Limitez les temps d’arrêt, améliorez l’efficacité et faites des économies avec l’IoT, la maintenance prédictive, la GenAI et les jumeaux numériques.

Wouldn’t it be great if your factory told you that a key piece of equipment was about to fail and was able to guide you in fixing it step-by-step in natural language? By drawing on developments in IoT, digital twins, Gen AI, and AI, Orange has developed a system that can do just that and help significantly reduce factory downtime.

It is a business imperative. Downtime on production lines is extremely costly and creates disruption throughout the supply chain, damaging both profitability and company reputation. And once a production line has gone down, recovery times are also increasing due to restart complexity and lack of skilled personnel. In fact, according to a report from Siemens, unplanned downtime now costs Fortune Global 500 companies 11% of their yearly turnover – almost $1.5 trillion[1].

One of the most common sources of production downtime is when equipment is out of action. This could be through equipment failure, quality problems or a too-long stoppage during unplanned maintenance. The latter could happen through missing parts or a lack of on-site skills to fix the problem.

Predictive maintenance, coupled with digital twins and generative AI, promises to identify faults before they happen and help machine operators find the best solution to keep machines working at full capacity. In fact, condition monitoring of machines and preventive maintenance can increase productivity by 6% and reduce maintenance costs by 40%[2].

Taking a holistic approach

But preventing downtime is about more than just predicting when failure might happen in a specific machine; it must address the whole factory – the equipment, people and environment. For example, many machines are interdependent; failure in one can precipitate failure in another. Older equipment will also need to be integrated into the overall model, even if manufacturers no longer support it.

In addition, looking at the impact on staff is essential, both in terms of worker safety and helping staff fix equipment. On-site technicians require accurate information to fix the machine, without necessarily having to wait for specialist experts to be dispatched, which can extend the downtime.

Orange is taking a holistic approach to these challenges by investing in AI-based solutions that will improve maintenance activities while preserving the well-being of operators.

Smart maintenance in action

As part of this initiative, Orange has developed an AI for smart maintenance solution that uses IoT infrastructure, digital twin platforms, a GenAI model and adapted data models. To show the solution in real-life, the Orange team has developed a demo that simulates a paper factory powered by a turbine. This solution is easily adaptable to any factory requirement and equipment.

In the demo, Orange has attached IoT sensors to a turbine to collect data such as vibration, gyroscopic data, temperature, and noise levels. The data collected by IoT sensors is sent to a digital twin, which mirrors the factory and links physical and virtual elements. It stores data such as the physical location of devices, their operational status, and any changes over time. This allows for continuous monitoring and analysis of factory operations.

Failure prediction

An AI model then uses the data to see if the turbine is operating outside of its baseline. Initial training to establish the baseline can be done in as little as two minutes, but training for longer periods improves model accuracy. The AI monitors changes in the digital twin and detects trends that indicate failure using thresholds computed during training to classify anomalies.

To help eliminate false positives, the system requires multiple consecutive results outside the baseline before classifying an anomaly as a potential failure. This would help prevent a gust of wind, for example, from increasing the vibration in the system and generating a failure alert. 

Digital twin at heart

The digital twin is a key part of the solution as it helps understand the current state of the factory and identify potential issues. While the demo is based on a simple paper factory, the digital twin is scalable and can handle large amounts of data from multiple devices and environments. It can be easily adapted to larger plants and more complex setups to provide comprehensive monitoring and maintenance capabilities.

By analyzing the data, the digital twin can identify patterns and connections that may not be apparent through manual analysis. It stores a history of the factory's data, enabling the replay of past events. This is useful for understanding the progression of issues and identifying the root causes of failures. For example, if a device fails, the digital twin can replay data from the past 48 hours to pinpoint when and how the failure started.

The solution can be adapted to work with existing digital twins, enhancing their capabilities and providing additional services, without requiring significant changes to the existing infrastructure.

Knowledge management

The other key part of the Orange AI for smart maintenance solution is the user interface into the factory systems that allows staff to identify and fix problems.

It draws on a knowledge management repository that stores factory data, including documents, maintenance reports, user manuals, and other relevant documents. This data is integrated with the information from the digital twin and fed into a large language model (LLM) that can provide answers based on the collected data.

The AI model provides actionable insights based on the collected data, helping users make informed decisions about maintenance and operations. It helps simplify complex information and make it accessible to users with varying levels of expertise. This helps mitigate the impact of skills shortages, because workers can start troubleshooting before calling an expert, potentially resolving issues independently.

Different levels of access are provided based on the user's role, ensuring that technicians, accountants, and other roles can access relevant information. For example, a technician can see fixing information, while an accountant can see order and payment details.

Accurate information

Preventing hallucinations is a key prerequisite of the system, and Orange uses prompt engineering to guide the AI model's responses. It also helps to adjust the AI's temperature settings to control the accuracy of the answers. For example, setting the AI's temperature to 0 can help provide basic answers without hallucinations. Finding the right balance between temperature and prompt engineering is crucial for accurate responses.

Workers can also interact with the LLM using voice commands, allowing hands-free operation. For example, a technician can ask the LLM for step-by-step instructions to fix an issue to help them troubleshoot and resolve issues effectively.

After fixing issues, workers generate reports that are integrated into the knowledge repository. This is used to retrain the AI model, enhancing its accuracy over time. This feedback loop ensures that the system learns from each maintenance activity, improving future prediction accuracy.

Integration with headsets

The demo included a VR/AR headset for workers used to display information about the digital twin of the factory. These headsets also help track workers in real-time, updating the digital twin with a 3D avatar of each worker, including detailed movements and positions. This allows the system to identify the closest user to the machine in question, for example.

The headsets allow workers to ask questions and receive answers via voice, enabling them to continue working effectively without needing physical manuals. This helps ensure that maintenance tasks are performed correctly and efficiently, reducing downtime and improving overall productivity.

Furthermore, the interactions are recorded and stored in the knowledge base for further use both by the system and users. By accessing previous maintenance procedures, workers can use the headset to simulate maintenance tasks and understand procedures before performing them in real life. Finally, the headsets also help monitor working conditions, ensuring that employees are adhering to safety protocols, such as wearing appropriate gear and avoiding excessive noise or vibrations.

Business benefits

As we have described, there are significant business benefits of this AI smart maintenance solution, including reduced downtime, improved worker safety, addressing the skills shortage and better business decisions through data-driven insights.

The solution is straightforward to deploy and can be adapted to many environments – including complex factories with thousands of data points. It can also be used in other environments such as construction to help with worker safety. We recommend starting with small-scale deployments and expanding to larger factory floors as the system proves effective.

Orange is looking to work with partners to further test the system in multiple real-world environments. If you are interested in collaborating with us, then please contact partner.business@orange.com for more information.

Recommended for you
Optimize industry by digitizing tools, processes and products
Smart Industry: Revitalizing the factory floor by leveraging maturity assessment for transformation