AI@HZDR in the Research Topic ENERGY
AI applications also play an important role in the raw materials sector. At the Helmholtz Institute Freiberg for Resource Technology (HIF) at HZDR, methods are developed and applied to map, to analyze and to model raw materials. In the future, these methods will be implemented in the FlexiPlant research infrastructure, where processing routes will be digitized and automated.
Source: gettyimages, NicoElNino
Machine Learning Supports Prospecting in Remote Areas
Traditional methods for exploring and characterizing mineral deposits in complex and remote regions are costly and demanding. There is therefore a need to bring together advanced technologies to develop efficient exploration methods and help explorationists to better identify resource-rich areas.
To this end, researchers at the Helmholtz Institute Freiberg for Resource Technology (HIF) at the HZDR are combining geology, artificial intelligence and remote sensing with machine learning methods. They are developing various methods to integrate data such as multispectral and hyperspectral imaging or laboratory analyses. This allows them to map mineralogical features and identify areas with high economic potential all over the Earth. Their methods enable the creation of efficient, robust, and reproducible algorithms that can support geologists to discover the metals needed for the energy transition.
Current research focuses for example on minimizing costly and time consuming laboratory analysis by developing computer vision and machine learning techniques and integrating innovative imaging technologies. They are able to classify and quantify mineral abundances in drill core samples or at the Earth´s surface by combining satellite, airborne, and unmanned aerial system data. The researchers also develop new techniques to monitor mining activities and their impact on the environment. To achieve this, they evaluate changes caused by mining activities using time series of satellite data archives. They also integrate data from different sources to obtain more precise environmental information.
Contact:
Prof. Dr. Pedram Ghamisi
Group leader Maschine Learning at HIF
Link to private website
Press Release: Using satellite data to improve mining safety
Hyperspectral remote sensing exploration in Greenland
Source: Dr. Lorenz, Sandra
Improving Earth Observation with Data from Many Sources
The group develops innovative methods towards a better monitoring of natural and anthropogenic processes at the Earth´s surface. The team is at the forefront of research in several key aspects of recent developments in machine learning. In Earth observation, they apply tasks such as semantic segmentation, which classifies each pixel in an image according to land cover types, and scene classification, which categorizes entire images or scenes as urban, forest, or water.
Semantic segmentation involves labeling each pixel in an image with a specific category like water or forest, while scene classification assigns an overall category to the entire image or scene. Both are key tasks in analyzing Earth observation data for land use, environmental monitoring, and more.
For example Explainable AI, which refers to techniques that make the decision-making processes of AI models transparent and understandable to humans. In Earth observation, it ensures that AI-driven insights are interpretable, helping users trust and validate results. They ensure the integration of human expertise into the AI-driven analysis process. Humans validate, correct, and guide the AI, leading to improved accuracy and more reliable outcomes, particularly for complex or ambiguous data. They integrate computer vision, which involves image analysis, with natural language processing to interpret and describe Earth observation imagery.
This integration allows for querying and summarizing satellite data in a format that is easy for humans to read, thereby improving accessibility and usability. They combine data from multiple sensors, including optical and radar, as well as data collected at different times, to create a more comprehensive view of the Earth. It enhances accuracy in monitoring environmental changes, such as land use or deforestation. They also identify differences in Earth's surface over time by comparing images from different dates. It is crucial for monitoring events like urban expansion, natural disasters, or climate impacts.
Contact:
Dr. Sandra Lorenz
Group leader Remote Sensing at HIF
Press Release: Three-dimensional view of the world with artificial intelligence
In-house developed front-end R interface
Source: HZDR/HIF
Improving Separation Processes through Precise Particle Characterization
Concentration processes in the raw materials industry are often faced with challenges arising from the complexity of materials being treated: micrometer sized particles, composed of several minerals – each of them with very specific physical properties. This leads to difficulties in meeting target concentrate compositions, or in recovering enough of the target metal. There is therefore a need for better understanding the process behavior of individual particles, accounting for their complexities. The aim is to improve the efficiency of processing plants, reduce costs and simulate customized separation cycles by identifying suitable process parameters for individual materials. Additionally, being able to forecast the outcome of separation processes under these circumstances is key to reducing uncertainties and better planning raw materials' extraction activities.
To this end, researchers at the Helmholtz Institute Freiberg for Resource Technology (HIF) at the HZDR are developing techniques for modelling raw materials' concentration processes in terms of single particles, based on their complex composition. While these models can be fed with state-of-the-art particle information collected with scanning electron microscopes, researchers at HIF are going beyond and developing 3D particle characterization technologies. Using X-ray computed microtomography, they collect detailed 3D particle data to improve particle-based separation models.
In their projects, the researchers are working on precise predictions of comminution processes. They investigate the influence of particle shapes on the flotation process and the optimization of separation circuits based on the characteristics of individual particles. This research is conducted for both primary (e.g., rocks) and secondary (e.g., electronic waste) raw materials. They strive to continuously advance the understanding and optimization of concentration processes, especially regarding the complexity inherent in real raw material.
Contact:
Dr. Lucas Pereira
Group leader Geometallurgy and Particle Based Process Modelling at HIF
Press Release: Improvement of mineral processing