The hidden signals that predict mining failures

The hidden signals that predict mining failures. 09.12.2025. Mining affects the environment directly, so knowing exactly what happens in water and soil is crucial. That is why TalTech Department of Geology postdoctoral researcher Bruno Grafe – who has worked both in mining research and in a German mining authority – believes that constant measurements can make a real difference. Over the last twenty years, sensors have become cheaper, smaller and far more capable. At the same time, Europe’s rules have not quite caught up, meaning real-time data is often treated as “extra”, not as official evidence. For researchers, that opens a challenge: to prove these tools are reliable enough to be trusted. Why “continuous data” matters Grafe opened with a simple idea: if you measure rarely, you see only snapshots. If you measure constantly, you see reality. He reminded the audience of the 2022 disaster on the Oder River, where 25-50% of the total fish population  died almost overnight. Real-time data could have helped identify the dangerous algae bloom earlier and understand how temperature, salt and fertiliser residues came together to trigger it. For mining companies, constant measurements have another value – they make processes more predictable and efficient. If you see changes the moment they happen, you can fix problems before they shut down the system. How real-time data uncovers early warning signs Grafe’s main story came from a gold mine on the West African coast. The problem there wasn’t cyanide, as many might expect, but gypsum. When sulphate and calcium are present at high concentrations gypsum crystals start to grow inside reverse-osmosis filters that clean the water– and those filters are expensive to replace. To avoid that a continuous Raman spectroscopy sensor platform (AISRAS®) built by Berlin based startup Watergenics was installed and tested over the course of one year. Simplified, it’s a small, robust box that shines a laser into the water and reads the light that comes back. From that, it can “see” what is dissolved in the water in real time. “We saw exactly when the mine switched between different water ponds – the sulphate level jumped instantly,” Grafe said. Lab samples taken once a day would never reveal this. Even more useful was what the sensor noticed before the system’s own automation did: the first signs of clogged filters. When sulphate levels rose slowly but steadily, it signalled that gypsum was building up. “For any operator, this is extremely valuable – it gives you time to react, adjust dosing or clean the system early,” he explained. The study has since been published, showing that the sensor’s accuracy is good for real industrial use. Estonia’s mining sector would also benefit from real-time data Grafe’s second example came from a large salt mine in Germany, where sensors track the salt content of a river before, during and after passing the mine. But sensors themselves can get dirty or drift out of calibration. To solve this, the team used a simple logic: if three sensors in the same river show very different values, one of them is probably wrong. With machine learning, they trained a model to recognise which changes were natural – such as rising water levels – and which were caused by a faulty sensor. Grafe put it plainly: “A sensor can work perfectly, but if it’s clogged, the data is still wrong.” The algorithm now alerts operators earlier than human monitoring alone. When asked about his future work here, Grafe said his goal is to bring more real-time data and automation into Estonia’s mining sector. “We need these tools – not only for environmental reasons but also because there’s simply not enough people to run operations the old way,” he noted.