Technological Thinking Triggered by the Epic Drama "Chernobyl": How to Avoid the Next Nuclear Disaster?
The popularity of the American TV series "Chernobyl" has made the audience feel the fear of nuclear accidents again. By applying artificial intelligence technology to nuclear protection and nuclear safety, it will bring fruitful results. Perhaps, the reasonable use of artificial intelligence to "constrain" the use of nuclear energy will give full play to the value of resources.
Game of Thrones finally came to an abrupt end in a disastrous plot. It is a pity that such a great drama ended like this. But HBO did not give the audience much time to relax, and another great drama appeared in people's vision.
The recently popular American TV series "Chernobyl" has only been updated to the third episode (out of a total of five episodes), but it has already received perfect scores on major rating websites, and has quickly become a hit TV series sought after by thousands of people.
Apart from its production advantages, one reason why it is so popular is the topicality of the Chernobyl nuclear incident itself, as it carries the memory of a catastrophic nuclear accident.
The Chernobyl nuclear power plant accident occurred on April 26, 1986 in northern Ukraine, which was then part of the Soviet Union.
On that day, the fourth generator unit of the power plant exploded, the nuclear reactor was completely destroyed, and a large amount of radioactive material leaked, becoming the largest accident in the nuclear power era (level 7). The radiation hazards were serious, resulting in 31 deaths within 3 months before and after the accident, 60,000 to 80,000 deaths in the following 15 years, and 134,000 people suffering from radiation diseases of various degrees.
In the nuclear power industry, is there any way for artificial intelligence to provide safety protection?
Efficiency or safety is never a choice
Serious nuclear power plant accidents have caused huge disasters. However, as one of the important energy sources in the world, nuclear energy still attracts people to explore its development.
The most attractive thing about nuclear energy is that it possesses amazing energy: the energy released by the fission of one kilogram of uranium-235 (a commonly used nuclear fuel) is equivalent to the energy released by burning 2,700 tons of coal.
But the greater the energy, the greater the uncontrollability, and once it gets out of control, the greater the damage it causes.
In previous accidents, improper human operation accounted for a large part of the causes. Therefore, how to use technology to achieve safer nuclear energy utilization is the most urgent problem to be solved.
The introduction of automated deployment and even artificial intelligence-related solutions has the opportunity to make nuclear energy a more docile wild horse in maintaining equipment production safety and safety monitoring.
Early detection of cracks in protective layers: machine vision
The United States is the world's largest commercial nuclear power supplier, providing about 20% of electricity. However, between 1952 and 2010, the United States had 56 incidents of varying severity, 19 of which were related to protective layer ruptures or failures. The cost of remediation reached $2 billion.
Those protective parts that should have been solid have suffered functional loss and safety damage due to aging and other reasons, including cracking, fatigue, embrittlement, wear, corrosion, oxidation, etc. This casts a shadow on safety.
To solve this problem, a research team from Purdue University developed a CRAQ (Crack Identification and Quantification) system.
The system combines graphics processing with deep learning, using analysis of video of the protective layer to identify changes in the metal texture to predict and target crack issues.
Nuclear reactors usually need to be submerged in water to maintain cooling, and due to factors such as high temperature and radiation, the components of the reactor cannot be directly inspected manually. The intelligence uses remotely recorded video of the underwater reactor surface to find cracks.
However, purely manual review is labor-intensive and prone to errors. In order to develop an efficient detection system, the researchers collected videos of 20 underwater specimens of internal nuclear power plant components, scanned the samples at 30 frames per second, and used convolutional neural networks to check for cracks in each frame.
The algorithm observed the crack from one frame to the next and was able to account for the changing configuration due to the moving camera, pinpointing the crack's location. The algorithm simulated the ability of human vision to carefully examine the crack from different angles, and was able to avoid the effects of shooting light.
The method also uses a dataset containing about 300,000 cracks and non-cracks to test the model. Tests show that the CRAQ system has a success rate of 98.3% in tracking cracks.
The distribution of radiation is predicted by models
The nuclear power plant can be put into production after going through layers of safety checks. However, when danger caused by unexpected factors occurs, in addition to the management finding ways to resolve and respond, another important aspect is to arrange the evacuation of people in a timely and reasonable manner.
Another nuclear accident in the world that was rated as level 7 was the nuclear power plant disaster in Fukushima, Japan in 2011. In order to minimize the impact of such incidents, machine learning and other related technologies are also used to assist in evacuation in response to nuclear leaks.
In July 2018, researchers from the University of Tokyo developed a set of machine learning-based tools to predict the geographical distribution of radioactive materials.
When a nuclear power plant accident occurs and radioactive material is released, it is crucial to evacuate nearby residents as quickly as possible. However, it is difficult to immediately predict where the leaked radioactive material will settle, thereby preventing people from being exposed to danger.
The research team trained a machine learning model using a dataset of near-ground wind conditions marked by meteorological simulations. The model was able to use algorithms to predict where radioactive materials will be distributed and the path of propagation, etc.
After being trained with historical weather pattern data, the tool consistently achieved over 85% forecast accuracy, reaching 95% accuracy when winter or predictable weather predominated. The model was also shown to be able to make accurate predictions up to 33 hours in advance.
This system can help evacuate immediately after a nuclear leak accident, so as to better take positive countermeasures when a disaster occurs.
Artificial intelligence makes nuclear energy safer
Ma Yilong once said in a conversation, "AI is much more dangerous than nuclear weapons."
The current focus on the use of AI in nuclear energy is not on protection or safety testing, but more on the dangers that AI can cause.
Of course, if AI is used in weapons research and other areas, it will inevitably pose a serious threat. But don’t forget that at this stage,Where technology is used is entirely the responsibility of humans.
The UK has reported on using AI and robots to help clean up the collection and treatment of nuclear power plant waste. This work, which endangers humans, can be easily solved using new technologies. This is the same as AI's efforts in nuclear protection, both of which are to create reassurance for people to use nuclear energy.
Perhaps, as long as it is used in the right place, AI can restrain nuclear energy and protect our home planet.