,allowExpansion)
AI-Powered Digital Twins for Maintenance, Anomaly Detection and Productivity
TOWARDS INTELLIGENT MANUFACTURING
The huge volume of data generated by industrial machinery presents a significant opportunity to unlock new business value. By harnessing the power of the Industrial Internet of Things (IIoT) and Artificial Intelligence (AI), companies can create intelligent ecosystems that not only monitor and control machinery but also proactively optimise processes, reduce costs, and foster innovation
The Pivotal Role of the Digital Twin
Digital Twins are virtual representations of a physical asset, process, or even an entire system. This technology serves as a bridge between the physical and digital worlds, enabling real-time monitoring, simulation, and control. Through the continuous collection of data from connected machinery, Digital Twins provide a dynamic and comprehensive view of operations. This allows for the derivation of new insights from the acquired data, the activation of actuators to perform actions in the real world, the automation of complex processes, and the simulation of "what-if" scenarios to test responses to potential disruptions without impacting live production.
Delivering Tangible Business Value
The integration of Digital Twins, IIoT, and advanced AI can accelerate the innovation of many manufacturing contexts. By improving the user experience for operators, enabling the prediction of anomalies to reduce waste, optimising processes to increase profitability, and extracting new value from existing machinery, these technologies are paving the way for a more proactive and efficient evolution for the industry. Reply’s approach not only enhances operational effectiveness but also opens up new avenues for servitisation, where the value proposition extends beyond the physical product to include a suite of connected services.
Modernising Legacy Systems Through Retrofitting
For legacy machinery, which often lacks innate connectivity, the practice of retrofitting offers a cost-effective pathway to integration within an intelligent IoT ecosystem. By augmenting older equipment with sensors and connectivity modules, it is possible to extract valuable data that would otherwise remain untapped. This data can then be used for a variety of applications, including the remote update of software and configurations, the collection of telemetry data, the monitoring of alarms and errors, and the tracking of energy consumption.
Leveraging AI for Proactive Operations
While the digital twin concept has been discussed in the manufacturing industry for years, the introduction of artificial intelligence has accelerated the use of this concept in different contexts. Now, the convergence of AI with a factory’s large output of data provides powerful tools for advanced analysis.
Machine learning algorithms can autonomously learn from data to identify patterns and make predictions without explicit programming. This is particularly valuable for tasks such as anomaly detection, where the goal is to identify outliers in data patterns that may indicate a developing issue, and predictive maintenance, which involves forecasting potential equipment failures to allow for pre-emptive servicing.
Another key application is cost forecasting, which enables the estimation of future operational costs based on historical data. The inherent scalability of these algorithms ensures that their performance improves with increasing data volumes, leading to the discovery of new patterns and improved adaptability. The predictive nature of these models facilitates a shift from reactive to proactive operational strategies.
Introducing Generative AI and Autonomous Agents
Generative AI, including technologies like Retrieval-Augmented Generation (RAG) systems and AI Agents, further enhances the capabilities of these intelligent ecosystems. AI workflows allow humans to define control logic that a large language model (LLM) can follow to resolve a given request. RAG systems are a specialised type of AI workflow that can retrieve information from a dedicated knowledge base, providing more contextualised and precise responses based on business- or project-specific data.
AI Agents represent a further leap in autonomy. Unlike AI workflows, AI Agents can independently and iteratively determine the optimal sequence of actions to achieve a specific goal, drawing upon a set of available tools. This dynamic and adaptable structure allows for a high degree of complexity to be managed.
Retrofitting a Legacy Pick-and-Place Machine
The pick-and-place machinery is used for placing electronic components onto a printed circuit board (PCB) and requires high speed and extreme precision. The project involved equipping the machine with sensors and connectivity to a cloud platform. A custom dashboard was developed to interact with the machine's Digital Twin, providing real-time visualisation of both raw sensor data and derived metrics, such as an estimated power value calculated by a linear regression model.
Furthermore, a RAG system was implemented to provide a conversational interface for accessing the machine's extensive manuals. This allows operators to ask questions and receive step-by-step instructions, complete with citations. The system also supports voice interaction, enhancing usability.
The architecture for this solution leverages a cloud-based platform, with a field gateway collecting telemetry data and transmitting it via MQTT to a data broker. The data is then processed through both hot and cold paths for real-time streaming and persistent storage in a specialised time-series database. The RAG component was built using Azure AI Studio with a GPT-4 model trained on the specific knowledge base of the pick-and-place machine. This architecture is designed to be cloud-agnostic, with the potential to be implemented using equivalent services from other providers.
The Generative Manufacturing Operations (GMO) Project
The project was undertaken to enhance the process of 3D printing metal parts. The project aimed to monitor and improve the printing experience for an EOS M 400 machine, which produces components from metal alloys. A key feature of this project is a sophisticated anomaly detection system designed to prevent costly failures by identifying prints that are likely to be defective.
The Digital Twin in this instance resides on a field gateway, which reads, preprocesses, and models data from the printer and its surrounding environmental sensors. This approach successfully identified distinct phases of the printing process, such as rest, heating, printing, cooling, and overheating. Rules were then established based on the duration of these phases; for example, a prolonged period of overheating is flagged as a potential anomaly. The anomaly detection model achieved an estimated accuracy of 70% on the available dataset.
The dashboard provides a comprehensive overview of the printing process and integrates two chatbots for a RAG system and an AI Agent. The RAG system, built on Amazon Bedrock Knowledge Bases, allows users to query documentation. The AI Agent, deployed locally, is equipped with tools to perform tasks such as retrieving print information and calculating energy consumption. The agent autonomously devises and executes a plan to fulfil user requests, iterating through different approaches until it arrives at a complete solution. It utilises a Common Objects in Context (COCO) dataset, and is then further trained on the BDD multi-object dataset, with initial inference tests conducted via a classical simulator. The next stage of development involves re-training this model directly on the converted event-based data accessing the capabilities of the neuromorphic hardware.
Concept Reply specialises in the research, development and validation of innovative solutions in the field of the Internet of Things (IoT), with a particular focus on the automotive, manufacturing and smart infrastructure sectors. Concept Reply is recognised as an expert in Testing and Quality Assurance. Thanks to its dedicated laboratories and an international team of professionals, the company is currently the trusted Quality Assurance partner for most of Italy’s leading banks, offering in-depth expertise in innovations and solutions within the global financial services market (both functional and technical – fintech), supported by observatories, partnerships and projects.