Integration of Industry 5.0 Technologies in the Concrete Industry: An Analysis of the Impact of AI-Based Virtual Assistants

Abstract

The construction industry, traditionally lagging behind in terms of digitalization, faces significant challenges in its transition to Industry 4.0, which is characterized by the use of advanced technologies such as artificial intelligence (AI), the Industrial Internet of Things (IIoT), and cloud computing. This article presents the development and implementation of an AI-based virtual assistant, designed to optimize the operation and maintenance of concrete production plants. The assistant helps reduce the margin of human error, improve operational efficiency, and facilitate continuous training for operators. These advancements foster a more collaborative and digitalized environment, while also generating environmental, economic, and social benefits: reduced material and energy waste, lower carbon footprint, increased workplace safety, and strengthened organizational resilience. The results show high accuracy in voice transcription (96%) and a 100% success rate in responding to technical queries, demonstrating its effectiveness as a support tool in industrial settings. Based on these findings, it is concluded that the incorporation of AI-based virtual assistants promotes a more sustainable and responsible production model, aligned with the Sustainable Development Goals of the 2030 Agenda, and anticipates the principles of Industry 5.0 by promoting symbiotic collaboration between humans and technology. This innovation represents a key advancement in transforming the concrete industry, contributing to productivity, environmental sustainability, and workplace well-being in the sector.

1. Introduction

The concrete industry is currently undergoing a critical transition. It must enhance operational efficiency while simultaneously complying with increasingly stringent environmental and digital requirements [1,2]. Unlike other industrial sectors where digital transformation is more mature, concrete production plants still rely heavily on manual procedures and exhibit low levels of automation [3]. This dependence reduces productivity, increases the risk of operational incidents, and limits product quality.

To address these limitations, Industry 5.0 has emerged as a new paradigm that builds upon the automation goals of Industry 4.0. It promotes human–machine collaboration and places operators at the core of digitalization strategies [4,5]. However, the implementation of these technologies is still hindered by legacy infrastructures, fragmented systems, and interoperability challenges [6,7].

Despite the availability of advanced technologies, their adoption in concrete production environments remains limited. This technological gap prevents the sector from fully leveraging the benefits of digitalization and customization offered by Industry 5.0. In this context, intelligent virtual assistants represent a promising solution thanks to their ability to support operators with real-time information, facilitate decision-making, and assist in managing complex tasks [8].

The main objective of this research is to design, implement, and validate an AI-based virtual assistant in a real concrete production plant, aiming to improve operational efficiency and foster human-centric digital transformation. This study assesses the system’s performance, identifies challenges to its adoption, and proposes a replicable implementation model grounded in Industry 4.0 and 5.0 principles.

The key contributions of this work are as follows:

• The deployment and evaluation of a virtual assistant (KobBot) in a real industrial environment.
• A systematic analysis of its operational benefits and integration limitations.
• A user experience assessment using structured surveys and statistical validation.
• An exploration of its alignment with Sustainable Development Goals (SDGs 8, 9, and 12).
• The proposal of a scalable and modular implementation model suitable for other industrial contexts.

The remainder of this article is organized as follows: Section 2 presents a focused review on the adoption of digital technologies and AI in concrete production, highlighting current practices and gaps. Section 3 details the methodology and tools used in the implementation. Section 4 reports the outcomes of the deployment, including technical validation and user feedback. Section 5 discusses the impact, limitations, and future directions of the assistant’s deployment, emphasizing its role in enabling human-centered digitalization in heavy industry. Finally, Section 6 discusses broader implications and outlines future perspectives for the use of intelligent assistants in industrial settings.

2. Research Context

The development of conversational virtual assistants in industrial environments has gained increasing relevance in recent years, driven by the need to improve interaction between operators and control systems. Within the framework of Industry 4.0, these systems aim not only to automate repetitive tasks but also to provide contextualized, comprehensive, and real-time assistance, enabling users to access critical information more intuitively and efficiently [9,10,11].

2.1. Operator-Centered Approaches and Industry 5.0

As industrial environments evolve toward greater collaboration, this work represents an approach aimed at integrating advanced digital technologies into production processes, incorporating concepts from Industry 5.0 such as operator centrality and human–machine interaction [12,13,14,15]. One of its implementations, the KobBot virtual assistant, combines natural language processing and understanding (NLP and NLU), automatic speech recognition (ASR), and machine learning techniques and connects with industrial systems via protocols such as OPC UA, MQTT, and RESTful APIs [16]. From a technical perspective, this type of system aims to overcome the limitations of purely automation-focused approaches, facilitating more natural interactions, providing contextualized assistance, and allowing efficient access to technical documentation during operation and incident resolution [4,17]. This approach aligns with Industry 5.0 principles by valuing the role of the operator in highly digitalized environments, although its effectiveness in real scenarios still requires systematic validation based on comparable empirical studies [12,13].

To operationalize the operator-centered approach, several reference frameworks propose interoperability structures and user–machine interaction levels. Lasi et al. propose an Industry 4.0 reference model that defines integration layers, including a semantic layer for human–machine communication, and emphasizes the need to adapt these architectures for collaborative scenarios [18,19]. Tao et al., for their part, describe a data-driven smart manufacturing approach that, by incorporating conversational AI modules and virtual assistants, demonstrated 12–15% improvements in operational efficiency and up to 10% reductions in maintenance costs in industrial case studies [20]. These frameworks highlight the lack of comparable empirical studies in concrete production and justify the need to systematically validate solutions like KobBot in real-world settings.

2.2. Use Cases and Related Technologies

Currently, most advancements in industrial virtual assistants have focused on robotic applications or structured manufacturing environments [5,21,22,23]. The Max system exemplifies how an NLP-based assistant can establish effective interaction with industrial robots using a RESTful architecture and human-inspired conversational strategies [16]. Voice control systems have also been proposed to improve ergonomics, reduce risks, and increase efficiency in critical tasks [24]. Recent reviews highlight the potential of AI-based conversational agents to enhance human–AI collaboration and outline future development directions [25]. In this context, a “white-label” approach has been suggested to encourage adoption in manufacturing SMEs by reducing costs and implementation complexity [26].

2.3. Current Limitations in Virtual Assistant Integration

The specialized literature indicates that despite advances in language-based AI, the adoption of virtual assistants in real industrial environments remains limited [27,28,29]. A key gap lies in integration with existing platforms, as most assistants lack direct connections to control systems or access to official technical documentation [8,30,31]. KobBot addresses this gap through its integration with plant infrastructure and its ability to generate contextualized responses based on system status and authorized technical sources [32]. Several recent studies have delved into the causes of these limitations. Among the most cited factors are the following:

1.

The lack of standards for integrating assistants with legacy OT infrastructures, requiring ad hoc gateways and increasing implementation costs [33,34,35].

2.

Latency and degradation in automatic speech recognition (ASR) accuracy in noisy environments, which hinders adoption on the plant floor [33].

3.

Cybersecurity and privacy risks from exposing process control data to external AI services [36].

4.

The absence of consensus frameworks for assessing the economic return and ongoing quality of the conversational agent in operation [8,37].

Despite these limitations, various recent technological advances offer new opportunities to overcome current obstacles in the integration of industrial virtual assistants. In particular, the deployment of edge computing architectures and digital twins has strengthened the connection between physical and digital systems [38,39,40,41], enabling the extraction of real-time insights from plant data [42]. These technologies allow an assistant like KobBot not only to query static information, but also to access updated operational data, thus improving its capacity for contextual interpretation and adaptive response.

2.4. Artificial Intelligence Applications in Industry

Beyond virtual assistants, artificial intelligence has become a core driver of industrial digital transformation [43], with applications ranging from predictive maintenance—using machine learning to anticipate failures and reduce downtime [2,44]—to the design of environmentally optimized concrete mixtures through generative and probabilistic models [45,46]. In structural monitoring, computer vision and deep learning have enabled automatic detection of cracks in concrete structures [27]. These advances support a more efficient, resilient, and data-driven industry, positioning AI as a key enabler of intelligent automation and decision-making [47,48].

2.5. AI and Digitalization in the Concrete Industry

The construction sector—and more specifically the concrete industry—has also begun exploring applications of Industry 4.0 [2,9,44], with a particular focus on artificial intelligence, although adoption remains lower compared to other sectors [17,40,49,50,51]. Models have been developed to optimize sustainable mixtures [46,52] and improve material performance [53,54], as well as to predict concrete properties using deep learning techniques [55,56].

In addition to the proposed models, several recent studies have provided empirical evidence of AI’s impact on real-world concrete production plants. Alshboul et al. demonstrated that implementing predictive maintenance models based on machine learning can significantly reduce unexpected failures in critical production equipment. In their study, the CatBoost algorithm achieved a prediction accuracy of 98.4%, improving operational reliability and reducing costs associated with downtime [2]. Držečnik et al. analyzed the use of digital twins in a medium-sized concrete plant, reporting substantial improvements in operational visibility, maintenance planning, and energy efficiency, which led to lower waste and safer operations [57]. Complementarily, Shahrokhishahraki et al. applied machine learning algorithms to optimize cement content in structural mixes intended for delayed stresses, achieving a 10% reduction in both cement usage and carbon emissions without compromising structural strength [58]. Finally, Liu et al. conducted a systematic review of AI’s impact on production, operations, and logistics management in modular construction, highlighting the effectiveness of these technologies in improving planning, traceability, and efficiency in industrialized environments comparable to concrete plants [59]. These studies reinforce the hypothesis that AI-based digital technologies can generate tangible, quantifiable benefits in operational efficiency, sustainability, and reliability in the concrete sector, fully aligning with the second objective of this research.

2.6. Gaps in Operational Digitalization

However, these applications have primarily focused on material design and analysis [60], without addressing the operational dimension of daily plant work or real-time assistance to operators. Some recent studies are beginning to identify this gap, noting that digitalization in concrete production still lacks tools oriented toward industrial system users [1,3].

Recent review studies emphasize that most “digital assistants” implemented in production and logistics remain focused on planning tasks or robotic system interaction, leaving a gap in situational support for plant operators [34,61]. Likewise, prototypes of Computerized Operator Support Systems (COSSs) show potential benefits in early fault detection and decision-making but remain scarce in validated cases within continuous process industries like concrete production [62]. In manual assembly environments, real-time visual assistance initiatives have shown improvements in cycle times and cognitive load, highlighting the opportunity to transfer similar approaches to concrete plants [63]. Overall, the literature agrees on the need to integrate assistants that combine contextualized access to plant data, predictive modeling, and a conversational interface capable of adapting to operating conditions—areas still largely unexplored in the construction sector.

3. Materials and Methods

This study proposes a technological implementation model aimed at optimizing operational management in concrete production plants through the incorporation of an intelligent virtual assistant. The model is framed within Industry 4.0 principles, promoting interoperability, modularity, decentralization, and user orientation [18]. Its practical application was deployed in a real industrial environment and structured into five main phases:

1.

Document Collection: A technical audit of the plant was conducted to gather all relevant documentation, including operating manuals, electrical diagrams, PLC configurations, alarm lists, and safety protocols. This documentation forms the technical knowledge base on which the AI system operates.

2.

Knowledge Vectorization: The collected documents were digitized and transformed into vector representations using semantic information retrieval techniques. These vectors, stored in a specialized database, enable high-precision contextual queries by the virtual assistant.

3.

Industrial Connectivity Configuration: Links between the KobBot system and the plant infrastructure were established through standard protocols such as OPC UA and MQTT. This enables key operational variables (e.g., temperatures, cycles, alarms, or machine states) to be interpreted by the system in real time.

4.

Deployment of the Virtual Assistant: The system was installed in the production environment, connecting to cloud-based language models (OpenAI, San Francisco, CA, USA) and accessing both plant data and the vectorized document base. The goal is to allow real-time contextual interaction with the operator.

5.

Validation Phase: A functional testing stage was carried out on site, where real operators interacted with the assistant in real situations. Use cases such as documentation queries, alarm interpretation, and incident resolution were analyzed. The results of this phase enabled iterative adjustments before the final deployment.

This model, which is replicable and scalable, allows the technological solution to be adapted to different concrete production plants with minimal adjustments. It also promotes the structuring of existing operational knowledge, making it accessible and interpretable by AI systems in line with a progressive operator-centered digitalization strategy.

3.1. Real Plant Analysis

The real plant under analysis is a concrete batching and mixing facility designed for medium- to large-scale projects, characterized by robust production capacity and efficient space utilization. The system is equipped with a twin-shaft mixer model FTS 4000/3000 (Frumecar, Murcia, Spain), which ensures a homogeneous and high-quality mixture, delivering a production capacity of $3 {\mathrm{m}}^3$ per cycle.

The aggregate storage system consists of four square-arranged hoppers with a total capacity of $30 {\mathrm{m}}^3$, expandable up to $80 {\mathrm{m}}^3$, which feed the mixer through an elevation conveyor belt. Cement dosing is handled by a scale with a capacity of $1500 \mathrm{kg}$, complemented by a water scale of $750 \mathrm{L}$ and an additional water dosing system with a flowmeter to increase precision in mix composition.

The water installation includes a $5.5 \mathrm{kW}$ pump and regulation system, while the pneumatic installation integrates a $7.5 \mathrm{kW}$ compressor to operate valves and gates. Overall, the plant requires a surface area of only $200 {\mathrm{m}}^2$, optimizing layout and facilitating integration into production sites. The total power consumption of the plant is $150 \mathrm{CV}$ ($111.85 \mathrm{kW}$).

This configuration demonstrates a balanced design that combines compactness with high operational performance, allowing continuous and controlled production of concrete. Such features make the plant suitable for demanding environments where both productivity and consistency of the mixture are critical, thereby serving as a representative case study for evaluating the impact of intelligent virtual assistants in real industrial scenarios.

3.2. General System Architecture

The KobBot system was designed with a modular and scalable architecture, structured around decoupled design principles and concurrent thread execution. This architecture allows for progressive integration of new functionalities and easy adaptation to different industrial environments and ensures high maintainability. Each component operates autonomously, communicating with others through controlled events, which reduces module dependencies and simplifies development and debugging.

The system’s core is a graphical user interface developed in Python 3.11 (Python Software Foundation, Wilmington, DE, USA) using the PyQt5 framework (Riverbank Computing, Trowbridge, UK). This interface supports both text and voice command input and provides visual and auditory feedback. It was designed considering operator needs in industrial environments, focusing on readability, ease of use, and clarity of displayed information. The adopted design pattern, based on the Model–View–Controller (MVC) model, keeps the presentation logic, event management, and interaction with AI and plant systems separate.

The system is organized into three functional layers. The presentation layer corresponds to the GUI and manages direct user interaction, displaying the query history and allowing switching between input modes. The intermediate processing layer orchestrates the system’s functionality, coordinating the voice recognition (ASR), natural language understanding (NLU), document access, and remote service connection modules. This layer ensures efficient and secure data transmission and that returned responses are properly integrated into the interface. Lastly, the service and connectivity layer establishes links with plant infrastructure via industrial protocols such as OPC UA and MQTT and connects with cloud-based language models to perform advanced semantic inference.

All layers are implemented with concurrent threads, allowing the system to remain fluid even under high operational load or network latency conditions. Automatic reconnection mechanisms and temporary local storage have been incorporated to ensure safe operation in environments with intermittent connectivity.

In addition to its technical modularity, the architecture was designed to facilitate future extensions. This allows for the integration of complementary modules such as computer vision systems, predictive analytics, or integration with digital twin platforms, as seen in the implementation environment shown in Figure 1. Thanks to this flexible structure, KobBot not only fulfills operational assistance functions but can evolve into a central component of distributed intelligence on the shop floor, aligned with Industry 4.0 digitalization principles and the operator-centered vision of Industry 5.0.

3.3. Functional Components of the System

The KobBot system is structured around a set of functional components that work in a coordinated manner, as detailed in Figure 2, to provide contextualized, real-time assistance to the plant operator. These components were designed to ensure a smooth user experience, facilitate integration with existing industrial infrastructures, and enable efficient management of technical information. The functional modularity allows the system to be adapted and scaled according to the specific needs of each installation, maintaining independence between the interface, language processing, and industrial connectivity.

• Graphical User Interface (GUI): This provides a chat-like conversational environment with visual controls, real-time feedback, and direct connectivity with voice and AI modules.
• Speech Recognition and Synthesis: Using Google Speech Recognition and gTTS (Google LLC, Mountain View, CA, USA), the system allows voice input and generates auditory responses, facilitating usage in noisy or hands-on environments.
• Cloud Processing (OpenAI): Operator queries are processed using cloud-based natural language models (OpenAI, San Francisco, CA, USA). These models have access to vectorized documentation and process data to deliver contextualized and accurate responses.
• Industrial Connectivity: Integration with plant systems is achieved through

  –

  OPC UA for real-time data reading from SCADA and PLCs;

  –

  MQTT for efficient and lightweight communication with distributed devices.

The separation between local logic (interface, communication, and events) and remote logic (language processing and document retrieval) allows the system to operate with low computational resource consumption and high adaptability to new installations. This modular design is key to agile deployment in different industrial environments, aligning with the principles of efficiency, customization, and human–machine interaction promoted by Industry 4.0.

4. Results

The evaluation of the KobBot system was carried out through a multidimensional approach, aimed at validating not only its technical performance but also its ability to integrate effectively into the real operational environment of a concrete production plant. The tests were designed to analyze critical aspects of the assistant’s functionality, from speech recognition accuracy to the relevance of technical responses, also considering direct user perception.

This process not only confirmed that the design objectives were met but also helped identify opportunities for improvement and refining the system’s deployment in real industrial scenarios.

4.1. Speech Transcription Accuracy

One of the first components evaluated was the speech recognition module, which is key to ensuring smooth and frictionless interaction. To assess this, the similarity between transcribed phrases and their original equivalents was measured, considering both lexical and semantic fidelity. The results, shown in Figure 3, reveal a generally positive trend, with high accuracy levels in most transcriptions. In the figure, individual test results are represented as discrete points, while blue dashed lines indicate the average performance across all evaluations. This performance confirms the system’s robustness in environments where voice input can offer a significant advantage over traditional interaction methods.

4.2. Technical Responses with Document Access

The second analysis block focused on validating one of KobBot’s key features: its ability to interpret complex technical queries and extract specific information from manuals and maintenance documents. To achieve this, the assistant’s responses were compared to the original document content, evaluating their similarity, relevance, and technical accuracy. As shown in Figure 4, the results were highly satisfactory, demonstrating that the assistant not only understands the context of the questions but is also able to deliver answers aligned with defined technical procedures.

This performance is especially valuable in operational scenarios where response time and information accuracy are critical to production continuity and operational safety.

4.3. Responses to General Questions

In addition to technical analysis, the assistant’s ability to answer general questions not necessarily linked to documents or specific operational contexts was evaluated. This scenario helps assess the system’s adaptability to less structured inputs. Figure 5 shows the distribution of similarity levels achieved, with an overall average of 81% and a low incidence of responses below 70%. While the results are satisfactory, room for improvement was identified in the interpretation of more ambiguous or open-ended queries. In the figure, individual test results are represented as discrete points, while blue dashed lines indicate the average performance across all evaluations.

4.4. User Experience Evaluation

Finally, a qualitative evaluation was conducted through a survey administered to real users in the plant environment. The goal was to gather first-hand impressions about the assistant’s usefulness, usability, and impact on daily operational workflows.

A total of 30 participants completed the survey. These participants were selected from workers at production plants where the assistant was actively in use, ensuring a representative sample based on job roles, experience levels, and frequency of interaction with the system.

The evaluation consisted of ten closed-ended statements, each rated on a Likert scale from 1 (strongly disagree) to 5 (strongly agree), allowing participants to express their level of satisfaction with different aspects of the assistant’s performance. The specific items presented were as follows:

• “The voice assistant provides accurate and useful answers to resolve my technical queries”.
• “Interaction with the voice assistant is intuitive and easy to carry out in my daily work”.
• “The assistant responds quickly to my queries, facilitating the continuity of operations”.
• “The information provided by the assistant is relevant and appropriate for the technical tasks I perform”.
• “The voice assistant operates reliably, without significant errors or technical problems during use”.
• “The assistant is able to correctly interpret unclear or ambiguous questions”.
• “The assistant functions properly under real plant conditions, such as noisy environments or interruptions”.
• “I feel comfortable interacting with the voice assistant compared to other consultation methods”.
• “The voice assistant is a useful tool for learning new tasks or understanding technical processes”.
• “My overall experience using the voice assistant has been satisfactory, and I believe it improves my work in the plant”.

The evaluation covered ten key dimensions, with an overall average score of 4.04 out of 5, indicating a high degree of user satisfaction. The results by dimension are as follows:

• Technical accuracy and usefulness (4.73): Provides effective answers and resolves complex queries.
• Usability (4.60): Intuitive and user-friendly interaction, supported by a smooth interface.
• Response time (4.17): Quick responses that minimize interruptions.
• Information relevance (4.37): Offers precise and useful data for technical tasks.
• Reliability (4.33): Stable system with no significant errors.
• Query interpretation (4.07): Good handling of ambiguous queries, with room for improvement.
• Adaptability (3.43): Performs reasonably well in noisy industrial environments.
• Interaction comfort (3.37): Considered a viable alternative to traditional methods.
• Learning tool (3.90): Useful for understanding technical processes and learning new tasks.
• Overall satisfaction (4.23): Significant improvement in plant operations.

Figure 6 summarizes the survey results, highlighting the users’ positive perception of the system, especially regarding technical accuracy (4.73) and usability (4.60), which supports the effectiveness of the user-centered design approach. The figure displays a radar chart where the blue line represents the average score for each evaluated dimension, and the red shaded area encompasses the range of responses, with its outer and inner edges corresponding to the maximum and minimum values, respectively. Statistical validation through one-sample t-tests confirms that all metrics significantly exceed the neutral baseline (p < 0.05), with technical accuracy (t = 47.37), usability (t = 36.52), and learning effectiveness (t = 30.79) demonstrating particularly strong statistical significance (p < 0.001). While dimensions such as query interpretation (4.07) and adaptability to noisy environments (3.43) indicate areas for improvement, their statistical significance (t = 9.76 and t = 6.21, respectively) confirms that even these lower-rated aspects perform meaningfully above neutral expectations, guiding future iterations to ensure the assistant continues evolving in line with real operational needs. The statistical details corresponding to the chart are provided in

Table 1. One-sample t-test results for user experience metrics (n = 30, test value = 3.0).

Metric	Mean ± SD	t-Statistic
Technical accuracy	4.73 ± 0.20	47.37 ***
Usability	4.60 ± 0.24	36.52 ***
Response time	4.17 ± 0.67	9.59 ***
Information relevance	4.37 ± 0.50	15.01 ***
Reliability	4.33 ± 0.49	14.87 ***
Query interpretation	4.07 ± 0.60	9.76 ***
Adaptability	3.43 ± 0.38	6.21 ***
Interaction comfort	3.37 ± 0.77	2.63 *
Learning tool	3.90 ± 0.16	30.79 ***
Overall satisfaction	4.23 ± 0.25	26.89 ***

Significance levels: * p < 0.05, *** p < 0.001 (df = 29).

.

4.5. Before-and-After Analysis in the Plant

To better understand the real impact of the AI-based assistant on daily operations, a comparative analysis was conducted in a representative concrete plant before and after its deployment. The analysis focused on quantifiable performance indicators such as the frequency of operational incidents, energy consumption, material waste, and estimated environmental impact. This approach not only allows us to assess technical improvements but also to infer broader implications regarding sustainability, cost-efficiency, and workplace well-being.

4.5.1. Incident Rates

Before the implementation of the assistant, concrete plants experienced a considerable number of recurring incidents related to automation, instrumentation, and operational procedures. Although not always critical, these problems frequently caused production interruptions and placed an additional burden on operators, who often had to consult extensive manuals or rely on specialized technical support. With the introduction of the assistant, capable of providing immediate responses based on digitized manuals and standardized protocols, a significant reduction in the frequency of these incidents was achieved, optimizing response times and improving operational continuity. This reduction is quantitatively illustrated in

Table 2. Comparison of maximum monthly incidents before and after the implementation of the assistant.

Type of Incident	Before Assistant	After Assistant
PLC/HMI failures	2	1
Software/ERP crashes	3	2
Scale communication errors	3	1
Scale miscalibration	3	1
Moisture sensor failures	3	1
Non-conforming mix quality	4	2
Hopper blockages	3	1
Returns due to setting time	3	1
Total maximum	24	10

, which compares the maximum number of monthly incidents reported before and after the assistant’s deployment.

The comparison shows that the assistant has significantly reduced the number of incidents, decreasing from a maximum of 24 to 10 per month. While it does not completely eliminate incidents—some of which are due to unavoidable mechanical factors—it has achieved a reduction of approximately 58% in those associated with procedural errors, delayed diagnostics, or lack of knowledge of protocols. This improvement has led to greater operational efficiency and reliability in the plant.

4.5.2. Environmental and Operational Impact Estimation

Beyond reducing the frequency of incidents, the implementation of the virtual assistant has had measurable effects on energy efficiency and environmental sustainability. In a concrete production plant, each incident typically results in approximately 30 min of inefficient operation due to equipment stoppages or batch reprocessing.

According to the actual plant configuration detailed in Section 3.1, the total installed power is 150 CV, equivalent to approximately $111.85 \mathrm{kW}$. Based on this figure, each incident is estimated to result in an energy loss of around $55.93 \mathrm{kWh}$ ($0.5 \mathrm{h}\times 111.85 \mathrm{kW}$).

Taking into account the monthly reduction from 24 to 10 incidents—a net decrease of 14—we estimate the following operational and environmental benefits, as detailed in

Table 3. Before-and-after comparison: energy, cost, and emissions.

Metric	Before Assistant	After Assistant
Estimated monthly incidents	24	10
Energy waste (kWh/month)	1342	559
Cost of energy waste (EUR)	201	84
CO2 emissions (kg/month)	309	129
Reduction (%)	-	58%

, which compares energy waste, costs, and emissions before and after the assistant’s implementation.

In addition to reducing energy waste, the assistant helps decrease raw material losses from non-conforming mixes or setting time errors. This indirectly contributes to more efficient use of cement, water, and aggregates—resources with high environmental and economic costs.

Furthermore, the assistant supports a safer and more inclusive working environment by reducing cognitive overload during high-stress situations and enabling less experienced personnel to make informed decisions, aligning with the human-centered vision of Industry 5.0.

5. Discussion

The implementation of KobBot in a live production environment constitutes a concrete step toward the realization of Industry 5.0 principles in heavy industry. Unlike conventional automation solutions, the assistant integrates artificial intelligence (AI), voice interaction, and contextual information access to support human-centered digitalization—a transition advocated in the recent literature on cyber–physical systems and smart construction technologies [4,5].

Previous studies have highlighted the persistent gap between digital innovation and field-level adoption in civil and concrete engineering [1,3]. KobBot addresses this challenge by offering an intuitive interface that connects operators with structured documentation, operational data, and real-time system feedback. This aligns with the shift toward individualized and operator-centered support systems discussed by [6].

Beyond usability, the assistant contributes to the evolving role of AI in sustainable development. As pointed out in the systematic review by [50], the influence of AI in civil engineering includes not only productivity but also sustainability, interconnectivity, and learning support. KobBot exemplifies this by promoting safer working conditions, reducing dependence on tacit knowledge, and fostering informal, on-demand learning within the plant environment.

Moreover, the implementation strategy proposed in this work aligns with the modular and scalable architectures recommended in the industrial AI literature [8,9]. The system was designed to be integrated with existing SCADA and ERP systems, reducing friction and avoiding costly overhauls. This pragmatic approach is essential in the context of SMEs, where budget and personnel constraints limit the feasibility of disruptive transformation projects [1].

Despite these advantages, limitations remain. Acoustic noise in the production environment still affects the assistant’s robustness—a known barrier in industrial voice applications [8]. Additionally, the reliance on external cloud services, although practical, raises valid concerns regarding data sovereignty and cybersecurity, especially in critical infrastructure scenarios [4].

In terms of broader societal impact, the system aligns with Sustainable Development Goals (SDGs), particularly SDG 8 (Decent Work), SDG 9 (Industry and Innovation), and SDG 12 (Responsible Consumption and Production), as discussed in [1,50]. The assistant promotes responsible knowledge usage, reduces incident risk, and empowers less-experienced operators, all of which contribute to more sustainable and resilient industrial ecosystems.

Future directions for research and deployment include the integration of multimodal capabilities such as computer vision [60], predictive maintenance models [11], and real-time digital twins [28]. These could extend the assistant’s role from passive information retrieval to proactive system optimization. Furthermore, hybrid cloud/on-premise architectures could help address current limitations while ensuring compliance with cybersecurity standards and operational independence.

In summary, this work confirms that intelligent virtual assistants can serve as key enablers of Industry 5.0 in the concrete sector, bridging human expertise with digital capabilities in a way that is scalable, sustainable, and operator-focused.

6. Conclusions

This study has addressed the key challenges of digital transformation in the concrete industry by designing and validating KobBot, an AI-based virtual assistant tailored to optimize operator interaction with industrial systems. Framed within the Industry 4.0 paradigm and aligned with the human-centered principles of Industry 5.0, the system responds to limitations such as low digitalization, reliance on manual processes, and fragmented access to technical knowledge.

The assistant demonstrated strong performance in real plant conditions, achieving high accuracy in speech recognition and technical responses and significantly reducing incident rates by 58%. These improvements translated into measurable operational benefits, including an estimated monthly saving of 783 kWh of energy, EUR 117 in cost, and 180 kg of CO₂ emissions—highlighting its potential to support more efficient, sustainable production.

Beyond technical gains, KobBot fosters safer and more inclusive workplaces by lowering the cognitive burden on operators and facilitating decision-making for personnel with varied experience levels. Its modular, interoperable architecture and capacity for contextual understanding position it as a scalable solution for smart, resilient industrial environments.

Ultimately, this work demonstrates that virtual assistants can play a transformative role not only in improving operational efficiency, but also in promoting responsible resource use, workplace well-being, and sustainable innovation in the concrete sector.