AI‑Powered IoT Framework for Predictive Maintenance and Fault

Detection in Healthcare Devices

Varsha Negi

, R. Ravi

, Venkata Ramana Banka

, S. Jeeva

, Vikram P.

and Syed Zahidur Rashid

Department of Computer Science, Shyam Lal College Evening, Shahdara, Delhi University, Delhi 110032, India

Department of Information Technology, J. J. College of Engineering and Technology, Tiruchirappalli, Tamil Nadu, India

Department of Computer Science and Engineering (AIML), CVR College of Engineering, Hyderabad 501510, Telangana,

India

Department of Management Studies, Nandha Engineering College, Vaikkalmedu, Erode - 638052, Tamil Nadu, India

Department of CSE, New Prince Shri Bhavani College of Engineering and Technology, Chennai, Tamil Nadu, India

Department of Electronic and Telecommunication Engineering, International Islamic University Chittagong, Chittagong,

Bangladesh

Keywords: an IoT, Predictive Maintenance, Healthcare Devices, Fault Detection, Explainable AI.

Abstract: In the age of digitized healthcare, maintaining and monitoring the operational effectiveness and reliability of

biomedical devices is fundamental to patient security and clinical effectiveness. Consequently, this article

provides a Secure & Scalable AIoT Framework for Real-Time Predictive Maintenance and Ethical Fault

Detection in Healthcare Devices that combines the concepts of AI and the Internet of Things (AIoT) to realise

intelligent monitoring, fault prediction and proactive maintenance. Specifically, the introduced framework

addresses critical limitations in today's systems by integrating high-precision data verification modules, strong

inter-operability via healthcare data standards and privacy-preserving AI models in accordance with HIPPA

and GDPR regulations. Thus lightweight accurate machine learning algorithms are used for low-power,

resource-constraint IoT devices, providing scalability and efficiency when potentially operating in event

environments with real-time analytics. Also, the framework observes ethical AI procedural using explainable

AI (XAI) and bias-mitigation techniques to ensure reliance and trust in critical decision making. Through

predictive alerts and visual insights, a user-centric dashboard enables the clinical workforce to act in a timely

manner. The system's modular architecture allows adaptive deployment across various healthcare

infrastructures, providing a comprehensive solution for intelligent device management that is future-ready.

Through experimental evaluations, we provide compelling evidence of marked improvements in fault

detection accuracy, prediction latency, and data security, validating its practicality for real-world clinical use.

1 INTRODUCTION

The explosion of digital transformation in healthcare

has opened the doors to a new era where artificial

intelligence (AI) and the Internet of Things (IoT)

have begun to reshape the landscape of patient

wellness, service experience, device durability and

operational excellence. Medical devices which

include everything from ventilators to infusion pumps

to wearable monitors are central to today’s clinical

workflows, and any unexpected failure can result in

postponed treatments, higher costs or even life-

threatening situations. In such high-stakes

environments, however, conventional maintenance

strategies such as reactive and scheduled servicing are

becoming increasingly inadequate, frequently noting

indications of device deterioration too late. To

surmount these drawbacks, this paper presents a

Secure and Scalable AIoT Framework for Real-Time

Predictive Maintenance and Ethical Fault Detection

in Healthcare Devices. By utilizing actual data feeds

from Internet of Things (IoT) connected medical

devices and utilizing sophisticated AI models, it

enables anomaly detection, potential failure

forecasting, and proactive maintenance actions. In

contrast to conventional systems, we propose a

framework that embeds high-integrity data pipelines

468

Negi, V., Ravi, R., Banka, V. R., Jeeva, S., P., V. and Rashid, S. Z.

AI-Powered IoT Framework for Predictive Maintenance and Fault Detection in Healthcare Devices.

DOI: 10.5220/0013867800004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 1, pages

468-477

ISBN: 978-989-758-777-1

that have embedded validation mechanisms toward

ensuring consistent and reliable input to the learning

models. Additionally, the aircraft integrates more

than just the technical aspects; it enshrines ethical AI

principles with explanation, equity, and privacy.

Using federated learning, differential privacy, and

regulatory standards such as HIPAA and GDPR

compliance, the framework guarantees safe handling

of sensitive patient-device data. Its lightweight

architecture offers productized deployability on low-

power edge devices, enabling it to be a fit for diverse

healthcare settings from massive hospitals to mobile

clinics.

Visual analytics combined with a unified

dashboard equips clinical and technical personnel to

screen device wellness perceptively, catch actionable

insights and answer maintenance signals ahead of

time. The solution is modular and interoperable, so it

can be adapted to diverse healthcare IT ecosystems,

resulting in scalability and future-proofing of the

solution for widespread adoption.

We detail the system design, implementation, and

evaluation of the proposed framework.

Comprehensive experiments confirm its effectiveness

in terms of real-time fault detection, predictive

maintenance, high model interpretability and data

security, thus providing a robust foundation for the

next generation of intelligent medical device

management systems.

2 PROBLEM STATEMENT

As smart medical devices and IoT-enabled equipment

becomes more prevalent in the healthcare ecosystem,

this dependence presents problems in ensuring these

systems remain reliable and safe and their

availability. Conventional maintenance strategies,

including regular inspections and reactive servicing,

are becoming inadequate to the needs of modern

healthcare environments, where the consequences of

device failure can be severe, even life-threatening.

Unexpected failures in devices such as infusion

pumps, ventilators, ECG monitors or diagnostic

equipment can result in delays in treatment,

compromised patient outcomes and additional

operational costs.

While many predictive maintenance models have

been established in the literature, most current

frameworks have critical limitations, including poor

data quality resulting from sensor unreliability,

inability to process data in real-time, inefficient

integration with healthcare IT standards, and a lack of

approaches that support data privacy and ethical AI

practices. Furthermore, very few systems can

effectively scale across varied healthcare

infrastructures or run seamlessly on low-power IoT

devices.

The remaining issue is that with no explainable

AI mechanisms in place, clinicians and technicians

cannot trust or understand the predictions, which

hampers clinical decision-making. In particular, the

failure to comply with regulations and ensure safe

handling of data aggravates these problems, since

interconnected medical devices will generate and

transfer sensitive patient data.

Therefore, it is essential to develop a secure,

scalable, and ethically-grounded AI-driven IoT

framework that enables real-time fault detection and

predictive maintenance of physical systems while

maintaining data privacy, transparency, and

operational efficiency. Given that this research

intends to fill this gap, we propose a comprehensive

and domain-specific approach that fits the

characteristics of the complex needs of the healthcare

ecosystem.

3 LITERATURE REVIEW

Artificial Intelligence (AI) combined with the

Internet of Things (IoT) known as an IoT has become

a major tool for achieving predictive maintenance and

fault detection in mission-critical systems,

particularly in healthcare. With medical equipments

becoming more digitized and connected, the need for

smart maintenance frameworks is becoming more

prominent.

Pech et al. 2 Rashid et al. (2021) In smart factory

settings, Rashid et al. (2021) argue for intelligent

sensors and predictive analytics, highlighting the

ability of AI-based maintenance strategies to

minimize operational downtimes and expenses.

Although this is an industrial setting, the principles

are translatable into the healthcare world, where the

stakes that is, patient safety are so much higher.

AI-based predictive maintenance systems: Key

factors of trustworthiness (Ucar, Karakose, Kırımça,

2024) This work exposed gaps (explainability, trust,

data integrity) between existing political models and

systems developed for social media content, an issue

that this paper address through ethical AI integrations

and privacy-preserving architectures.

In their paper Sandu (Sandu (2022)) focused on

AI driven framework for the predictive Maintainence

of the Industrial IoT. His approach relied on fault

prediction based on real-time anomaly detection

algorithms but has not sufficiently focused on

AI-Powered IoT Framework for Predictive Maintenance and Fault Detection in Healthcare Devices

469

healthcare-specific needs, such as the sensitivity of

patient data, or the inter-operability of devices, topics

that are comprehensively addressed in this study.

Khalid (2024) explored the use of AI-enabled

digital transformation in the sustainable operations

arena. His focus on scalability and energy-efficient

design resonates with the lightweight model design of

this paper that allows deployment on resource-

constrained medical IoT devices.

He et al. (2019) studied how AI technologies can

be implemented in real-world medicine, identifying

regulatory, technical and ethical challenges

preventing AI uptake. Informed by this

understanding, the present work incorporates

privacy techniques that are compliant with

GDPR/HIPAA and explainable AI tools to engender

trust in clinicians.

Dhameliya and Patel presented the predictive

maintenance of general IoT systems using the

machine learning models. However, their solution

failed to tackle challenges such as interoperability

with Electronic Health Record (EHR) systems or

deployment in edge environments gaps that this

research intends to fill.

Shajahan and Ramesh: An IoT health monitoring

system with predictive analysis (2019). This was

effective for real-time monitoring; however, it did not

provide for scalability or ethical AI features, again

highlighting the need for a more holistic framework

as presented in this paper.

Pesapane et al. (2018) analyzed regulatory

questions for using AI as a medical device,

highlighting ethical dangers and legal ambiguities.

However, our proposed framework comes to

overcome them by embedding regulatory compliance

in the design process and guaranteeing explainable

decisions.

According to Pessin (2025), the role of connected

devices in predictive diagnostics is expanding. His

piece describes how these networked medical

devices help reduce risk to patients through

monitoring in real time. This paper extends that work

to propose a secure system for large-scale such

monitoring.

Davies (2025) studied the race for hospitals to go

‘smart’ with AI and IoT. He cited concern, however,

around scalability and standardization — which this

framework explicitly addresses through modular,

standards-compliant architecture.

Focusing on Intelligent Transportation Systems,

Iyer (2021) provided some key insights in edge

deployment and resource aware AI design, which are

also applied by this framework to achieve efficient

semantic processing over medical devices.

From an industry perspective, Goja (2022)

presented the AIoT architecture design and

emphasized adaptability and interoperability. Our

work realizes this vision in the healthcare vertical

characterized as an end to end view across devices,

and instant insights that span diverse grid

infrastructure across a hospital.

Huber-Straßer et al. (2018) imagined the future

value chains of robotics and AI. Albeit speculative,

their focus on ethical frameworks and human-

machine trust dovetails nicely with this research’s

aim of integrating fairness and interpretability into

healthcare AI.

As Ghosh (2020) discussed, when AI and IoT

integrate we get AIoT, but true real-time intelligence

at the edge is needed. The proposed research widens

the basis of that application by using a mendacious

AIoT in high subsidiarity clinical settings with

protection privacy and decision transparency.

Lin et al. (2019) showcased the AIoT

applications by smart agriculture that indirectly verify

the cross-domain applicability of AIoT models. This

data-driven approach led us to develop a multi-tier

data validation mechanism with a focus on high

prediction accuracy in this paper.

Chu et al. (2019) reviewed AIoT in sports science,

demonstrating that AIoT has the ability to monitor

performance metrics instantaneously. We extend

these principles in this work to monitor medical

device status and failure trends within clinical

settings.

Anumandla (2018) introduced a simplified

predictive maintenance approach based on IoT and

machine learning. Yet it did not possess a privacy

model and explainability framework, which this

paper combines to address healthcare applications.

Singh (2025) provided an overview of AI-driven

sensors for predictive maintenance, including

context-aware sensing. This notion of sensor fusion is

the strategy we adopted in our system to extract

operational and environment data of devices.

Cheng et al. In line with this concept, Dey et al.

(2019) proposed an edge based AIoT system that

supports real-time analytics and this architectural

decision was also following in this work to reduce

latency, and dependability on centralized servers.

Lin et al. It presented a new perspective that

describes how low-latency decision-making inside

AIoT improves system responsiveness (2019). Their

model motivated our system’s use of lightweight and

edge-optimized models for real-time error

notifications.

Pessin (2025) and Davies (2025) addressed

industry level phenomena regarding predictive

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

470

diagnostics and reiterated the need for fault resilient

critical systems, which we provide in our framework

through integrated predictive alerts and maintenance

scheduling.

Khalid (2024) and Ucar et al. (2024) singled out a

lack of explainability in traditional models. This

study directly responds to that gap with Explainable

AI (XAI) tools like SHAP, which provide insights

into model predictions in an interpretable manner.

Dhameliya & Patel (2020) and Shajahan &

Ramesh (2019) considered general IoT environments

but did not provide specialization for healthcare.

Paper extends their foundation into a specialized,

regulation compliant, healthcare framework.

4 METHODOLOGY

This section presents the methodology adopted for the

development of a Secure and Scalable AIoT

Framework for predictive maintenance and fault

detection in healthcare devices. The approach

combines AI, IoT, edge computing, and explainable

AI (XAI) within a modular architecture that ensures

real-time data processing, device interoperability, and

adherence to privacy regulations.

4.1 Framework Design

The framework is built around the integration of IoT-

enabled healthcare devices with AI-powered models

for fault detection and maintenance prediction. Key

components include:

The system integrates with various healthcare

devices, such as ECG monitors, infusion pumps, and

wearable sensors, each transmitting real-time

operational data via IoT protocols like MQTT and

HTTP. Device data includes operational parameters

(e.g., temperature, pressure) and environmental

factors (e.g., humidity, battery status).

The system follows a modular architecture where

distinct components (data acquisition, preprocessing,

predictive model, alerting system) interact through

APIs. This enables easy integration with different

medical devices and seamless updates without

affecting the core functionalities.

To reduce latency and bandwidth dependency, a

significant portion of data processing is carried out on

edge devices. Lightweight AI models are deployed on

local hardware (e.g., Raspberry Pi, medical-grade

edge servers) for real-time fault detection and

maintenance predictions, ensuring low-latency

decision-making.

The system employs XAI techniques such as SHAP

(SHapley Additive exPlanations) to make the AI

models transparent and interpretable for healthcare

practitioners. This allows clinicians to trust the

predictions and understand why certain maintenance

actions are recommended. Table 1 shows the system

parameters.

Table 1: System Parameters.

Parameter

Details

Data Sources

IoT-enabled healthcare

devices (e.g., ECG monitors,

infusion pumps)

Data

Preprocessin

g Techniques

Noise Filtering, Missing

Value Imputation, Outlier

Removal

Feature

Extraction

Time-domain (mean,

variance), Frequency-

domain (Fourier Transform)

Machine

Learning

Models

Random Forest, Gradient

Boosting, LSTM

Edge

Computing

Deployment

Raspberry Pi, Medical-grade

Edge Servers

Security

Measures

Federated Learning,

SSL/TLS Encryption

Explainabilit

y Techniques

SHAP (SHapley Additive

exPlanations)

4.2 Data Preprocessing and Feature

Engineering

4.2.1 Data Cleansing

Raw sensor data collected from healthcare devices is

subject to preprocessing steps, including noise

filtering, missing value imputation, and outlier

removal. The goal is to prepare the data for reliable

AI model training and real-time prediction.

Feature Extraction

Relevant features are extracted from the sensor data,

such as:

Time-domain features (e.g., mean, variance,

skewness)

Frequency-domain features (e.g., Fourier

Transform)

Statistical features (e.g., moving average,

standard deviation)

4.2.2 Data Normalization

All features are scaled using Min-Max scaling to

ensure that the AI models can process data efficiently

AI-Powered IoT Framework for Predictive Maintenance and Fault Detection in Healthcare Devices

471

and converge quickly during training. The framework

uses Apache Kafka and Apache Flink for streaming

data processing, enabling real-time analysis and

immediate prediction of potential faults. Predictive

maintenance signals are generated continuously,

based on real-time sensor data. Table 2 shows the data

preprocessing techniques.

Table 2: Data Preprocessing Techniques.

Preprocessing

Step

Description

Noise

Filtering

Removal of signal noise

using median filtering and

smoothing techniques.

Missing

Value

Imputation

Replacement of missing

values using mean

imputation or

interpolation.

Outlier

Removal

Detection and removal of

outliers using Z-score or

IQR method.

Normalizatio

Min-Max scaling to

standardize feature values

between 0 and 1.

4.3 Predictive Maintenance Models

A combination of machine learning and deep learning

models is employed for fault detection:

Random Forest and Gradient Boosting Machines

(GBM) are used for supervised learning, leveraging

labeled maintenance data to predict potential failures.

Recurrent Neural Networks (RNNs), particularly

Long Short-Term Memory (LSTM) networks, are

used to model time-series data and predict future

device health status.

The models are trained using historical device failure

data and operational logs. The training process

involves:

Splitting the data into training, validation, and test

sets.

Hyperparameter tuning using grid search or random

search to optimize model performance.

Cross-validation to assess model robustness and

avoid overfitting.

Models are evaluated based on:

Accuracy, Precision, Recall, and F1-score for

classification tasks (e.g., detecting device failure).

Mean Absolute Error (MAE) for regression tasks

(e.g., predicting device remaining life).

AUC-ROC curve to measure model performance in

imbalanced datasets.

4.4 Privacy and Security Measures

To address privacy concerns, the framework

incorporates federated learning, where model training

occurs locally on edge devices without transferring

sensitive patient data to centralized servers. Only

model updates are shared, ensuring compliance with

HIPAA and GDPR.All communication between IoT

devices and the cloud server is encrypted using

SSL/TLS protocols. Device data is stored in

encrypted databases, ensuring that sensitive patient

information remains secure. The framework enforces

strong access control policies, ensuring that only

authorized personnel (clinicians, technicians) can

access predictive maintenance results and device

diagnostics. Table 3 shows the privacy and security

measures.

Table 3: Privacy and Security Measures.

Security

Measure

Description

Federated

Learning

Ensures data never leaves

edge devices; only model

updates are shared.

Data

Encryption

Use of SSL/TLS for secure

data transmission between

devices and servers.

Access

Control

Role-based access control to

limit system access to

authorized personnel.

Regulatory

Compliance

Compliance with HIPAA and

GDPR for data privacy and

security.

The framework is deployed in a simulated

hospital environment, where IoT-enabled medical

devices are integrated, and real-time performance

metrics are monitored.

4.5 Performance Metrics

System performance is assessed based on:

Fault detection accuracy (measured by precision,

recall, and F1-score).

Prediction latency (time taken for fault detection from

sensor data input to alert).

Scalability (ability to handle multiple devices and

large-scale data inputs).

Data security (measured by the number of successful

penetration tests and compliance audits).

Healthcare professionals involved in the deployment

provide feedback on the usability of the system,

focusing on the dashboard interface and the

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

472

interpretability of AI-generated insights. Table 4

shows the experimental setup.

Table 4: Experimental Setup.

Parameter

Details

Dataset

Healthcare device failure logs

(e.g., ECG, infusion pumps)

Number of

Devices

50 devices (including ECG

monitors, infusion pumps,

diagnostic tools)

Edge

Device

Configurati

Raspberry Pi 4 with 4GB

RAM, running Python,

TensorFlow

Cloud

Service

Configurati

AWS EC2 for model training

and aggregation

Evaluation

Metrics

Accuracy, Precision, Recall,

F1-Score, AUC-ROC

Training

Time

3 hours for 100 epochs on 10

devices

4.6 Ethical Considerations

The research adheres to ethical AI principles by

ensuring that all machine learning models are

explainable and transparent.

Figure 1: Training and Validation Loss Comparison Over

Epochs.

Additionally, privacy-preserving techniques such

as federated learning and differential privacy are

employed to safeguard patient data. The system

design ensures that all predictive maintenance

recommendations are non-intrusive and clinician-

approved, ensuring that AI does not replace medical

decision-making but rather supports informed

actions. Figure 1 and 2 shows the training and

validation loss comparison over epochs and IoT

Framework for Predictive Maintenance and Fault

Detection in Healthcare Devices.

Figure 2: Iot Framework for Predictive Maintenance and

Fault Detection in Healthcare Devices.

5 RESULTS AND DISCUSSION

This section provides the findings of the experiments

conducted to evaluate the Secure and Scalable an IoT

Framework for Real-Time Predictive Maintenance

and Ethical Fault Detection in Healthcare Devices.

The framework’s performance is evaluated in the

aspects of fault detection accuracy, predictive

maintenance prediction, real-time processing

capability, and system scalability. Furthermore, this

discussion expands on the significance of these

findings, how they can relate to healthcare

environments, and the benefit of applying the

proposed framework to real life.

5.1 Fault Detection Accuracy

The main goal behind the framework is to predict

device failure precisely and start preventive

maintenance. Appendix A summarizes the dataset

used to train and evaluate the models, which

comprised historical failure data from various

healthcare devices, including ECG monitors and

infusion pumps. The Random Forest and GBM

models had precision: 91%, recall: 88%, and F1-

score: 89%; the LSTM network had precision: 93%,

AI-Powered IoT Framework for Predictive Maintenance and Fault Detection in Healthcare Devices

473

and recall: 90%. The proposed AIoT framework is

effective in realizing the real-time detection of faults,

which can be applied for predictive maintenance in a

healthcare environment. The proposed machine

learning–based approach greatly improves the

system's ability to detect anomalies, which can be

very subtle and often go unobserved due to the

thresholds used in rule-based maintenance systems.

The LSTM model outperforms other models because

time-series analysis is crucial in fault detection, as

medical device faults therefore occur gradually over

time than immediately.

5.2 Prediction of Predictive

Maintenance

The predictive maintenance capability of the

framework is evaluated by considering the time

between the first anomaly detection (by the

framework) and the moment the system predicts that

the maintenance action should be taken. Predictive

maintenance using LSTM predictions could be used

to anticipate device malfunctions with a lead-time of

up to 72 hours in advance of failure, to schedule

maintenance activity without disrupting critical care

operations. On the other hand, the conventional

maintenance models generally require short notice,

and they result in either unplanned downtimes or

excessive repairs.

The efficacy of the framework is further validated

with a Mean Absolute Error (MAE) of 2.1 hours

achieved successfully in maintenance interval

predictions, which remains well within operational

tolerance limits. By offering this early diagnosis,

healthcare facilities can minimize unscheduled

downtime and maximize the utilization of their assets,

leading to improved health outcomes for patients.

5.3 Real-Time Processing and Latency

Healthcare systems need to process real-time data

because of its immediate effect on decision making

that can also be lifesaving. We proposed a very fast

fault detection system. With edge computing now

implemented, the system demonstrated 0.3 seconds of

processing latency at its best from data input to fault

detection, which guarantees alerts are generated in

time to prevent catastrophic device failures.

By avoiding the network traffic and further

remote data processing time found in cloud systems,

the edge deployment strategy provides a drastic

increase in responsiveness. Clinicians receive real-

time alerts, allowing them to intervene quickly

without having to wait for the information to be

processed on a centralized server. This low-latency

design is essential in critical care environments

where time-critical decisions impact patient safety.

Table 5: Model Evaluation Metrics.

Model

Precision

Recall

F1-Score

Accuracy

AUC-ROC

Random

Forest

0.91

0.88

0.89

0.92

0.94

Gradient

Boosting

Machine

0.89

0.85

0.87

0.90

0.92

Long Short-

Term Memory

(LSTM)

0.93

0.90

0.91

0.94

0.95

5.4 Scalability and Performance of the

System

To test the scalability of the system, we simulate the

integration of multiple devices (e.g., ventilators,

infusion pumps, diagnostic tools, etc.). It was been

consistently performed well regardless number of

connected devices. Fault detection accuracy and

prediction latency continued to hold steady as we

scaled from 10 to 50 devices, which shows that this

method can operate in large scale healthcare

environments without the loss of performance.

Framework that is flexible in its architecture,

allowing for changes, but is also modular enabling

the easier additions of devices to the network. This is

particularly useful for hospitals and health care

providers who have large inventories of IoT-enabled

devices and it allows the framework to be

implemented without requiring an overhaul of the

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

474

existing infrastructure. Table 5 shows the model

evaluation metrics.

5.5 Ethical Considerations and Data

Privacy

One of the major advantages of the proposed

framework is its privacy-preserving design, achieved

through federated learning. By keeping sensitive

patient data on local edge devices and sharing only

model updates, the system ensures that personal

health information is never exposed to unauthorized

access. This method complies with major data

protection regulations, including HIPAA and GDPR,

safeguarding patient privacy.

In addition, the Explainable AI (XAI) integration

ensures that the system’s predictions are transparent

and interpretable. Healthcare professionals can view

the rationale behind each prediction, helping them

make informed decisions about device maintenance.

This transparency is critical for building trust in AI

systems within the healthcare sector, where data

integrity and accountability are paramount.

5.6 Discussion

The results highlight the significant potential of AI-

powered IoT frameworks for transforming healthcare

maintenance strategies. By moving beyond

traditional reactive maintenance methods, this system

enables proactive device management, which can

reduce unplanned downtime and enhance patient

care. The combination of machine learning and edge

computing ensures high accuracy in fault detection

while maintaining the responsiveness required in

real-time healthcare settings.

The predictive maintenance capabilities of the

system ensure that healthcare devices are maintained

before failures occur, improving their reliability and

lifespan. This is particularly important in critical

healthcare environments, where device failure can

have dire consequences for patient safety.

Furthermore, the system’s scalability makes it

suitable for hospitals of all sizes, from small clinics to

large healthcare networks.

The ethical design of the framework, with its

focus on data privacy, transparency, and regulatory

compliance, ensures that the system can be safely and

responsibly deployed in real-world healthcare

settings. Federated learning enables the system to

operate without compromising patient privacy, which

is a significant concern in today’s digital health

landscape.

However, there are limitations. The system’s

dependence on high-quality data means that it may

struggle in environments where sensor data is noisy

or incomplete. Additionally, while the system

performs well in controlled environments, future

work should focus on real-world testing in a variety

of healthcare settings to fully assess its robustness

under diverse conditions.

5.7 Summary

In conclusion, the proposed An IoT framework offers

significant improvements over traditional

maintenance models by providing real-time

predictive maintenance, enhancing device reliability,

and ensuring patient data privacy. The results

demonstrate that the system performs efficiently and

effectively, with high accuracy in fault detection and

low-latency predictions. With its scalability and

modular design, the framework offers a future-proof

solution for healthcare providers, allowing them to

integrate it seamlessly into existing IoT

infrastructures while maintaining compliance with

ethical and privacy standards.

6 CONCLUSIONS

Describing this research, they say: "We propose a

Secure and Scalable AIoT Framework for Real-Time

Predictive Maintenance and Ethical Fault Detection

in Healthcare Devices. In this work, we propose a

framework that incorporates Artificial Intelligence

(AI) and Internet of Things (IoT) technologies to

address proactive device management, fault

detection, and predictive maintenance in the context

of the healthcare environment. Utilizing machine

learning models like Random Forest, Gradient

Boosting machines (GBM), and LSTM networks,

the system accurately forecasts device failures ahead

of time, leading to enhanced operational efficiency

and a reduction in unpredictable downtime in critical

healthcare environments.

The framework utilizes edge computing with

local data processing for making predictive

maintenance decisions in real-time with low latency.

Not only that, but this improves the responsiveness

of the system, too, while reducing the dependence on

cloud-based infrastructure — making mobile health

(mHealth) suitable for a range of healthcare

environments. Explainable AI (XAI) instills

transparency and trust by ensuring that healthcare

professionals can understand the system’s predictions

and act accordingly. Moreover, implementing

AI-Powered IoT Framework for Predictive Maintenance and Fault Detection in Healthcare Devices

475

federated learning with strong data encryption

mechanisms allows patient data to be processed

securely, adhering to HIPAA and GDPR regulations

while fostering privacy and security for the patients.

The system's accuracy in detecting faults, its li

optimal lead time in predicting maintenance needs,

and its scalability for implementation at any scale

were all validated through extensive testing, making

it an ideal tool for healthcare organizations.

Moreover, its modular structure enables smooth

incorporation into current health care frameworks,

where upgrades wouldn't necessitate major updates,

leading to long-term sustainability.

Yet despite its benefits, the framework depends

on high-quality, consistent data. Future work could

explore the system's performance in environments

with noisy and/or incomplete sensor data. In addition,

testing in real-world conditions within multiple

healthcare systems is necessary to prove this

robustness and for its application to the clinical

setting.

Reflecting upon the AIoT framework, this

research promises transitively; into the healthcare

space to improve predictive maintenance, promising

higher device reliability, scalable clinical

performance, and ultimately optimizes operational

costs while improving the healthcare experience

overall. This framework combines advanced AI

techniques, ethical considerations, and real-time

processing capabilities that make it a useful asset in a

changing digital healthcare environment.

REFERENCES

Anumandla, V. (2018). Predictive Maintenance using IoT

and Machine Learning. International Journal of

Engineering Research & Technology (IJERT), 7(6), 1-

6. Retrieved from https://www.i jert.org/research

/predictive-maintenance-using-iot-and-machine-

learning-IJERTV7IS060001.pdf

Cheng, C. C., Shih, C., Chou, W. Y., Ahamed, S. I., &

Hsiung, P. A. (2019). Artificial Intelligence of Things

in Sports Science: Weight Training as an

Example. Computer, 52(11), 38-47. https:// doi.org

/10.1109/MC.2019.2905891

Chu, W. C., Shih, C., Chou, W. Y., Ahamed, S. I., &

Hsiung, P. A. (2019). Artificial Intelligence of Things

in Sports Science: Weight Training as an Example.

Computer, 52(11), 38-47. https://doi.org/10.1109 /M

C.2019.2905891

Davies, M. (2025). Medical centres compete to achieve

'smart hospital' status. Financial Times.

Retrieved from https://www.ft.com/content/2805edfd-

36db-4a58-b93f-411a18c6e003

Dhameliya, D., & Patel, D. (2020). Predictive Maintenance

using Machine Learning. International Journal of

Engineering Research & Technology (IJERT), 9(7),

1046-1050. Retrieved from

https://www.ijert.org/research/predictive- maintenance

-using-machine-learning- IJERTV9IS070546.pdf

Ghosh, I. (2020). AIoT: When Artificial Intelligence Meets

the Internet of Things. Visual Capitalist.

Retrieved from https://www.visualcapitalist.com/aiot-

when-ai-meets-iot/

Goja, A. (2022). The Architect's Guide to the AIoT.

Cisco. Retrieved from https://www.cisco.com/c/en/us/

solutions/internet-of-things/aiot-architect-guide.html

He, J., Baxter, S. L., Xu, J., Xu, J., & Zhou, X. (2019). The

practical implementation of artificial intelligence

technologies in medicine. Nature Medicine, 25(1), 30-

36. https://doi.org/10.1038/s41591-018-0307-0

Huber-Straßer, A., Schüller, M., Müller, N., von der Gracht,

H., Lichtenau, P., & Zühlke, H. M. (2018). Rethinking

the value chain. A study on AI, humanoids and robots -

Artificial Intelligence: Possible business application

and development scenarios to 2040. KPMG. Retrieved

from

https://home.kpmg/xx/en/home/insights/2018/09/rethi

nking-the-value-chain.html

Iyer, L. S. (2021). AI enabled applications towards

intelligent transportation. Transportation

Engineering, 5, 100083. https://doi.org/10.1016/j.tren

g.2021.100083

Khalid, M. (2024). Energy 4.0: AI-enabled digital

transformation for sustainable power networks.

Computers & Industrial Engineering, 193, 110253.

https://doi.org/10.1016/j.cie.2024.110253

Lin, Y. J., Chuang, C. W., Yen, C. Y., Huang, S. H., &

Huang, P. W. (2019). AIoT Applications in Smart

Agriculture: A Case Study on Cabbage. 2019 IEEE

International Conference on Artificial Intelligence

Circuits and Systems (AICAS), 193-196.

https://doi.org/10.1109/AICAS.2019.8771536

Pech, M., Vrchota, J., & Bednář, J. (2021). Predictive

Maintenance and Intelligent Sensors in Smart

Factory: Review. Sensors, 21(4), 1470. https://doi.or

g/10.3390/s21041470

Pesapane, F., Volonté, C., Codari, M., & Sardanelli, F.

(2018). Artificial intelligence as a medical device in

radiology: ethical and regulatory issues in Europe and

the United States. Insights into Imaging, 9(5), 745-753.

https://doi.org/10.1007/s13244-018-0645-y

Pessin, G. (2025). Networked devices help head off medical

woes and speed recovery. Financial Times.

Retrieved from https://www.ft.com/content/74badf1b-

6876-4146-a6f8-34e610a61b7d

Sandu, A. K. (2022). AI-Powered Predictive Maintenance

for Industrial IoT Systems. Digitalization &

Sustainability Review, 2(1), 1-14.

Retrieved from https://www.researchgate.net/publicati

on/380969569_AI-

Powered_Predictive_Maintenance_for_Industrial_IoT

_Systems

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

476

Shajahan, T., & Ramesh, M. V. (2019). IoT based health

monitoring system with predictive analysis. 2019

International Conference on Communication and

Signal Processing (ICCSP), 1409-1413.

https://doi.org/10.110 9/ICCSP. 2019.8697908

Ucar, A., Karakose, M., & Kırımça, N. (2024). Artificial

Intelligence for Predictive Maintenance Applications:

Key Components, Trustworthiness, and Future Trends.

Applied Sciences, 14(2), 898.

https://doi.org/10.3390/app14020898

AI-Powered IoT Framework for Predictive Maintenance and Fault Detection in Healthcare Devices

477