamic data collection.

The system and software system allowing the continuous and robust field can actively navigate the environment, detect elevated emission operation, and self-calibration functions. Experimental results zones, and reposition itself in real-time for more accurate and show that our device attains approximately 90% accuracy of localized monitoring of hazardous gases or potentially haz- the benchmarking LI-7700 instrument in detecting CH4 leaks, ardous leak zones.

To enable such autonomous functionality, while taking only a few percentage of the cost compared to the system must possess a comprehensive understanding and its high price. predictive simulation of its deployment environment. Digital • Continuous Gas Emission Monitoring Network: Lever- Twin (DT) provides this capability by establishing a real- aging the lightweight Message Queuing Telemetry Transport time, data-driven digital twin world (DTW) of the physical (MQTT) protocol, we design a bidirectional communication world.

Unlike traditional simulation, the DTW mirrors the layer that supports real-time data streaming, intelligent sens- dynamics of the real physical system in real time, enabling ing control, and remote device management, even in harsh intelligent decision-making based on both historical data and environments. the current system state. Additionally, the DTW can predict • AI-Powered Modeling, Visualization and Analysis: Em- future phenomena through a comprehensive understanding powered by our DT framework which integrates an intelligent of system dynamics, including “what-if” scenarios generated sensing data analysis layer and a physics-based multiscale by its own decision-making capabilities.

As an emerging weather-gas modeling layer, a web-based visualization inter- paradigm, DT offers a powerful technical framework for face with interactive heat maps and multi-parameter correla- bridging the gap between the physical world and DTW in tion views is created. This visualization platform enables the many practical applications. For example, DT has been used real-time modeling and visualization of the emission dynamics in urban traffic management to support predictive simulations over targeted infrastructures. for real-time decisions like adjusting traffic light logic [10]. • Real-World Implementation at Scale: We implement Beyond transportation, DT also finds applications in smart our AIMNET architecture through both a small-scale distributed buildings (e.g., energy optimization, temperature control), sensing network with 23 static field nodes over an oil and gas precision manufacturing (e.g., predictive maintenance, per- field, and a mobile-based plume scanning network mimicking formance monitoring) [11], and some human-centric sys- the envisioned large-scale distributive network for proving the tems [12].

However, the integration of DT into intelligent concept. The static deployment primarily supports large-scale environmental monitoring, particularly for carbon-based gas visualization, while data trustworthiness is validated through emission detection, remains unexplored. To the best of our mobile detection using a high-precision LI-7700 benchmark knowledge, our work is also the first monitoring framework sensing instrument. that employs a basin-scale distributed sensing network for detecting methane plumes.

By fusing IoT-based sensing with II. AIMNET A RCHITECTURE AI-driven analytics and physics-informed simulations, a DTW AIMNET is architecturally structured into three primary can continuously interpret, predict, and respond to the envi- layers: i) Physical world: front-end data collection layer, ii) ronment. Specifically, this framework enables the integration Bidirectional information feedback links: middle-end commu- of the advanced Weather Research and Forecasting model with nication layer, and iii) Digital twin world: back-end analytics greenhouse gases (WRF-GHG) modeling, commonly used layer, as illustrated in Fig. 1.

The main functions of the to predict and retrieve regional, low-resolution greenhouse physical layer are environmental data collection and trans- gas dynamics, and geometry-resolving large-eddy simulation mission. The device’s microcontroller intelligently coordinates (LES) to resolve methane emissions and identify their sources each electronic component for low-power, adaptive, and au- with unprecedented spatiotemporal resolution and accuracy. tonomous field operation.

The communication layer is built Furthermore, it enables context-aware visualization interfaces upon the lightweight MQTT protocol and supports reliable that facilitate rapid decision-making in complex scenarios. bidirectional information exchange between field devices and In essence, DT addresses the core deficiencies of existing the DTW. It enables the upstream transmission of real-time monitoring systems, laying the foundation for regional-scale sensor data to the analytics platform and the downstream monitoring and early hazard detection. delivery of control commands from operators to edge devices.

We introduce AIMNET , an IoT-empowered DT platform The DTW functions as a cloud-based platform for data-driven designed to deliver intelligent monitoring and real-time de- secondary application development and intelligent manage- cision support. AIMNET comprises three core components as ment. It integrates an interface for real-time and historical introduced in Section II.

By seamlessly integrating customized data visualization, along with AI-driven modules for data hardware, resilient networking, and AI-empowered analytics, enhancement, anomaly detection, and predictive modeling. AIMNET addresses the critical deficiencies of existing systems, This enables actionable environmental insights and supports offering a robust, scalable foundation for next-generation en- the development of advanced applications for hazard detection vironmental monitoring. and decision-making.

SUBMIT TO IEEE INTERNET OF THINGS MAGAZINE 3

Fig. 1: The large-scale Digital Twin gas monitoring framework integrated with a distributed IoT gas sensing network, advanced gas transport modeling and data visualization technologies.

A. Physical World: Front-end Data Collection Layer

The physical layer serves as the data source, acquiring and transmitting high-fidelity environmental data from the field. Achieving this requires overcoming significant challenges, including the need for long-term operational reliability, energy efficiency, and data accuracy in harsh, uncontrolled outdoor environments. To meet these demands, we develop a fully integrated, custom-built IoT sensing device, as illustrated in Fig. 2.

Unlike conventional commercial gas detectors that Fig. 2: Prototypes of IoT gas sensing instrument: left one rely on off-the-shelf development boards or pre-packaged features a completed system with battery and power manage- sensor modules, our approach provides full control over the ment modules; middle one is the revised version isolating the hardware stack and enhances mechanical durability. We retain core sensing control from the battery module for enhanced only the essential functional components (i.e., sensing front- field operating robustness; right one shows the total power end, cellular modem, and voltage converter) and consolidate consumption of our device. them onto a unified, application-specific PCB.

This integration significantly reduces hardware complexity, minimizes idle and peak power consumption, lowers the bill of materials cost, can be affected by environmental factors, such as humidity, and decreases overall size without sacrificing functionality. As temperature, and atmospheric pressure [13]. To mitigate this, Fig. 2 shows, our device achieves a power consumption of just we move beyond traditional signal processing and develop 1404 mW , compared to the LI-7700’s 8 − 41 W . a custom, regression-based machine learning (ML) model to The hardware architecture is designed for autonomous field refine data as introduced in II-C. operation.

In detail, the control and sensing functions are managed by an Adafruit Feather M0 microcontroller, which serves as the central hub. For high-fidelity sensing, it utilizes B. Bidirectional Information Feedback Links: Middle-end an NDIR technology sensor, chosen explicitly for outdoor Communication Layer detection due to its robust resilience and significantly re- The AIMNET communication layer, bridging the physical duced susceptibility to environmental influences compared to world and DTW, addresses the critical challenge of reliability electrochemical sensors.

Furthermore, this sensor offers long- in regional IoT deployments. It enables reliable data collection term reliability with significantly lower power consumption from the physical world and supports timely operational compared to other optical instruments. A solar-powered energy suggestions from the DTW to physical entities.

In harsh harvesting module provides the power, ensuring consistent environments, such as rural oil fields, intermittent connectivity performance even for an entire week without recharging. This and limited hardware can hinder data transmission. Our archi- enables a low-power, narrow-band SIM7070G communication tecture provides a resilient and secure framework explicitly module to reliably and efficiently transmit data. engineered for these demanding conditions.

A cornerstone of our design is the tight integration of Reliability, ensuring stable data streaming and timely com- hardware and software to ensure data credibility. While optical mand delivery, is the foundation of DT systems. However, NDIR sensors are cost-effective and sensitive, their accuracy harsh networking conditions in the field often result in high

SUBMIT TO IEEE INTERNET OF THINGS MAGAZINE 4

packet loss and delayed command delivery, and commer- 2) DT-enabled Modeling and Analysis: The heart of this cial IoT platforms can exacerbate the problem due to their layer is a DTW that dynamically fuses data from the physical lack of transparency and control over critical performance world and physics-based multiscale models to create a real- parameters. To overcome these limitations, AIMNET uses a time, virtual representation of the monitored environment. self-hosted MQTT broker.

We chose MQTT for its efficient Specifically, we use the WRF-GHG to simulate near-real-time publish-subscribe architecture, which avoids the high latency meteorological conditions and greenhouse gas concentrations of HTTP’s request-response model for each data exchange and at a resolution of approximately 1 km. These outputs then offers greater reliability than UDP-based protocols like CoAP. serve as boundary conditions for a geometry-resolving LES, This is because MQTT provides robust message acknowl- which enables much finer-scale simulations at resolutions edgment and session recovery, which are essential for real- ranging from 0.5 to 5 m.

This multiscale representation is time digital twin synchronization. This approach provides fine- continuously updated with live sensor data, accurately re- grained control by assigning each IoT device a unique topic for flecting the evolving physical conditions of the region. By both publishing sensor data and receiving commands.

In detail, the closed-loop synchronization between physical and virtual for forward data transmission, each IoT device is assigned components, the DTW recalibrates in response to environmen- a unique MQTT topic. The central MQTT broker classifies tal changes (e.g., wind direction, humidity, temperature), con- incoming data based on these topics, allowing the DTW sistent with the concept of “living DTs” [14]. The DTW also to reconstruct the real-world environmental dynamics using integrates simulation and inverse modeling techniques to op- the topic-associated data streams.

For reverse management, timize sensor placement, identify likely emission sources, and such as delivering operational suggestions or enabling remote evaluate mitigation strategies, which enhances the system’s configuration of device parameters, IoT devices are designed capability to assess gas leaks under varying meteorological to receive commands either through designated MQTT topics conditions. or via direct operator SMS messages.

To balance energy efficiency and timeliness, we use a duty-cycled policy, 5-minute sampling followed by 1-minute uplink transmission. We further evaluated the communication performance to verify reliability under real-world conditions. Transport latency (device-to-dashboard once sent) is seconds- scale, while dashboard freshness remains within 6 minutes.

With MQTT QoS 1 and session recovery enabled, the system achieves a packet loss below 0.1% for transmission intervals of ≥4 seconds, with each node maintaining an average uplink bandwidth below 1 kbps.

Fig. 3: Real-time and historical visualization. C. Digital Twin World: Back-end Analytics Layer 3) Data Visualization: In an era demanding increasingly The last layer serves as the intelligence core of AIMNET , granular and responsive environmental oversight, traditional functioning as both the data center and the DTW.

It hosts the data analysis methods often fall short of providing the dynamic MQTT broker, manages data storage, supports visualization, insights necessary for effective decision-making. To bridge this and runs AI-driven analytics to enable real-time environmental critical gap, we develop an interactive environmental monitor- monitoring, predictive analytics, modeling, and system-wide ing dashboard as depicted in Fig. 3 that transforms disparate decision-making. sensor readings into actionable intelligence.

At its core, the 1) AI-empowered Calibration: To enhance data complete- system features a robust backend that supports high-volume ness and mitigate the influence of environmental factors on data ingestion from diverse sources, including fixed sensors, sensor readings, we deploy an ML model trained on a com- mobile platforms, and external weather stations, providing a prehensive dataset comprising measurements from both labo- comprehensive understanding of the environment.

The user- ratory and field environments. The dataset includes CH4 and facing interface offers an intuitive experience, integrating real- CO2 concentrations along with corresponding environmental time geospatial maps with interactive time-series visualiza- parameters, enabling the model to learn correction patterns tions. This synchronized display enables users to identify under diverse conditions.

Model performance is validated subtle trends and anomalies across customizable time win- using two independent datasets: one collected in an indoor dows and immediate localization. Both real-time and historical environment and the other in an outdoor, dynamic setting. data are presented through interactive heat maps and multi- Results indicate that the model reduces the mean absolute parameter correlation views. AI modules enhance system deviation in methane readings to approximately ±3 ppm and intelligence by detecting anomalies and imputing missing data, achieves coefficients of determination of R2 = 0.948 for effectively addressing both short-term gaps and longer out- the indoor dataset and R2 = 0.908 for the outdoor dataset, ages by leveraging neighboring sensor information.

Integrated enabling DT to maintain stable performance even under highly weather model overlays enrich sensor data with atmospheric variable field conditions. context, enabling users to trace potential emissions sources.

SUBMIT TO IEEE INTERNET OF THINGS MAGAZINE 5

Advanced features, such as auto-refreshing, sensor isolation particularly under highly dynamic conditions, since air must controls, and optional infrastructure overlays (displaying oil traverse the intake tubing and mixing chamber before reaching and gas facilities or calibrated methane measurements), further the sensing cell. The geographic information system-based enhance analytical precision. These capabilities provide a comprehensive and actionable view of the evolving environ- ment, empowering operators to detect hazards early and make proactive decisions with confidence.

III. P RELIMINARY I MPLEMENTATION AND E VALUATION

To evaluate the trustworthiness of AIMNET , we deploy both static and dynamic operational modes. While both CH4 and CO2 are measurable, the implementation focuses on CH4 due to its greater challenges, though the method also applies to broader CO2 emission monitoring. In the static mode, a total of 23 IoT devices are deployed onto the power posts along the road over the Anadarko basin (Fig. 4), which includes environmentally complex regions such as oil wells.

These Fig. 5: Mobile CH4 measurement. devices were installed approximately 2.5 meters above the ground and spaced at intervals of 0.5 miles to optimize both figure on the right provides a detailed spatial visualization of solar energy harvesting and high-resolution gas detection. The methane detection locations. It further confirms that AIMNET static network has been continuously operational from March successfully identifies all high-concentration methane areas 2024 to July 2025. that are also detected by the LI-7700.

The only missed region is a mid-level concentration zone (approximately 4 ppm) located in the lower left corner. The 15 ppm false positive occurred on the highway, likely due to rapid vehicle motion. In detail, our mobile platform incorporates a vehicle-mounted system, a drone-equipped optical gas imaging camera, and an integrated weather station to collaboratively capture real-time methane plume dynamics.

We further apply our multiscale modeling framework over the same study site, nesting the geometry-resolving LES model within the WRF-GHG simulation domain. To illustrate the value of explicitly resolving built structures, here we first increase the WRF-GHG resolution to 32 m, approaching Fig. 4: Experimental deployment (Map from Google Maps). LES-scale turbulence representation but without the ability to resolve building geometry.

We then conduct a geometry- For the dynamic mode, a mobile methane monitoring system resolving LES simulation at a finer horizontal resolution of is tested around the City of Norman Water Reclamation Facil- 5 m, using lateral boundary conditions from the 32-m WRF- ity to demonstrate a practical IoT-based solution for detecting GHG simulation. and modeling local and regional methane leakage. To validate the accuracy of our device, we employ a high-precision LI- 7700 sensor as the reference standard and conducted compar- ative measurements using AIMNET .

Fig. 5 presents the testing results. The upper two images show the output from AIMNET , while the lower image displays the reference measurements. It can be observed that AIMNET successfully captures all the major methane concentration peaks detected by the LI-7700.

This indicates that when the methane concentration exceeds 5 ppm, our sensor is capable of accurately identifying the elevated levels. Only one false positive is reported, with a concentration reading of 15 ppm. We observe a time offset between them, attributable to their fundamentally different sensing architectures: the LI-7700 directly measures methane Fig. 6: Physics-based multiscale modeling system. concentration in the ambient air stream with negligible res- idence time, whereas AIMNET utilizes an active pumped- Fig. 6 compares the results from both simulations.

While inlet designed for static environmental monitoring. This con- both models reasonably capture CH4 concentrations near the figuration introduces a small but systematic transport delay, emission sources (anaerobic digesters), substantial differences

SUBMIT TO IEEE INTERNET OF THINGS MAGAZINE 6

emerge farther downwind, particularly in regions behind build- analytics a formidable challenge. Traditional models like Inte- ings. The geometry-resolving LES successfully reproduces the grated Moving Average are ill-equipped to handle the nonlin- meandering behavior of the CH4 plume, a feature not captured ear, non-stationary dynamics of gas dispersion, especially with by the coarser WRF-GHG simulation.

In addition, the high- noisy or incomplete data [15]. Furthermore, operators face resolution geometry-resolving LES can spatially differentiate cognitive overload from cluttered visualization interfaces that emissions from multiple, closely spaced sources that would fail to convey spatial-temporal patterns and data uncertainty otherwise be lumped together within a single WRF-GHG grid clearly. cell. Compared to WRF-GHG, the LES also aligns more AIMNET addresses these challenges with a robust, intelli- closely with field measurements taken along the road east gent analytics core that enables high-frequency monitoring of the emission sources (Fig. 5), providing a more accurate and real-time visualization.

Its modular, scalable architecture representation of plume dispersion. This improved agreement leverages distributed computing tools such as Apache Kafka is critical for the reliable identification and quantification and Spark for high-volume, sub-second data processing, with of emission sources and their strength. Since WRF-GHG the potential for lower-latency stream-first integrations.

To is typically run at even coarser resolutions in real-world, overcome the limitations of the conventional statistical models, operational deployments (e.g., 1 km), these results highlight AIMNET employs advanced ML pipelines, like long short- the value of a multiscale modeling system, particularly when term memory networks, for accurate gas dispersion modeling. applied to oil and gas infrastructure with complex geometries.

The DT framework enables accurate, physics-informed leak localization by comparing real-time sensor data with LES IV. A DDRESSING K EY C HALLENGES WITH AIMNET outputs, dynamically integrating live environmental data, back- ground gas concentrations, and weather conditions, thereby Achieving fully autonomous and self-regulating carbon- distinguishing true anomalies and reducing false positives. based gas monitoring and hazard prediction remains chal- The system supports semantic temporal fusion for aligning lenging, particularly in large-scale deployment and accurate asynchronous data streams.

Insights are delivered through a detection. Next, we outline the key research challenges and web-based, interactive visualization platform, designed for fu- describe how AIMNET is specially designed to address them. ture integration with Large Language Model-powered natural Reliable Sensing & Edge Intelligence: language queries (e.g., “What is the most likely emission Reliable gas sensing is always a significant challenge for source near node 17?”), transforming fragmented data into large-scale deployment.

Low-cost sensors enable broad spa- clear, actionable intelligence. tial coverage but suffer from limited sensitivity, signal drift, Trustworthy & Secure AI Integration: and cross-gas interference, where issues are exacerbated by Deploying AI in large-scale environmental monitoring holds environmental factors such as humidity and temperature that transformative promise but also faces key challenges. Reliable corrupt the data.

For example, in oil fields or agricultural models require extensive, high-quality labeled data, which is zones, CH4 often coexists with CO2 and volatile organic com- often scarce due to the rarity of critical events and noisy pounds, hindering its detection accuracy. Conversely, while field data (e.g., methane leaks, sensor tampering). Real-time high-precision instruments like tunable diode laser absorption performance necessitates balancing inference accuracy and spectroscopy and Fourier-transform infrared spectrometers of- computational efficiency, particularly for edge computing. fer superior accuracy, their cost, power requirements, and Moreover, in safety-critical and regulated settings, AI outputs bulk render them impractical for the widespread, distributed must be transparent, explainable, and auditable. networks essential for regional-scale monitoring.

To address these challenges, AIMNET adopts a modular AIMNET directly addresses this dilemma through an innova- AI framework that leverages edge-cloud cooperation and is tive blend of custom hardware design and ML-driven calibra- deeply integrated with the DT system. The DT framework tion. The system continuously ingests real-time sensor streams provides historical context, environment-aware feedback, and and environmental metadata, applying ML-based correction data traceability visualization, all of which enhance model mechanisms to account for sensor drift and environmental reliability and facilitate continuous performance evaluation. variability.

This closed-loop framework enhances sensing re- For secure and compliant AI deployment, AIMNET incorpo- liability and eliminates the frequent manual recalibration. rates two key mechanisms: application-layer encryption to Moreover, leveraging historical data and predictive modeling, protect data beyond transport-level security, and DT-Honeypot AIMNET validates readings, reconstructs missing or corrupted to proactively detect and analyze malicious behaviors.

Finally, values due to communication or sensor faults, and identifies AIMNET supports AI lifecycle management by enabling model anomalous gas patterns. These self-healing capabilities enable retraining, recalibration, and concept drift detection over time, long-term autonomous operation and precise detection of leaks ensuring long-term reliability, trustworthiness, and compliance and hazardous events, making it ideal for scalable carbon- of AI-driven functions. based gas monitoring in diverse and dynamic field settings.

Scalable Real-Time Analytics & Visualization: V. C ONCLUSION Large-scale distributed sensors generate massive data streams, creating significant hurdles for scalable, real-time We developed AIMNET , an IoT-powered Digital Twin (DT) emission detection. High data throughput, coupled with in- system designed for continuous gas emission monitoring and consistent formats and asynchronous inputs, makes sub-second early hazard detection.

AIMNET transcends the limitations of

SUBMIT TO IEEE INTERNET OF THINGS MAGAZINE 7

current monitoring approaches, which are often geographically [12] J. Chen, C. Yi, S.

D. Okegbile, J. Cai, and X.

Shen, “Networking constrained and lack intelligence, by providing a scalable, architecture and key supporting technologies for human digital twin in personalized healthcare: A comprehensive survey,” IEEE Communica- resilient, and fully integrated solution. The three-layer ar- tions Surveys & Tutorials, vol. 26, no. 1, pp. 706–746, 2023. chitecture facilitates a seamless data flow from sensors to [13] W. Dong et al., “Environmental effects on ndir-based ch4 monitoring: analytics.

Key innovations of AIMNET include the development Characterization and correction,” Environmental Science & Technology, 2025. of a low-power, self-calibrating IoT sensing instrument, a [14] F. Elghaish, S. Matarneh, M.

R. Hosseini, A. Tezel, A.-M.

Mahamadu, stable bidirectional communication framework, and an AI- and F. Taghikhah, “Predictive digital twin technologies for achieving net driven analytics core built on the digital twin world that en- zero carbon emissions: a critical review and future research agenda,” Smart and Sustainable Built Environment, 2024. ables predictive modeling and real-time decision support. Then [15] P.

M. Khiabani, G. Danala, W.

Jentner, and D. Ebert, “Challenges in data we verified AIMNET credibility based on actual deployment. integration, monitoring, and exploration of methane emissions: The role By holistically addressing the challenges, AIMNET serves as of data analysis and visualization,” in 2024 IEEE Workshop on Energy Data Visualization (EnergyVis). IEEE, 2024, pp. 1–6. a concrete DT blueprint for the next generation of smart environmental monitoring systems.

ACKNOWLEDGEMENTS

This work was supported by the DOE iM4 (Innovative Methane Measurement, Monitoring, and Mitigation) program under the contract number: DE-FE0032285, and NSF CPS- 2331105. The authors utilized AI-assisted language polishing tools to improve the readability of the manuscript. These tools were not used for generating scientific content; all conceptual development, technical content, analyses, and all scientific conclusions are entirely original.

R EFERENCES

[1] J. Hansen, M. Sato, P.

Kharecha, G. Russell, D. W.

Lea, and M. Siddall, “Climate change and trace gases,” Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, vol. 365, no. 1856, pp. 1925–1954, 2007. [2] T. A.

Jacobson, J. S. Kler, M.

T. Hernke, R. K.

Braun, K. C. Meyer, and W.

E. Funk, “Direct human health risks of increased atmospheric carbon dioxide,” Nature Sustainability, vol. 2, no. 8, pp. 691–701, 2019. [3] S. M.

Agarwal, M. Sharma, and S. Fatima, “Vocc: a database of volatile organic compounds in cancer,” RSC advances, vol. 6, no. 115, pp. 114 783–114 789, 2016. [4] R.

A. Alvarez, D. Zavala-Araiza, D.

R. Lyon, D. T.

Allen, Z. R. Barkley, A.

R. Brandt, K. J.

Davis, S. C. Herndon, D.

J. Jacob, A. Karion et al., “Assessment of methane emissions from the us oil and gas supply chain,” Science, vol. 361, no. 6398, pp. 186–188, 2018. [5] H.

Zhang, B. Li, M. Karimi, S.

Saydam, and M. Hassan, “Recent advancements in iot implementation for environmental, safety, and production monitoring in underground mines,” IEEE Internet of Things Journal, vol. 10, no. 5, pp. 4567–4580, 2023. [6] X.-M. Hu, W.

T. Honeycutt, C. Wang, B.

Weng, B. Zhou, and M. Xue, “Observation and simulation of methane plumes during the morning boundary layer transition,” Journal of Geophysical Research: Atmo- spheres, vol. 130, no. 7, p. e2024JD042317, 2025. [7] S.

Dhingra, R. B. Madda, A.

H. Gandomi, R. Patan, and M.

Danesh- mand, “Internet of things mobile–air pollution monitoring system (iot- mobair),” IEEE Internet of Things Journal, vol. 6, no. 3, pp. 5577–5584, 2019. [8] Y. Song, H. Shin, S.

Koo, S. Baek, J. Seo, H.

Kang, and Y. Kim, “Internet of maritime things platform for remote marine water quality monitoring,” IEEE Internet of Things Journal, vol. 9, no. 16, pp. 14 355–14 365, 2021. [9] U.S. Environmental Protection Agency, “40 CFR Part 58, Appendix D – Network Design Criteria for Ambient Air Quality Moni- toring,” https://www.law.cornell.edu/cfr/text/40/appendix-D to part 58, 2016, accessed: July 2, 2025. [10] H.

Xu, A. Berres, S. B.

Yoginath, H. Sorensen, P. J.

Nugent, J. Severino, S. A.

Tennille, A. Moore, W. Jones, and J.

Sanyal, “Smart mobility in the cloud: Enabling real-time situational awareness and cyber-physical control through a digital twin for traffic,” IEEE Transactions on Intelli- gent Transportation Systems, vol. 24, no. 3, pp. 3145–3156, 2023. [11] National Academy of Engineering and National Academies of Sciences, Engineering, and Medicine, Foundational research gaps and future directions for digital twins. Washington, DC: The National Academies Press, 2024.

R EFERENCES

R. Lyon, D. T.