Integrated system for the intelligent optimization of production and maintenance processes and the creation of digital super-technicians

Abstract

The invention relates to an integrated system that combines an advanced hardware-software system to optimize production and maintenance processes, supporting the creation of digital super-technicians. A network of sensors acquires operational data from machines and plants, transmitting it to a local server, which performs initial processing before transferring it to a cloud infrastructure for advanced management and self-learning. Operators interact with the system through a plurality of user interfaces, receiving instructions and providing contextual input. An advanced software module, based on industrial artificial intelligence algorithms processes the data using a multimodal approach, classifying anomalies and determining whether to act autonomously or involve experts through a Human-ln-The-Loop (HITL) model. A generative Al agent continuously refines the analyses, while a multi-technique diagnostic engine detects faults using advanced technologies. Finally, a Digital Twin unit compares real and virtual machines, automatically optimizing operational parameters to improve efficiency and reliability.

Description

[0001] Title: "INTEGRATED SYSTEM FOR THE INTELLIGENT OPTIMIZATION OF PRODUCTION AND MAINTENANCE PROCESSES AND THE CREATION OF DIGITAL SUPER-TECHNICIANS”

[0002] DESCRIPTION FIELD OF THE INVENTION

[0003] The present invention relates to an integrated hardware-software system for the intelligent optimization of production and maintenance processes and the creation of digital super-technicians.

[0004] PRIOR ART

[0005] As is known, predictive maintenance is a type of preventive maintenance carried out following the detection of one or more parameters, which are measured and processed using appropriate mathematical models in order to determine the remaining time before failure.

[0006] In recent years, strongly driven by Industry 4.0, intensive activities have developed aimed at collecting data and information from components and machinery.

[0007] The problems related to the collection and processing of such data are manifold. It is sufficient to mention that data are made available in various formats and therefore are not coordinated with each other, that the data often remain isolated, and that, in practice, the data flow frequently stops at simple visualization and consultation.

[0008] All of this leads to difficulties in ensuring accurate and reliable analysis of operational data in industrial contexts, increasing the risk of errors and negatively impacting the ability to predict and interpret data.

[0009] Furthermore, it is known that the advent of generative artificial intelligence has enabled advanced automation of analysis and decision-making but has also introduced new challenges related to the management of sensitive data and their security.

[0010] In particular, the use of generative models on cloud or shared platforms may expose critical information to risks of unauthorized access or data leakage.

[0011] There is therefore a need to protect sensitive corporate data during the processing and use of generative artificial intelligence, avoiding risks of exposure to external systems. In addition, the evolution of Al on Edge, i.e., the execution of artificial intelligence (Al) algorithms directly on edge devices, has opened new possibilities for local data processing, reducing dependence on the cloud and improving system responsiveness. However, this technology requires optimized hardware capable of ensuring rapidprocessing times without compromising energy efficiency or operational safety. There is thus a need to develop devices capable of supporting advanced algorithms locally, ensuring high performance and reliability in industrial processes.

[0012] Among the objectives of the present invention, the following can be included:

[0013] To develop highly specialized technical intelligence capable of effectively and efficiently enhancing operational and production processes.

[0014] To improve the accuracy of analyses and predictions in industrial contexts, enabling effective interaction between automated systems and human operators to refine anomaly classification and optimize the solutions adopted.

[0015] To ensure the protection and confidentiality of industrial data, preventing their exposure to external systems while simultaneously providing efficient processing of information to support operational activities.

[0016] To increase the efficiency and autonomy of industrial processes, reducing the need for dedicated resources through the use of advanced systems for analysis, learning, and operational optimization.

[0017] Other objectives and advantages of the invention will become apparent from the following description.

[0018] BRIEF SUMMARY OF THE INVENTION

[0019] These and other objectives are achieved by an integrated hardware-software system for the intelligent optimization of production and maintenance processes and the creation of digital super-technicians, wherein the said system includes at least:

[0020] a network of sensors configured to acquire operational data relating to the operation of machines and industrial plants;

[0021] one or more local servers and / or a cloud infrastructure set up for the collection, processing, and structuring of the data acquired by the sensors;

[0022] a plurality of user interfaces to interact with the system, provide contextual input, receive operational instructions, and access processed information;

[0023] an advanced software module, based on industrial artificial intelligence algorithms, configured to:

[0024] a) process the acquired data using a multimodal approach combining interpretive, predictive, and generative Al techniques;

[0025] b) classify the detected patterns distinguishing between normal operating conditions, critical faults, and moderate deviations in order to determine the level of intervention required;c) activate a Human-ln-The-Loop (HITL) model for the involvement of practitioners and experts if pattern classification requires human analysis;

[0026] d) generate, in the case of Al decision autonomy, predictive operational actions for the optimization of industrial processes;

[0027] a generative artificial intelligence agent operating within the company perimeter, configured to iteratively learn from collected operational data and HITL interactions, providing technical support, document generation, reporting, and continuous training to operators;

[0028] a multi-technique analysis engine for predictive maintenance, which integrates diagnostic techniques to detect faults and optimize machine performance;

[0029] a self-learning architecture configured to collect feedback from the execution of operational actions, update analysis models, and refine the system's predictive capabilities;

[0030] a Digital Twin unit, configured to compare real operating conditions with simulated ones, enabling the automatic optimization of industrial machine operating parameters, where the advanced software module processes data acquired from the sensor network through the multi-technique analysis engine, dynamically interacting with the local server and / or cloud infrastructure and user interfaces to generate predictive actions, while the generative artificial intelligence agent contributes to the continuous optimization of operations through an iterative retaining process and updating of technical knowledge.

[0031] In particular, the artificial intelligence agent is proactive, meaning that it can directly intervene in feedback mechanisms even without human intervention by modifying the actions to be carried out. Furthermore, the artificial intelligence agent can also make requests to an operator, for example, if it detects anomalies or non-standard patterns in the examined data sets, for instance, in order to “understand” the events by evaluating them in the operational context, to “retain” them for future assessments, and so on. Conversely, an operator can always query the artificial intelligence agent for any type of analysis, evaluation, or task, thereby establishing a truly bidirectional relationship between the operator and the assistant. It should also be noted that, more generally, one of the functionalities of the artificial intelligence is causal inference for the determination of cause-effect relationships. This functionality is applicable to the Digital Twin to simulate and subsequently decide whether to implement thehypothesized scenarios, or simply for cause-effect research to determine the reasons for problems, inefficiencies, or unmet targets.

[0032] In addition, it should be noted that the presence of even a remote cloud does not imply exposure of the data to third parties, since delocalized private clouds can be used; however, it is important that the data remain protected and unaltered or unalterable by external agents.

[0033] The present invention offers numerous advantages.

[0034] • Ease of installation and integration, being flexible and integrable with all existing hardware and software systems within the company to enable intelligent management.

[0035] • Capitalization of experience, where experiences are retained and made available to all resources involved in the production process, ensuring operational continuity and fostering collaboration.

[0036] • Universal asset management is enabled, allowing the management of all types of assets, whether recent or older, maximizing the effect of the smart factory in terms of performance and sustainability, while providing a holistic and ecosystemic view. • Elimination of the shortage of specialized technical personnel: SPS addresses the widespread issue of the lack of trained and specialized technical staff or high turnover, providing targeted operational support and advanced tools that optimize the use of available resources.

[0037] • Improved operational efficiency: SPS identifies anomalies and problems with greater precision and provides predictive operational solutions in real time, contributing to increased Overall Equipment Effectiveness (OEE) through targeted and timely interventions. SPS provides operational tasks to prevent issues and optimize performance. In this context, in the areas of production and quality, this also includes the failure to achieve a target, for example, when one or more expected performance levels are not met, with the aim of improving production performance through appropriate targeted corrective actions.

[0038] • Cost reduction: It optimizes maintenance planning, limits premature component replacements, and increases performance.

[0039] • Technical reliability: It combines advanced diagnostic techniques, academic knowledge, operational technical experience, and the operational context to provide more robust prediction and management.• Sustainability: Prevention of sudden failures, reduction of waste, and energy optimization. Furthermore, it makes operator activities more sustainable through the continuous and automatic generation of insights and tasks.

[0040] The invention represents a significant advancement over the prior art, leveraging cutting-edge technologies for the intelligent management of machines and plants, keeping humans at the center and providing a unique and adaptable solution.

[0041] Additional features and advantages of the invention can be inferred from the dependent claims.

[0042] BRIEF DESCRIPTION OF THE FIGURES

[0043] Additional features and advantages of the invention will become apparent from the following description, provided by way of example and without limitation, with reference to the figure illustrated in the accompanying drawing, wherein:

[0044] - Figure 1 is a block diagram illustrating the operation of the invention.

[0045] DETAILED DESCRIPTION OF THE FIGURES

[0046] The present invention represents a significant evolution of a previously patented system, expanding its functionalities and introducing unique features that enhance its effectiveness and adaptability. The system, also referred to as SPS - Smart Predictive System, combines advanced hardware and software to generate, acquire, process, analyze, and predict data relating to machines and industrial plants.

[0047] The invention primarily functions to support operators in four strategic operational areas: predictive maintenance, production / process, quality, and energy efficiency. Through a sophisticated industrial artificial intelligence system, the system issues targeted operational tasks, providing operators with practical guidance to optimize processes and prevent critical issues. Its ability to aggregate data and transform it into concrete actions makes it an indispensable tool for improving overall efficiency (OEE) and ensuring operational continuity. The new system configuration is based on a holistic and ecosystemic vision, oriented toward human-machine integration to promote the smart factory. On the human side, it integrates academic knowledge with operational experience, creating a shared heritage that enriches both human resources and the company. On the machine side, it facilitates the integration of both new and legacy systems to maximize benefits and support sustainable and efficient production. Knowledge, experience, and operational data merge to create corporate know-how and sustainable operations in predictive maintenance, production process optimization, quality management, and energy efficiency, offering innovative solutionsthat combine advanced analysis and prediction techniques with intelligent operational support systems. This solution represents the corporate super-technician, always active and available to provide support. SPS makes machines intelligent while keeping humans at the center,

[0048] ensuring reliability and integrating with traditional corporate management systems, using available information to enable intelligent, proactive, and predictive management.

[0049] The invention will be described below with particular reference to Figure 1 , noting that the invention utilizes the following resources.

[0050] Advanced Software: The software of the present invention represents the core of the system, based on industrial artificial intelligence algorithms selected from standard libraries and internally optimized / configured for specific operational requirements. It integrates interpretive, predictive, and generative approaches to analyze and process the collected data and information, transforming them into concrete and customized actions.

[0051] Interaction with the HITL (Human-ln-The-Loop) system ensures constant validation of analyses, improving the reliability of diagnostics and predictions.

[0052] As is known, the "HITL system" refers to the "Human-in-the-Loop" model, where human intervention is integrated into the automated process or artificial intelligence system. In a HITL context, the human operator can monitor or supervise processes in real time, enhancing precision, reliability, and overall outcomes. It is particularly useful in applications where the machine requires assistance to make decisions in complex or unforeseen situations beyond the initial programming.

[0053] The private generative Al component, named "Franco," operates exclusively within the SPS (or company) perimeter, preserving the security of corporate data and providing comprehensive support through technical responses, reporting, and continuous training for operators.

[0054] The invention also employs Advanced Hardware designed to ensure maximum flexibility and compatibility with a wide range of industrial contexts.

[0055] Such hardware includes a data concentrator and a series of accessories, which integrates with existing or newly installed sensors and connects to machine controllers to create a complete operational context.

[0056] This allows the generation, collection, and processing of data and information using a non-invasive approach, maximizing energy efficiency and improving fault diagnosis.The hardware is available in multiple optimized versions, prepared for Al on Edge, to ensure fast and secure local processing.

[0057] More specifically, the invention integrates an intelligent, multi-technique solution for predictive maintenance, encompassing a wide range of advanced diagnostic techniques, including Anomaly Detection, MCSA / ESA, vibration analysis, ultrasound, oil and lubricant analysis, motion amplification, dispersed magnetic flux, and others. These techniques can be combined to ensure precise and reliable diagnostics, supporting predictive maintenance and improving operational efficiency. In particular, the MCSA / ESA technique allows precise and reliable diagnostics combined with energy efficiency, through a simple and non-invasive installation. SPS multi-technique predictive maintenance is applicable to both fixed and variable operating machines and plants, thanks to its ability to learn behaviors (machine learning) and classify operating conditions.

[0058] The invention provides four operational systems in a single solution, offering a unified approach that covers the four strategic operational areas: predictive maintenance, production / process, quality, and energy efficiency.

[0059] This integration within a single platform ensures a continuous and complete flow of information and operational solutions, contributing to the optimization of business processes and sustainability. Furthermore, this approach allows for the extraction of maximum value from data without replicating or multiplying hardware and software infrastructures. Correlations between information and the operational context make the system reliable and effective.

[0060] The HITL system active in the present invention represents a bidirectional approach in which operators interact with the industrial Al to validate and refine analyses. This process allows the collection of real-time feedback, improves anomaly classification, and customizes solutions according to the operational context,

[0061] combining academic knowledge with technical experience. The integration between HITL and the software of the invention ensures analytical robustness, making predictions and diagnostics highly reliable. The network and information are managed through automated processes that minimize the need for dedicated resources.

[0062] The invention uses artificial intelligence algorithms appropriately selected from standard libraries and internally configured, leveraging interpretive, predictive, and generative approaches. This technical assistant, based on private generative Al and trained in industrial and manufacturing technical and production topics, operatesexclusively within the perimeter of the invention, ensuring data security, corporate privacy, and the reliability of the information it manages and / or generates.

[0063] For the sole purpose of simplifying the description, this virtual technical assistant will hereinafter be referred to as FRANCO. It capitalizes on the company's experiences, data, and operations, providing technical responses, reports, and on-the-job training. Furthermore, it supports the integration of new resources and offers operational flexibility, adapting to specific needs and continuously improving through interaction with the HITL system.

[0064] The integration of Franco, that is, the generative artificial intelligence (GenAI) based on an open-source LLM, represents a key element within the SPS system of the invention, contributing to the development of advanced technical intelligence in support of industrial operations.

[0065] This creates a system capable of enhancing operational and production processes with a level of automation and optimization never previously achieved. For this reason, the SPS system of the invention has been designed as an integrated ecosystem that combines advanced hardware, capable of generating and acquiring data streams through dedicated accessories, with HITL (Human-ln-The-Loop) technology, which enables dynamic interaction between operational data and human knowledge. At the core of the system is a structured knowledge base that combines real operational data with human expertise, and a self-learning loop designed to continuously improve the predictive and operational capabilities of the system.

[0066] Within this architecture, “Franco” plays a crucial role as a multi-modal technical artificial intelligence. Its operation is based on the ability to reprocess and contextualize technical information, leveraging an open-source LLM model that is continuously updated and optimized in technical and production domains. Through a dynamic learning process, Franco integrates operational data acquired from the SPS hardware and information derived from HITL interactions, thereby building a highly specialized knowledge base. A key aspect is continuous retaining, which ensures that technical skills remain up to date and can quickly adapt to changes in production processes, maintaining knowledge always aligned with operational requirements.

[0067] The integration of Franco delivers tangible operational benefits, translating into direct support for operations, particularly in multi-technique predictive maintenance, production process optimization, quality improvement, and scrap reduction, as well as significant energy efficiency gains. This innovation not only automates themanagement of technical knowledge but also enhances operational accuracy, reduces response times, and ensures uninterrupted production continuity without the need for manual updates to technical skills.

[0068] It is therefore essential to highlight the iterative retaining methodology, which allows Franco to learn from operational technical data and human knowledge, ensuring continuous evolution of its capabilities. Integration with the HITL system and the selflearning loop represents a distinctive element, guaranteeing that the developed technical intelligence is constantly updated and refined through experience. Furthermore, the use of an open-source LLM in the industrial domain constitutes a significant innovation, as it enables the creation of highly specialized and adaptable technical intelligence for operations.

[0069] The combination of these elements provides the system with a significant competitive advantage, making this approach unique within the Industry 4.0 and 5.0 landscape. The introduced innovation not only improves the management of industrial operations but also establishes a new standard in the use of artificial intelligence for maintenance, process optimization, quality, and energy efficiency, defining a scalable model applicable to diverse production contexts.

[0070] In the implementation detail, the algorithm illustrated in Figure 1 begins with a data collection and pre-processing phase (block 10), during which information originating from or generated by sensors and industrial systems (block 20) is acquired, cleaned, and standardized to ensure consistency and reliability in the subsequent analysis. Naturally, the term “sensors” in the context of the present invention can also be understood as including computer vision techniques as an evaluation element, which may complement or replace numerical data through various methods, for example by detecting vibrational phenomena through image motion analysis or by analyzing thermographic images.

[0071] Subsequently, an artificial intelligence (Al) module processes the collected data, applying predictive and interpretive models to identify possible anomalies and assign a confidence level to the results obtained (block 30).

[0072] Once anomalies are identified, the system proceeds with the classification of the associated patterns (block 40), distinguishing between critical events, transient irregularities, and normal operating conditions, in order to determine the required level of intervention.At this point, the system makes an operational decision: it determines whether the detected anomaly can be managed autonomously by the Al, that is, if it involves known patterns (block 50), or if human intervention via the Human-ln-The-Loop (HITL) model is required, for example in the case of unknown patterns (block 60).

[0073] If the system opts for expert involvement, the expert analyzes the problem, corrects any classification errors, and provides feedback, thereby improving the accuracy of the Al model for future processing.

[0074] This can be carried out by internal operators (block 70) or, in more complex cases, with the support of external experts and academics (block 80).

[0075] If the anomaly can be managed autonomously, the Al suggests possible solutions (block 90), simulates their impact on the system, and proposes the most effective actions to resolve the issue.

[0076] Once the solution is determined, the system proceeds with executing the corrective action, updating operational parameters and recording the intervention in the corporate database to ensure traceability and improve future analyses.

[0077] Finally, a feedback and continuous learning process is initiated (block 100), in which data from the intervention are integrated into the Al model, refining predictive capabilities and making the system increasingly precise and efficient over time.

[0078] The cycle concludes, ready for a new iteration in monitoring and optimizing industrial processes.

[0079] APPLICATION EXAMPLE 1

[0080] The system of the invention, integrated with advanced artificial intelligence and predictive diagnostic technologies, was applied at Montello Spa, a company operating in the recycling sector, to monitor the conditions of a 500 kW electric motor connected to a densifier.

[0081] One of the key techniques employed is Electrical Signature Analysis (ESA), which analyzes electrical signatures (current and voltage) to detect anomalies in machinery, ensuring:

[0082] Complete diagnostics on motors and generators

[0083] Non-invasive, without modifications to the equipment

[0084] Early detection of faults in bearings, windings, and mechanical imbalances Universal applicability to different types of industrial plants.

[0085] In particular, Electrical Signature Analysis (ESA) is one of the advanced predictive monitoring and diagnostic techniques that analyzes the electrical signatures (such ascurrent and voltage) of machinery and electrical equipment to detect anomalies and early-stage faults. This methodology provides a comprehensive overview of the operational conditions of electrical systems and the powertrain, optimizing maintenance and improving reliability.

[0086] Integrated with artificial intelligence solutions and production intelligence systems, such as SPS (Smart Predictive System), ESA not only allows the identification of faults but also suggests concrete corrective actions, contributing to improved overall efficiency, cost reduction, and ensuring sustainable and reliable production.

[0087] The installation of the system of the invention included the following components:

[0088] SPS-lbox-C

[0089] SPS-DAQ-P

[0090] SPS-CTVH-600-40

[0091] The Al of the system of the invention performed a Fast Fourier Transform (FFT) analysis on the data streams, identifying a severe anomaly in the bearing of the densifier motor. Following the diagnosis, the system of the invention generated an operational task for the replacement / overhaul of the motor.

[0092] Verification and Intervention

[0093] To confirm the system’s diagnosis, an on-site vibration analysis was conducted by a third-party company, which verified the problem. During the disassembly of the motor, severe damage to the bearing rollers occurred, as predicted by the system.

[0094] After the replacement, a new frequency analysis confirmed the resolution of the issue, and the system automatically closed the task, storing the experience to optimize future diagnoses.

[0095] In addition, the system of the invention can utilize an interactive Human-ln-The-Loop (HITL) approach, allowing human operators to:

[0096] Provide feedback on operating conditions via web and mobile interfaces Receive instructions and data from the system

[0097] Interact with the system through voice messages, images, and documents in various formats.

[0098] The system of the invention also makes use of a Generative Al, named Franco, that is a generative Al agent which:

[0099] Assists in operations by responding to technical inquiries.

[0100] Generates support documents.

[0101] Contributes to operator training.The integration of Electrical Signature Analysis (ESA) with the system of the invention has demonstrated the ability to detect faults in advance, reduce downtime, and improve operational efficiency. The case study confirms the value of Al-driven solutions in predictive maintenance, with a positive impact on reliability, sustainability, and industrial cost reduction.

[0102] In the present invention, a technical expert may, in order to meet contingent and specific needs, implement further modifications and variants, all of which remain within the scope of protection of the invention as defined by the following claims.

Claims

CLAIMS1. An integrated hardware-software system for the intelligent optimization of production and maintenance processes and the creation of digital supertechnicians, where the system includes at least- a network of sensors configured to acquire operational data relating to the operation of machines and of industrial plants;- one or more local servers and / or a cloud infrastructure set up for the collection, processing and structuring of the data acquired by the sensors;- a plurality of user interfaces to interact with the system, provide contextual input, receive operational instructions and access processed information;- an advanced software module, based on industrial artificial intelligence algorithms, and configured to:- a) process the acquired data using a multimodal approach combining interpretive, predictive and generative Al techniques;- b) classify the detected patterns, distinguishing between normal operating conditions, critical faults and moderate deviations, in order to determine the level of intervention required;- c) activate a Human-ln-The-Loop (HITL) model for the involvement of practitioners and experts, if pattern classification requires human analysis;- d) generate, in the case of Al decision autonomy, predictive operational actions for the optimization of industrial processes;- a generative artificial intelligence agent operating within the company perimeter, configured to iteratively learn from collected operational data and HITL interactions, providing technical support, document generation, reporting and continuous training to operators;- a multi-technique analysis engine for predictive maintenance, which integrates diagnostic techniques to detect faults and optimize machine performance;- a self-learning architecture configured to collect feedback from the execution of operational actions, update analysis models and refine the system's predictive capabilities;- a Digital Twin unit, configured to compare real operating conditions with simulated ones, enabling the automatic optimization of industrial machine operating parameters, where the advanced software module processes data acquired from the sensor network through the multi-technique analysis engine, dynamically interacting with the local server and / or cloud infrastructure and user interfaces to generate predictive actions, while the generative artificial intelligence agent contributes to the continuous optimization of operations through an iterative retaining process configured to automate the management of technical knowledge and keep system skills updated in alignment with the operational needs of the production processes.

2. System as at claim 1, in which the diagnostic techniques are chosen from Anomaly Detection, Electrical Signature Analysis (ESA), vibrational analysis, ultrasound, oil and lubricant analysis, motion amplification, dispersed magnetic flux and combinations of these.

3. System as in claim 1 , in which the advanced software module is configured to dynamically select and optimize industrial artificial intelligence algorithms from a standard library, adapting them to specific operational needs by means of machine learning and iterative learning techniques.

4. A system as in any of the previous claims, in which the sensor network includes vibration, thermal, acoustic, current, voltage, magnetic flux and chemical sensors for advanced detection and analysis of the operating conditions of industrial machines.

5. System as in any of the previous claims, in which the local server is configured to perform Al on Edge processing, reducing analysis latency and optimizing bandwidth management with the cloud infrastructure.

6. A system as in any of the previous claims, in which the advanced software module implements a Human-ln-The-Loop (HITL) model, configured to collect input from operators via web interfaces, mobile applications, voice messaging, HMIpages and document interrogation systems, improving anomaly classification and prediction reliability.

7. System as in any of the previous claims, in which the generative artificial intelligence agent is configured to operate in an environment isolated from the public cloud, guaranteeing corporate data security, operational privacy and protection of intellectual property.

8. System as in any of the previous claims, in which the multi-technique analysis engine for predictive maintenance is configured to dynamically combine different diagnostic techniques according to the type of machinery being monitored, optimizing the reliability of diagnosis by means of an adaptive information weighting system.

9. System as in any of the previous claims, in which the Digital Twin is configured to automatically update the operating parameters of the real machines, modifying the operating settings to minimize deviations between simulated and real behavior.

10. System as in any of the previous claims, in which the self-learning loop is configured to collect and process operational feedback generated by the execution of predictive actions, progressively improving the accuracy of the system and reducing the need for manual corrective action.

11. System as in any of the previous claims, wherein the advanced software module is configured to issue targeted operational tasks, including predictive maintenance suggestions, production process optimizations, quality improvement strategies and energy efficiency actions, ensuring proactive management of industrial operations.

Images