Maintenance 4.0 for HVAC Systems: Addressing Implementation Challenges and Research Gaps

Highlights

What are the main findings?

• There is a lack of research on integrating Industry 4.0 into ventilation system maintenance planning.
• Existing reviews cover sensors, AI/ML, and big data in HVAC, but not integrated Maintenance 4.0.

What are the implications of the main findings?

• Effective ventilation is crucial for indoor air quality, health, and energy efficiency.
• AI-driven analytics enable proactive, data-driven maintenance for optimized schedules, failure prediction, and improved performance.

Abstract

This article explores the integration of Maintenance 4.0 into HVAC (heating, ventilation, and air conditioning) systems, highlighting its essential role within the framework of Industry 4.0. Maintenance 4.0 utilizes advanced technologies such as artificial intelligence and IoT sensing technologies. It also incorporates sophisticated data management techniques to transform maintenance strategies into HVAC and indoor ventilation systems. These innovations work together to enhance energy efficiency, air quality, and overall system performance. The paper provides an overview of various Maintenance 4.0 frameworks, discussing the role of IoT sensors in real-time monitoring of environmental conditions, equipment health, and energy consumption. It highlights how AI-driven analytics, supported by IoT data, enable predictive maintenance and fault detection. Additionally, the paper identifies key research gaps and challenges that hinder the widespread implementation of Maintenance 4.0, including issues related to data quality, model interpretability, system integration, and scalability. This paper also proposes solutions to address these challenges, such as advanced data management techniques, explainable AI models, robust system integration strategies, and user-centered design approaches. By addressing these research gaps, this paper aims to accelerate the adoption of Maintenance 4.0 in HVAC systems, contributing to more sustainable, efficient, and intelligent built environments.

1. Introduction

Ventilation systems in public and private sectors have a great role in controlling pollution levels, enhancing personal comfort, and increasing productivity and illness prevention. Quoting the United Nations Children’s Fund (UNICEF) statement, “Outdoor and indoor air pollution are directly linked with respiratory infections and other diseases that account for 15% of all under-five deaths” [1]. Additionally, inefficient ventilation systems lead to poor air quality. This has been linked to health problems in schools, for both students and teachers [2], such as headaches, nausea, and respiratory irritation. Indoor air quality (IAQ) management is a critical aspect of maintaining healthy and comfortable indoor environments, particularly in the context of modern ventilation systems. In this respect, maintaining the structural performance of ventilation systems, especially in public sectors, is crucial to attain an optimal work environment.

The traditional method to sustain ventilation systems often lacks proactive action due to the absence of real-time data. This can lead to the delayed identification of system failures and reduced performance, impacting not only the ventilation system but also personnel. For example, the New York City Department of Education encountered difficulties in effectively tracking air quality in its numerous school buildings, sparking concerns about the potential impact on student health and academic performance [3]. To address these drawbacks, scholars and practitioners integrate sensing elements and advanced technologies, such as artificial intelligence (AI), machine learning (ML) and the Internet of Things (IoT), to unlock new possibilities for monitoring and improving the performance of ventilation systems which can be added under the implication of Industry 4.0. By leveraging data-driven models and real-time monitoring capabilities, these technologies offer innovative solutions for predicting, diagnosing, and prescribing maintenance measures to improve IAQ.

This article provides the status quo of the existing solutions suggested by the implementation of the Industry 4.0 model to monitor, control, and maintain the ventilation systems. Further, it explores the impact of maintenance strategies, pollutant categories, and opportunities presented by IoT platforms. Through a comprehensive analysis of existing literature and studies, it aims to highlight the potential of the implementation of the Industry 4.0 model in revolutionizing IAQ management and optimizing ventilation system performance. The optimization of these systems entails real-time decision-making. These critical decisions include maintenance decisions, which are crucial for avoiding sudden failures or degraded performance of the ventilation systems. The possibility for the Industry 4.0 model to make suitable maintenance decisions has been defined as modern Maintenance 4.0 (M4.0) [4,5].

The integration of Industry 4.0 technologies—such as real-time data collection through sensing elements, advanced data management systems, analytics, cloud computing, and AI for predictive maintenance—has transformed various sectors. Despite this advancement, a significant research gap exists in the specific application of these technologies to the planning and maintenance of ventilation systems.

To date, ten specified reviews (see Table 1) have addressed various components of Industry 4.0 within the context of ventilation systems, categorized into three main areas:

• Sensing elements: Reviews focusing on sensor applications for monitoring IAQ and environmental parameters [5,6,7].
• AI and machine learning: Studies compiling ML and AI models for fault detection in heating, ventilation, and air conditioning (HVAC) systems [6,7,8].
• Big data and predictive maintenance: Reviews examining big data analytics and ML techniques aimed at improving predictive maintenance in ventilation systems [9,10,11,12].

While these reviews highlight important advancements, there is a notable absence of focused research on the concept of Maintenance 4.0—specifically, how Industry 4.0 technologies can be effectively integrated into the maintenance planning of ventilation systems.

This research aims to fill this gap by providing a comprehensive review that not only synthesizes existing literature on the application of Industry 4.0 tools but also explores their potential integration into maintenance strategies for ventilation systems. By addressing this underexplored area, this article aims to enhance the understanding and implementation of Maintenance 4.0, ultimately contributing to more efficient and effective ventilation system management.

Table 1. Summary of existing reviews.

Ref.	No. of Review Articles	Covered Period	Main Objective
[5]	150	1999–2022	Review of the development and application of IoT-based IAQ monitoring platforms.
[13]	34	2013–2023	Comprehensive evaluation of low-cost air pollution sensing technologies compared to high-quality instruments for IAQ monitoring. The review finds that while low-cost sensors show moderate correlations with reference instruments and are adequate for qualitative analysis, high-quality instruments are still crucial for accurate measurements.
[14]	68	1999–2022	Review outlining the relationship between hospital IAQ (IAQ) and factors such as building design, operation, and occupant behavior.
[6]	211	1996–2023	The paper provides a comprehensive review of current computing-based fault detection and diagnosis (FDD) methods for HVAC systems, classifying them into knowledge-based and data-driven approaches. It identifies key topics such as data availability, quality, approach generality, capability, interpretability, and modeling efforts. The review discusses the current state of FDD, highlights challenges like dealing with complex fault situations, improving model fidelity, and handling multiple faults, and suggests future research directions to enhance FDD methods’ applicability and effectiveness in real buildings.
[7]	279	2002–2022	Review of AI-based FDD methods for HVAC systems.
[8]	400	2005–2021	Review of the use of AI big data analytics in building automation and management systems (bamss).
[9]	161	1993–2024	Review of predictive maintenance from a data mining perspective. It highlights the importance of optimizing maintenance timing and type to maximize system availability and minimize resource usage.
[10]	189	1999–2021	Review and practical guidance on applying machine learning for air quality mapping.
[11]	55	2002–2021	Review of the existing algorithms used for predictive maintenance in HVAC systems.
[12]	112	2007–2022	It highlights how deep learning can optimize building performance, reduce energy consumption, improve predictive maintenance, and monitor equipment. The paper also discusses challenges, such as the need for public datasets, and emphasizes the importance of deep learning for predictive maintenance of thermal storage air-conditioning (TS-AC) systems to enhance sustainability and cost-efficiency.

Table 1. Summary of existing reviews.

Ref.	No. of Review Articles	Covered Period	Main Objective
[5]	150	1999–2022	Review of the development and application of IoT-based IAQ monitoring platforms.
[13]	34	2013–2023	Comprehensive evaluation of low-cost air pollution sensing technologies compared to high-quality instruments for IAQ monitoring. The review finds that while low-cost sensors show moderate correlations with reference instruments and are adequate for qualitative analysis, high-quality instruments are still crucial for accurate measurements.
[14]	68	1999–2022	Review outlining the relationship between hospital IAQ (IAQ) and factors such as building design, operation, and occupant behavior.
[6]	211	1996–2023	The paper provides a comprehensive review of current computing-based fault detection and diagnosis (FDD) methods for HVAC systems, classifying them into knowledge-based and data-driven approaches. It identifies key topics such as data availability, quality, approach generality, capability, interpretability, and modeling efforts. The review discusses the current state of FDD, highlights challenges like dealing with complex fault situations, improving model fidelity, and handling multiple faults, and suggests future research directions to enhance FDD methods’ applicability and effectiveness in real buildings.
[7]	279	2002–2022	Review of AI-based FDD methods for HVAC systems.
[8]	400	2005–2021	Review of the use of AI big data analytics in building automation and management systems (bamss).
[9]	161	1993–2024	Review of predictive maintenance from a data mining perspective. It highlights the importance of optimizing maintenance timing and type to maximize system availability and minimize resource usage.
[10]	189	1999–2021	Review and practical guidance on applying machine learning for air quality mapping.
[11]	55	2002–2021	Review of the existing algorithms used for predictive maintenance in HVAC systems.
[12]	112	2007–2022	It highlights how deep learning can optimize building performance, reduce energy consumption, improve predictive maintenance, and monitor equipment. The paper also discusses challenges, such as the need for public datasets, and emphasizes the importance of deep learning for predictive maintenance of thermal storage air-conditioning (TS-AC) systems to enhance sustainability and cost-efficiency.

This paper is organized as follows: the next section explains the implementation of Maintenance 4.0 in ventilation systems, and the way it is implemented in the industry. Section 4 presents sensing elements to collect online data. This includes the different connection methods between these sensors and the cloud. Section 5 explains the management process for the data when it is received in the cloud. Finally, the AI techniques and algorithms that are used to predict data are described.

2. Research Methodology

To achieve the objective of this research, a systematic research methodology was conducted. This methodology included data collection for the related literature documents. Then, these documents were critically analyzed, key themes and knowledge gaps were identified, and a novel theoretical framework was developed to address these gaps and advance the field.

The necessary data for this paper were obtainable through the following steps. First, an initial search was conducted on the Scopus database using the following strings:

(“(air AND quality) OR (CO₂ AND emissions) OR (air AND ventilation)” AND “(artificial AND intelligence) OR (neural AND network) OR (machine AND learning) OR (deep AND learning)” AND “(maintenance) OR (industry 4.0) OR (maintenance 4.0)”).

Second, inclusion and exclusion criteria were considered to filter the data gathered from the search process (see Figure 1). All documents published before 2014—covering a ten-year range—were excluded. The remaining documents were then further filtered based on document type and language. Only reviews and articles were included, and any language other than English was excluded. Third, the screening process was performed to eliminate the irrelevant documents. This was followed by the forward snowballing stage to ensure that all relevant documents were included in the research library. Forward snowballing involves reviewing the reference list in every individual document in the research library to extract relevant papers not available in the library. Finally, if there are any duplicates, they should be removed. The resulting documents have been analyzed to study the trajectory of the research in Maintenance 4.0 for ventilation systems, and the top countries working in this area.

2.1. Current Research Trajectory

In a statistically based overview, the research on the implementation of an Industry 4.0 configuration in maintenance planning for ventilation systems has been slowly growing in the last ten years. For more details, the number of articles published on this topic has grown from a single article in 2014 to 34 in 2023, with an average of fewer than 5 articles per year, see (Figure 2). Over the past decade, China and the United States have led research on applying Industry 4.0 to ventilation system maintenance planning. Together, these countries have published a total of 56 research documents. India, South Korea, and Canada have also made significant contributions in this field, as illustrated in (Figure 3).

2.2. Maintenance vs. Industry Revolutions

The evolution of maintenance processes, strategies, and planning is necessarily connected to the industry revolution phases, beginning with Industry 1.0 and moving through to Industry 4.0. However, Nagpal et al. [15] has philosophically considered the evolution of maintenance processes related to different maintenance strategies, assuming that Maintenance 4.0 is an evolution of Maintenance 1.0, which is considered to be after the failure of the machine, then Maintenance 2.0, which is equal to the preventive maintenance which uses time data to define the necessary periods to maintain machines or systems. During that period, the operators and practitioners faced several problems because of unforeseen ventilation failures. In this respect, it was necessary to predict failures, which opened the great research stream into predictive maintenance (Maintenance 3.0). Progress continued until Maintenance 4.0 arrived, which is proactive maintenance for fault elimination. However, such an assumption lacks reality and may be inconsistent with the chronological change in the four-industry revolution. Based on the literature and historical evidence in this article, the authors agree with the ideology that industry evolution and maintenance evolution have grown from the same tree. Figure 4 shows that, eventually, maintenance is an action to sustain the structural integrity of industrial entities, so industry development entails advancement in maintenance procedures and strategies, beginning from a run-to-failure strategy to reaching the configuration of Industry 4.0 (i.e., Maintenance 4.0).

Maintenance 4.0 and Predictive Maintenance (M3.0) are transforming IAQ management by leveraging IoT, data analytics, and ML to optimize ventilation system performance. Maintenance 4.0 enables real-time monitoring of air quality parameters such as CO₂ levels and particulate matter, facilitating proactive maintenance strategies that enhance safety and efficiency [16]. Predictive Maintenance (M3.0) employs ML models and digital twin technology to forecast equipment failures, minimizing downtime and operational costs while maintaining stable indoor environments [17,18]. Despite these advancements, challenges such as data quality, high initial investment, and the need for skilled personnel must be addressed for effective implementation [19].

2.3. Maintenance 4.0 Structure

Building on Section 2.2, the structure of M4.0 differs based on the corresponding application. For example, Nagpal et al. [15] proposed a simple structure for M4.0. As shown in Figure 5, the basic idea in M4.0 stems from the data collection method and the utilization of AI with limited details on data collection except for the historical data.

A more detailed structure was proposed by Farah et al. [20]. Farah intensified the selection of the sensing elements and the proper installment to ultimately obtain accurate data. To obtain these data, sensing elements should be connected to a strong and compatible network. The collected data (i.e., from sensors, historical data, and the ARP system should then be processed in real-time-based processes. Next, ML or AI algorithms are utilized to predict machine failure. Finally, a dashboard is designed to visualize the process to help humans to read and understand the performance of machine(s) and understand the behavior of the controlled machine.

In this paper, the following structure is proposed which mainly consists of a sensing element, which is the source of the data; in our case, it is “air parameters”. After that, the data were stored on a server and visualized in real time. Next, the data are preprocessed, which includes filtering and preparation before the data analysis and prediction using AI algorithms. Then, the results are represented on dashboards, and in the case that anything goes wrong, the system will send alerts. Finally, the system is connected through the IoT, as shown in Figure 6.

3. Maintenance 3.0: Predictive Maintenance

As mentioned earlier, although there is no evidence to suggest that the evolution of maintenance processes coincides directly with the industrial revolutions, there is no doubt that the increasing complexity of industrial systems has driven the development of advanced maintenance strategies and methods. In addition to the first and second maintenance phases, Maintenance 3.0 has gained significant attention from researchers. The third industrial revolution was crucial in shaping the programs and methodologies used to maintain structural integrity. To avoid confusion, this section focuses on the predictive maintenance in the HVAC systems.

Several frameworks have been developed to maintain the efficiency of ventilation systems. Most of these systems were reviewed by Es-sakali et al. [11]. They categorized the models into knowledge-based, analytical models, and data-driven models. Deeper insights into data-driven models have been reviewed by Esteban et al. [12] and Sanzana et al. [21]. The reviews discussed research in predictive maintenance until 2022. In this section, the overlooked articles are investigated and discussed to provide a deeper insight into M 3.0, which is the gate of M4.0. For instance, Tian et al. [22] devised a framework for the predictive maintenance of HVAC systems utilizing autoencoders to accurately classify system health conditions, aiming to enhance maintenance practices through improved health condition classification. On the other hand, Bouabdallaoui et al. [23] presented a flexible predictive maintenance framework applicable across various building facilities, utilizing ML techniques.

ML and deep learning have been combined with disruptive technologies such as the digital twin (DT) to develop a predictive maintenance framework for indoor climate systems [24]. The deployment of the DT has shown remarkable efficiency in the maintenance planning of the HVAC system, particularly air handling units (AHUs) [18,25]. Hosamo et al. [26] utilized a digital twin approach to enhance building occupant comfort and improve overall building performance through predictive maintenance strategies. This study underscores maintenance’s importance, particularly predictive maintenance, in optimizing HVAC system performance and extending equipment lifespan.

Another research stream has been initiated to study the impact of external factors on ventilation systems. For instance, Sanzana et al. [24] investigated the impact of external weather metrics on TES-AC systems, emphasizing the necessity of comprehending how weather data influence system sustainability and operational efficiency. Mirza also explored computational intelligence applications for the predictive maintenance of TES-AC systems, acknowledging weather conditions’ significance in devising effective maintenance strategies. Tancev [18] addressed drift components and variability in low-cost electrochemical sensor systems for air quality monitoring, proposing predictive maintenance strategies to improve measurement reliability. Based on weather data, Zhao et al. [27] created a prediction model for energy consumption in district cooling systems, with an aim to optimize HVAC system efficiency without intricate building information, thereby offering a straightforward yet accurate approach for maintenance control within district cooling systems.

Building on these advancements, the rapid development in AI, IoT, and cloud computing has led to the emergence of Maintenance 4.0. Fault diagnosis techniques are the gateway to this new era, paving the way for more intelligent and proactive maintenance strategies, which are discussed in the next section on HVAC systems.

4. Fault Diagnosis Systems

Recent advancements have led to significant improvements in fault diagnosis systems. Machine learning and deep learning techniques have been successfully applied to enhance accuracy and efficiency in detecting and diagnosing faults. The effectiveness of these models depends on the quality and type of data they analyze. The key data sources used in predictive fault diagnosis include the following:

• Environmental parameters: air temperature, humidity, CO₂, and particulate matter (PM2.5, PM10) [28];
• Energy consumption metrics: real-time power usage, efficiency loss trends, and operational cycles [29];
• Equipment performance data: vibration levels, pressure differentials, airflow rates, and historical maintenance logs [30].

These input data provide the foundation for predictive analytics models, which are further refined through feature selection and preprocessing techniques.

Notable approaches include multi-scale convolutional neural networks, interpretable mechanism mining, and convolutional neural networks. These innovations have the potential to optimize maintenance practices and reduce energy consumption.

For example, Wu et al. [31] introduced a fault diagnosis method tailored for HVAC systems with imbalanced data, employing a multi-scale convolution composite neural network to enhance accuracy. Similarly, Yun et al. [32] contributed a data-driven fault detection and diagnosis scheme for HVAC systems, aiming to enhance maintenance reliability and performance compared to traditional methods, particularly for AHUs. Movahed et al. [33] proposed a data-driven framework for fault detection in HVAC systems using ML algorithms, with a focus on aiding maintenance activities by identifying faults and their root causes. Khan et al. [34] proposed a proactive attack detection system for HVAC systems using ML models, focusing on enhancing security without explicitly discussing maintenance aspects. Elnour et al. [35] introduced a fault diagnosis approach using 2D CNNs for single actuator faults in HVAC systems, stressing the importance of fault diagnosis for preventive maintenance. Yang et al. [36] developed a machine learning-based methodology for prognostics in central heating and cooling plant (CHCP) equipment, aiming to improve fault detection, diagnostics, and prognostics. Chen et al. [37] created a deep learning-based method for fault diagnosis in HVAC systems, prioritizing interpretability and generalization to enhance fault detection accuracy. Although maintenance is not explicitly addressed, the focus on interpretable mechanisms mining aids in refining fault diagnosis processes. Albayati et al. [38] introduced a semi-supervised ML approach for fault detection and diagnosis in HVAC systems, highlighting proactive maintenance to prevent catastrophic failures and reduce energy costs. Similarly, Martinez-Viol et al. [39] developed a semi-supervised ML model to evaluate the effectiveness of transfer learning methodologies in fault detection and diagnosis for AHUs. They emphasized the importance of reliable fault detection and diagnosis in reducing energy consumption and maintenance costs. Lee et al. [40] developed an AI-driven fault detection and diagnosis (AI-FADD) system for HVAC equipment to improve fault detection reliability and reduce maintenance costs, emphasizing the financial benefits of reducing false alarms and enhancing fault classification accuracy. Dixit et al. [41] developed a novel fault diagnosis method using a conditional auxiliary classifier GAN and meta learning to enhance fault detection accuracy in rotary machines, aiming to improve maintenance practices. Also, Albayati et al. [42] studied early adopters of FDD technologies for rooftop HVAC systems, aiming to optimize operation and minimize maintenance costs.

5. Sensing Elements and IoT Devices

As illustrated in Figure 5 and Figure 6, sensing elements and IoT devices are crucial tools to collect real-time data for the HVAC system. This real-time data collection is the backbone of the Industry 4.0 setup and thus M4.0 under the same setup. This section summarizes the field of air quality monitoring using traditional and wireless sensor methods. Also, it assesses the effectiveness of different sensors, pinpoints possible problems, and suggests improvements.

5.1. Types of Sensors

Sensor connectivity is crucial for real-world data collection in IoT-based HVAC diagnostic systems, especially for air quality and pollution monitoring, and as a key driver of Maintenance 4.0 within these systems. A review of the literature reveals diverse approaches: some researchers use wired sensors, others wireless, and some a combination of both, often driven by availability. Notably, a significant number of publications fail to specify sensor connectivity. Specifically, an analysis of 88 research articles on pollution monitoring (Figure 7) showed that 40 did not define the IoT sensor’s connectivity mode, 30 utilized wireless sensors, 10 employed a combination of wired and wireless, and the remainder used wired sensors. This distribution highlights the increasing prevalence of wireless technology in pollution monitoring, likely due to its advantages in monitoring and real-time data transfer.

Sensor connectivity in IoT deployments can vary, but sensors typically connect using a range of protocols chosen based on factors such as range, bandwidth, power consumption, and cost. Wireless communication is prevalent, utilizing various network types. Personal area networks (PANs), like Bluetooth [43] and Zigbee [44], are suited for short-range, low-power applications such as wearables and home automation. Local area networks (LANs), primarily using WiFi [45], are appropriate for higher bandwidth applications within a limited area, such as industrial automation and smart buildings. Wide area networks (WANs) offer long-range connectivity; LoRaWAN [46] is optimized for low-power communication in environmental monitoring, while cellular technologies, including NB-IoT [47] and LTE-M [48], are suitable for wide-area coverage and mobile applications like asset tracking and smart metering. Other wireless protocols, such as the proprietary XBee [49] are also found in industrial settings. In contrast to wireless options, Ethernet remains the standard for wired LAN connections, offering high bandwidth and reliability in industrial and enterprise environments [50].

A review of the literature, based on the defined search criteria, revealed the following distribution of connection technologies: WiFi (twenty-one articles), Zigbee (nine articles), Cloud (three articles), XBee radio (three articles), and Bluetooth (eight articles), please see Figure 8. Within the wired category, all six instances utilized Ethernet connections. One article employed both wired and wireless connectivity, while 14 instances did not specify the connection type. By displaying the relative frequency and distribution of each connection type, this hierarchical graphic efficiently highlights the most often utilized technologies within the dataset.

These protocols enable seamless communication and data exchange in the IoT ecosystem.

Table 2. Summary of research related to IoT.

Research	Focus on IAQ and Monitoring	Focus on Building Automation, Maintenance, and Optimization
[18]	✔
[51]	✔
[52]	✔
[53]	✔
[54]	✔
[26]		✔
[22]		✔
[25]		✔
[34]		✔
[55]		✔
[23]		✔

summarizes the research related to IoT.

5.2. IoT Sensors for Different Pollutant Categories

The data analysis offers important insights into the presence of different pollutants in pertinent literature within the context of Maintenance 4.0, as shown in Figure 9 and Table 3. Particulate matter (PM—forty-one occurrences), methane (CH₄—four occurrences), sulfur dioxide (SO₂—seventeen occurrences), temperature (twenty occurrences), humidity (fifteen occurrences), ozone (O₃—twenty-three occurrences), oxygen (O₂—two occurrences), nitrogen dioxide (NO₂—thirty occurrences), carbon monoxide (CO—thirty-eight occurrences), and carbon dioxide (CO₂—twenty-six occurrences) are among the pollutants that are displayed in the accompanying bar chart. It is noteworthy, nonetheless, that certain contaminants are not included in the list because of their rare incidence. This choice was made in order to select the most important contributors and streamline the visualization. This methodological technique guarantees lucidity and concentrates on the most common contaminants, thus augmenting the usefulness of the information in comprehending the ecological consequences of Maintenance 4.0 procedures.

The emergence of IoT platforms presents several opportunities to revolutionize the management of IAQ [56]. Recent research highlights the growing impact of IoT technology in air quality monitoring and management. Studies by Wei Hu et al. [18] and Vajs et al. [57] showcase the potential of IoT sensor networks for real-time data collection on various air quality parameters like temperature, humidity, and PM2.5. These data are crucial for evaluating, predicting, and optimizing indoor and outdoor air quality. Notably, Vajs et al. emphasizes the use of low-cost sensor networks, making air quality monitoring more accessible for nationwide projects.

Table 3. Frequency of pollutants.

Environmental Parameters	Sample of References
Temperature	[11,21,23,25,38,52,53,56,57,58,59,60,61,62,63,64,65,66,67,68]
Humidity	[21,38,69]
CO	[6,13,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83]
O3	[84,85,86,87,88,89,90,91,92,93]
NO2	[51,81,88,89,94,95]
PM	[96,97,98,99,100,101,102,103,104,105,106,107,108]
O2	[38,71]
CH4	[13,75,76,109]
CO2	[110,111,112]
SO2	[84,93,106]

Table 3. Frequency of pollutants.

Environmental Parameters	Sample of References
Temperature	[11,21,23,25,38,52,53,56,57,58,59,60,61,62,63,64,65,66,67,68]
Humidity	[21,38,69]
CO	[6,13,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83]
O3	[84,85,86,87,88,89,90,91,92,93]
NO2	[51,81,88,89,94,95]
PM	[96,97,98,99,100,101,102,103,104,105,106,107,108]
O2	[38,71]
CH4	[13,75,76,109]
CO2	[110,111,112]
SO2	[84,93,106]

Furthermore, research explores the effectiveness of IoT in air pollution detection and mitigation strategies. For example, Bainomugisha et al. [51] demonstrate the development of the AirQo system, utilizing IoT technology for low-cost and effective air quality monitoring in urban environments, particularly beneficial for resource-constrained settings. Additionally, Saravanan et al. [70] investigate the integration of IoT with ML models, specifically bidirectional recurrent neural networks (RNNs), to enhance the accuracy of air pollution detection. This integration offers a significant advancement in air quality monitoring compared to traditional methods. Beyond monitoring, Tancev et al. [54] explore the role of IoT in smart city initiatives. Their work focuses on integrating low-cost sensor systems into the IoT network for air quality monitoring. Interestingly, the study investigates using mobile sensor systems, such as those on public transport or drones, for potential recalibration based on exchanged data. While the paper does not explicitly address sensor drift or optimal recalibration timing, it presents a promising avenue for further exploration.

The convergence of IoT technology with other emerging technologies is transforming building automation and maintenance practices. Studies by Hosamo et al. [25,26] showcase the potential of digital twin technologies, which integrate IoT sensors with BIM, cloud computing, and AI. These digital twins enable real-time monitoring of system performance, facilitating predictive maintenance strategies for optimal building operation and occupant comfort. Hosamo et al. [26] specifically focus on applying this framework to AHUs in HVAC systems. However, implementing these data-driven approaches requires adequate data collection. Tian et al. [22] highlights the challenge of applying such techniques to systems without embedded IoT sensors. Bouabdallaoui et al. [23] demonstrate how various IoT devices, like meters and vibration sensors, can be used to collect relevant data from building facilities.

Security considerations are also crucial in the age of IoT. Khan et al. [34] emphasizes the importance of analyzing data generated by IoT devices in HVAC systems to proactively detect cybersecurity threats. This highlights the need for robust security measures when integrating IoT into building systems. Beyond predictive maintenance, IoT data can also be used for broader building optimization goals. Villa et al. [55] propose using anomaly prediction models that leverage IoT sensor networks and BIM to identify potential issues in building systems. This focus on data-driven insights can contribute to more sustainable building maintenance practices.

6. Data Management, Collection, and Analysis Techniques

Efficient data management is vital for optimizing ventilation system performance and maintenance. This section outlines the key techniques for collecting, storing, cleaning, and analyzing sensor data, providing insights that enhance decision-making for HVAC system optimization.

The data management process in ventilation systems starts with data collection, where data is gathered from various sensors and IoT devices, and ERP and other relevant data are collected. Next, data management involves maintaining these data in databases or cloud storage. Data cleansing is performed to ensure quality, reducing noise and managing missing values. The cleaned data are then analyzed using techniques such as ANNs in the data analysis phase to generate results. Finally, data visualization presents these insights in a clear and simple format, allowing facility management administrators to optimize the performance and maintenance of HVAC systems.

Classifying Articles Based on Common Analysis Techniques

It is necessary for data extraction to apply advanced analysis methods for the proper sorting of articles. Sentiment analysis determines emotional tone, which is used to understand public opinion, while topic modeling involves the identification of recurrent themes for the organization of content. Entity identification allocates mentioned entities and boosts the information collection process. Such methods contribute to simplifying decision-making and data analysis, thus producing holistic views, allowing the right judgment to be made. The most common techniques are the following:

Descriptive analysis: Summing up and characterizing the key elements of a dataset or phenomenon without drawing any conclusions or making predictions.

Predictive analysis: The use of previous data to forecast future outcomes or patterns, which is frequently done using statistical algorithms and machine learning approaches.

Diagnostic analysis: Looking at the underlying causes or circumstances that contributed to a given outcome or occurrence to understand why it occurred.

Prescriptive analysis: Recommending actions or treatments to attain desired results based on descriptive, predictive, and diagnostic analyses to inform decision-making.

The distribution of publications across several analysis techniques is shown in the bar chart Figure 10 depicts. Descriptive analysis accounts for the largest amount, 38.3%, concentrating on data summarization. Predictive analysis follows closely at 37.0%, indicating a desire to foresee trends. Diagnostic analysis accounts for 22.7%, stressing attempts to discover root causes, whereas prescriptive analysis accounted for 28.5%, indicating a growing interest in prescribing measures based on analytical insights. This distribution emphasizes a multidimensional approach to studies in various disciplines, with a focus on understanding, forecasting, diagnosing, and prescribing measures to improve processes and systems.

Several studies have demonstrated the effective application of these techniques. For instance, Bouabdallaoui et al. [23] utilized a prescriptive analysis framework to offer maintenance recommendations for building ventilation systems. While this approach provided valuable insights, it lacked a thorough evaluation of model development time and profitability. In contrast, Schneising et al. [109] focused on preventive maintenance in air handling units (AHUs) by employing diagnostic techniques to predict potential faults and optimize performance. However, like the previous study, it highlighted the need for further economic analysis to assess its practical implications. Other research has combined descriptive and diagnostic techniques to both characterize system performance and identify potential issues [29,108]. For example, a model-based predictive control methodology for HVAC systems used descriptive analysis to outline the key elements of the control system, while diagnostic techniques revealed limitations in scalability and performance under varying environmental conditions. In real-time monitoring of indoor aerosol concentrations, descriptive analysis helped assess the sensor network’s performance, but diagnostic analysis identified challenges such as the lack of standardized data integration and insufficient discussion of monitoring system limitations. Studies that combine prescriptive and diagnostic techniques have shown promise in optimizing maintenance strategies. One such approach in facility management used prescriptive analysis to recommend actions while employing diagnostic techniques to identify the underlying causes of system inefficiencies [25]. However, scalability and real-time applicability remain areas requiring further development.

In order to implement Maintenance 4.0, a robust data management process that encompasses collection, integration, analysis, and interpretation is required. Research across various domains reveals critical gaps and opportunities within this process.

Studies on HVAC systems [25] and facilities maintenance management (FMM) processes [109] highlight the need for real-time data collection from diverse building types and climates. These data, encompassing sensor readings, system performance metrics, and environmental factors, must be collected consistently and reliably. Integrating data from multiple sources, such as sensor networks, smartphone sensors, and building management systems, poses significant challenges. Studies on indoor aerosol concentrations [108] and road anomalies [23] emphasize the lack of standardized integration solutions, hindering efficient data processing. Then, extracting meaningful insights from integrated data requires sophisticated analysis techniques. While studies on road anomalies [29] point to limitations in feature extraction from multiple coordinate axes, research on fault detection and diagnosis [32] showcases the potential of machine learning and neural networks for identifying anomalies and predicting failures. Interpreting analysis results and translating them into actionable insights is crucial for optimizing maintenance strategies. Studies on economic viability [51,90,100] underscore the importance of evaluating cost-effectiveness and profitability, while research on calibration methods [63] and feature extraction techniques [47] emphasize the need for reliable calibration models and generalizable findings across different system types.

Table 4. Existing review gaps in implementing data analysis techniques.

Article	Citations	Research Focus	Gaps
[29]	359	Development and application of a model-based predictive control methodology for HVAC systems in buildings to optimize thermal comfort and minimize energy consumption.	Discussing limitations in real-world implementation. Further validation and testing across a wider range of building types and climates.
[113]	133	Utilizing ML techniques for classifying road anomalies based on smartphone sensor data.	Limited extraction methods from multiple coordinate axes. Lack of scalability and real-time applicability.
[25]	115	Optimize predictive maintenance and dynamic maintenance strategies in FMM processes.	Dependency on developer experience. Lack of standardized data integration solutions.
[108]	102	Evaluating the effectiveness of a sensor network platform for real-time monitoring of indoor aerosol concentrations.	Absence of detailed recommendations and future work. Lack of discussion on the challenges of monitoring systems.
[109]	94	Developing a scientific algorithm to simultaneously retrieve XCO and XCH4 from shortwave infrared spectra recorded by the TROPOMI instrument onboard the Sentinel-5 Precursor satellite.	Uncertainties with retrieval algorithm with low surface reflectance or residual cloudiness.
[40]	88	Developing a generic framework for predictive maintenance in buildings, incorporating literature review, interviews with FM experts, and case study demonstration.	Minimal exploration of the time required for model development and profitability. Inadequate consideration of the characteristics of buildings and their impact on the effectiveness and scalability.
[18]	75	Proposing a data-driven FDD scheme for AHU, specifically addressing undefined states to enhance maintenance reliability.	Limited evaluation of economic viability and cost-effectiveness for integrating the proposed scheme into AHU maintenance, emphasizing the need for further analysis.
[59]	72	Development of a feature extraction technique for temperature and power data in TU systems, enabling automatic fault prediction and diagnosis.	Limited to a specific type of TU (fan coil unit).
[114]	71	Development of a semi-autonomous micro aerial vehicle (MAV) system equipped with image capture capabilities and neural network-based CBC detection for vessel structures.	Weak integration between range sensors and feature extraction.
[67]	65	Evaluating calibration methods for air pollution low-cost multi-sensor platforms, comparing LR, MLR, and ANN techniques.	Ensure reliable calibration. Identifying suitable calibration models for different types of sensors with ANN model, especially for CO sensors.

summarizes the research progress on the different data collection and data analysis models for HVAC systems.

7. Artificial Intelligence Techniques (AI)

As mentioned earlier, Maintenance 4.0 can be defined as a variant of Industry 4.0 setups. In this respect, AI is a key component in the setup of Maintenance 4.0.

In the HVAC literature, AI has been used to build advanced models to evaluate air quality in various places [29,115]. In addition, it has been used to calibrate and maintain the sensor systems responsible for monitoring air quality. When talking about commonly used, low-cost PM sensors, machine learning algorithms have been widely used to calibrate these devices, improving their accuracy and reliability [57]. In addition, we should point out the use of AI models to detect, predict, and diagnose faults in heating, ventilation, and air conditioning equipment, which has played a crucial role in maintaining IAQ and reducing maintenance costs. In general, applications of artificial intelligence in the field of air quality show their potential to improve monitoring, forecasting, and maintenance processes, which ultimately contribute to improved air quality management.

Comparison of AI Techniques for Air Quality Prediction

A variety of artificial intelligence techniques have been used to predict air quality, including precise machine learning models and advanced deep learning architectures.

Table 5. Comparison of AI techniques.

AI Technique	Description	Strengths	Limitations	Example Applications
Traditional machine learning	Includes models like ANN, linear, and logistic regression.	Relatively simple to implement; suitable for initial exploration of relationships.	Limited capacity to capture complex, non-linear environmental relationships; often requires extensive feature engineering for optimal performance.	Predicting air quality index and pollutant concentrations [48]; air quality forecasting using decision tree [111].
Advanced machine learning	Includes models like Ada Boost, SVR, RF, KNN, and MLP regressor.	More sophisticated; can capture complex relationships with less feature engineering; adaptable to dynamic environmental conditions.	May require larger datasets for training; interpretability of complex models can be challenging.	Air quality prediction in smart cities using LSTM model [110].
Hybrid deep learning	Combines multiple deep learning architectures, such as CNN-BiLSTM, for enhanced accuracy and predictive capabilities.	Can achieve high accuracy for a wide range of pollutants; effective for multi-step-ahead forecasting.	Limitations in predicting specific pollutants (e.g., primary pollutants like NO2, SO2, and CO) may require further model refinement.	Pollutant concentration prediction using CNN-BiLSTM in Tianjin, China [74].
Decentralized AI networks	Employs a network of smart sensors, each equipped with an ANN, for localized prediction and adaptable forecasting.	Captures local microclimates and urban environments effectively; adaptable to changing conditions; enhances responsiveness to local variations.	Requires coordination and management of multiple sensors; data synchronization and communication between sensors can be challenging.	Real-time air quality monitoring and prediction in urban environments [94].

illustrates the strengths, limitations, and typical applications of various AI techniques used to predict air quality.

The table presents a spectrum of AI techniques employed in environmental monitoring, particularly focusing on air quality prediction. Traditional machine learning, including models like ANNs, linear regression, and logistic regression, offers a simple starting point for exploring relationships within data [47,110]. These models are relatively easy to implement but struggle to capture complex, non-linear environmental interactions and often require extensive feature engineering for optimal performance. Advanced machine learning techniques, such as Ada Boost, SVR, RF, KNN, and MLP regressors, provide a more sophisticated approach [23,80,115]. They can capture complex relationships with less need for feature engineering and are adaptable to dynamic environmental conditions. However, they may require larger datasets for training, and their interpretability can be challenging.

Hybrid deep learning, combining architectures like CNN and BiLSTM, offers enhanced accuracy and predictive capabilities [18,81]. This approach can achieve high accuracy across a wide range of pollutants and is effective for multi-step-ahead forecasting. However, it may face limitations in predicting specific primary pollutants, necessitating further model refinement. Decentralized AI networks utilize a network of smart sensors, each equipped with an ANN, to provide localized predictions and adaptable forecasting [101]. This model effectively captures local microclimates and urban environments, enhancing responsiveness to variations. However, it requires careful coordination and the management of multiple sensors, as well as effective data synchronization and communication between them.

Major problems and challenges facing accurate air quality predictions are data issues and a lack of data quality. BRITS-ALSTM for the imputation of missing data: The BRITS-ALSTM model, a deep learning framework, handles missing air quality data values by leveraging global dependencies and local multivariate correlations in time series data [82]. By combining BRITS and LSTM with the attention mechanism, this model enhances the accuracy of data evaluation. This makes it particularly valuable for areas where data are limited, such as the Qinghai-Tibet Plateau. To address data gaps and improve the reliability of air quality predictions, additional techniques such as Kalman filtering and data augmentation are being explored [28,116].

Furthermore, the transformation achieved by AI in air quality administration and hardware diagnostics has prompted more successful mediations and viable and phenomenal upkeep systems. Ongoing contamination control: To further develop contamination control measures, artificial intelligence-fueled frameworks can use accessible continuous air quality information. For instance, AI calculations can anticipate and recognize places with high contamination and animate designated mediations, for example, adjusting traffic flow or industrial emissions [117].

HVAC predictive maintenance: AI reasoning plays a significant part in upgrading the dependability and productivity of HVAC systems. By utilizing and dissecting sensor information, AI consciousness models can foresee gear breakdowns, which helps in proactive upkeep and lessens free time [55]. This technique depends on decreasing working expenses, decreasing energy utilization, and augmenting the existence of the framework.

Indoor air quality administration: To screen and foresee indoor air quality (IAQ) progressively, simulated intelligence-based frameworks like “Vayuveda” influence the Web of Things and artificial intelligence to accomplish this [102]. These frameworks can give significant pieces of knowledge to work on indoor conditions and upgrade general wellbeing by coordinating high-level AI models like PS, GA, and RF.

8. Boosting the Implementation of Maintenance 4.0 in HVAC Systems

The increasing use of AI methods holds significant potential for advancing HVAC system maintenance, including air quality monitoring, predicting equipment failures, and optimizing maintenance strategies. However, several challenges and gaps must be addressed to fully realize the benefits of Maintenance 4.0 in this domain.

8.1. Existing Gaps and Challenges

Several key challenges hinder the widespread and effective implementation of Maintenance 4.0 in HVAC systems:

Data availability and quality: A significant obstacle is managing missing information and ensuring data quality. Accurate and reliable AI models depend on comprehensive and clean datasets. The lack of standardized data collection protocols and the presence of noisy or incomplete data can significantly impact model performance.

Model interpretability and transparency: Understanding how AI models arrive at their predictions is crucial for building trust and facilitating informed decision-making. Many AI models, particularly complex deep learning models, operate as “black boxes,” making it difficult to understand their internal workings. This lack of transparency hinders adoption and limits the ability to identify and correct potential biases or errors.

Model generalizability and transferability: Models trained on specific datasets or under particular environmental conditions may not generalize well to other contexts. Ensuring model robustness and applicability across diverse HVAC systems, building types, and climates is a major challenge. This includes the challenge of integrating IoT-based IAQ platforms with existing building management systems, requiring careful tuning and compatibility solutions.

Integration with existing systems: Integrating advanced AI-driven maintenance solutions with existing BMS and other building systems can be complex. Compatibility issues, data exchange protocols, and the need for middleware solutions pose significant integration challenges.

Economic evaluation and scalability: There is a gap in the economic evaluation of maintenance optimization methodologies, particularly regarding their scalability and real-time applicability. This lack of a clear cost–benefit analysis can hinder investment and adoption.

Occupant integration: Effectively integrating occupant feedback and preferences into IAQ management systems presents a challenge. User-centered design, intuitive interfaces, and training programs are needed to empower occupants to monitor and contribute to IAQ management.

Lack of explainable AI (XAI) solutions: The absence of dedicated XAI solutions for IAQ management and prediction highlights the need for focused research in this area.

8.2. Proposed Solutions for the Future

To overcome these challenges and unlock the full potential of Maintenance 4.0 in HVAC systems, the following solutions are proposed:

Advanced data management techniques: Future research should prioritize developing techniques for effectively managing missing information. This includes imputation methods, data augmentation strategies, and robust data preprocessing pipelines. Enhanced data sources for environmental monitoring, including integrating additional sensors and measuring devices, are also crucial.

Transparent and explainable AI: Focus should be placed on utilizing transparent and understandable AI strategies. This includes exploring techniques like rule-based systems, decision trees, and attention mechanisms in deep learning models to enhance transparency and interpretability. Research should also focus on developing new XAI methods tailored for IAQ management and prediction.

Robust model validation and generalization: Rigorous model validation across diverse datasets and environmental conditions is essential to ensure robustness and generalizability. Techniques like cross-validation, transfer learning, and domain adaptation should be employed to improve model performance in unseen scenarios. Future studies may also compare and integrate advanced data sources with existing predictors for more accurate predictions.

Seamless system integration: Developing standardized protocols, middleware solutions, and open APIs can facilitate a smoother integration of AI-driven maintenance solutions with existing BMS and other building systems. Collaboration between technology providers and building operators is essential for achieving interoperability.

Comprehensive economic analysis: Conducting thorough economic evaluations of Maintenance 4.0 methodologies, including scalability and real-time applicability assessments, is crucial to demonstrate the value proposition and encourage wider adoption.

User-centered design and occupant engagement: Employing user-centered design principles to develop intuitive interfaces and providing training programs for occupants can improve their engagement and contribution to IAQ management. Developing “occupant-in-the-loop” platforms is essential.

Optimization of machine learning models: Future work should focus on optimizing machine learning models for fault detection, energy consumption prediction, and environmental monitoring to enhance accuracy and efficiency. This includes exploring new algorithms, feature engineering techniques, and hyperparameter optimization strategies.

Long-term forecasting and predictive maintenance: Future research could involve the long-term forecasting of environmental parameters and equipment failures using advanced machine learning models and considering various scenarios to improve predictive maintenance strategies and optimize maintenance scheduling.

By addressing these challenges and implementing the proposed solutions, the full potential of Maintenance 4.0 can be realized in HVAC systems, leading to improved energy efficiency, enhanced IAQ, reduced maintenance costs, and more sustainable and healthy built environments.

8.3. Advantages of Maintenance 4.0 in HVAC Systems

While challenges such as data quality, integration complexity, and model transparency remain, Maintenance 4.0 offers significant advantages:

• Energy efficiency: Reduces energy consumption by optimizing HVAC operation.
• Cost savings: Minimizes unplanned downtime and maintenance costs through predictive analytics.
• Improved IAQ: Enhances IAQ by ensuring timely maintenance and performance optimization.
• Scalability: IoT-driven solutions can be integrated into smart building systems for enhanced automation.
• Addressing the existing gaps while leveraging these advantages will accelerate the adoption of Maintenance 4.0, leading to more sustainable and intelligent HVAC management.