- Use case ecosystem
- Key facts
- Advances beyond the State of the Art
- Open source/proprietary
- Target users
- Business impact
- Potential societal or environmental impacts
Air Fusion Sp. z o.o. (AF) is an EU-based subsidiary of a U.S. software company. As an end-user of the AI-SPRINT 36-month research and innovation action, Air Fusion brings innovations to inspection and maintenance by transforming the infrastructure inspection sector with is proprietary AI-machine learning for automated damage detection, classification and change detection.
This use case combines the efforts of:
- TT Analysis: End-user developing a novel solution by optimally using more powerful resources in the performance of first-level image analysis at the edge and interventions triggered by edge processing using more powerful AI cloud-based systems.
- 7BULLS: Enhancing the monitoring platform and advanced scheduling solutions for accelerated devices for model training and retraining.
- Cloud & Heat: Providing the infrastructure and computing resources (including GPUs) for testing and validating the use case.
- Politecnico di Milano: Supporting the definition and optimisation of the AI models.
Wind turbines have a lifespan of around 20–25 years, according to research conducted by the Imperial College in London, but this lifespan can be drastically reduced when a turbine is damaged.
Wind power generator failure is often linked to the degradation of blades, generators and gearboxes and costs to fix damaged components can be extremely high, ranging from $300,000 to replace a blade to $5 million to repair the entire station.
Conducting a proper maintenance and inspection is crucial to reduce the risks of wind turbine breakage. This is where artificial intelligence systems come into play by processing massive amounts of data and using it to better predict and analyse when and what type the maintenance is needed over time.
The AI-SPRINT use case on predictive maintenance and inspection of wind turbines uses AI models for detecting damages through the collection of images by drones during their flight paths and sending them to the edge-cloud channel for analysis.
This is a crucial step in accelerating inspection time by drastically reducing the time spent by operators to analyse damage or maintenance requirements, as well as the likelihood of human error by using machine learning systems.
- Enable the best interaction of cloud-based analysis and local processing using lighter data pattern recognition routines.
- Increase the reliability of windmill plants and enable predictive maintenance.
- Exploit privacy preserving solutions, whenever a potential problem is detected.
The underlying architecture maps into the reference AI-SPRINT architecture, including design and Run-Time tools specific to the Maintenance and Inspection Use Case. The deployable infrastructure consists of (1) Training nodes used to train the models (2) edge devices running VM nodes, which will typically run on a laptop and (3) a lightweight edge device on the drone itself.
All models will be trained on a public or private cloud, the inference will run on either an edge device or the cloud, also through FaaS, depending on the inference type .
At Run-Time level, SPACE4AI-R and Krake orchestrate the computation; MinIO provides local storage which can be synchronised with the cloud; IM and EC3 manage container deployment and elasticity, respectively.
Currently most of the inspection companies are leaning towards drone-based asset inspections. The data acquisition process is manual or semi-automated. Data is downloaded manually from the drone to the laptop and then to the cloud. The use case goal is to improve that process by:
- making sure that photos taken in the process are of good quality;
- providing quick feedback on major issues found on the blades;
- limiting the amount of data that needs to be processed .
As a result the whole process will provide better quality data, in a shorter time.
The tools used are open source. Resulting platform is proprietary.
- Drone operators working on wind farms;
- Data analysts preparing damage reports;
- Wind farm operators;
- Inspection companies.
This AI-SPRINT use case will significantly improve the efficiency of AI models, bringing new market opportunities for the entire damage identification workflow. Air Fusion will be able to take to market novel AI-enabled products, spanning telco towers, power transmission lines, gas pipelines and the energy footprint of buildings by using distributed AI facilities. Competitive edge will come from operational excellence through seamlessly distributed computations from cloud to edge.
The time series processing part of the AI-SPRINT framework could significantly accelerate the development of new backend modules analysing measurement data generated as current and historical images of damage to numerical measures reflecting the defined parameters and their evolution over time.
Drone operators thanks to the quick feedback on the photos taken during the inspection can redo some parts of the inspection without additional trip to the wind farm (instantaneous feedback).
Through the development of artificial intelligence models developed over the 3-year lifespan of the AI-SPRINT project, this use case will help reduce the environmental effects caused by malfunctioning wind power stations. It will also significantly contribute to energy efficiency and environmental sustainability. By fostering technology-based advances in maintenance and inspection, AI-SPRINT expects to contribute to the United Nation’s Sustainable Development Goal 9 on industry, innovation and infrastructure.