Machine learning can be the secret to better battery life, as computers predict the best factors for effective design.

You’ve probably asked yourself this question countless times when replacing batteries in a cell phone or flashlight or buying a new expensive car battery. Guessing how long these batteries will power your device or vehicle is not only inconvenient but also expensive.

So far, researchers and battery manufacturers have had only one surefire way to test battery life – keep cycling until the battery is finally finished. Battery cycling involves a full charge and then a discharge. Unfortunately, this method can take years and is expensive, says electrochemist Susan Babinets in a press release from the Argonne National Laboratory about a new study she co-authored. This study uses the benefits of machine learning, teaching computers to recognize patterns in data to make predictions about new data. In this case, computers were able to accurately predict how long different types of batteries will last.

Scientists from the U.S. Department of Energy’s Argonne National Laboratory in Lemont, Illinois, have collected experimental data from 300 batteries representing six different chemicals, including the types and locations of atoms that make up the cathode structure of the battery. “Different types of cathodes can store more or less energy and can degrade faster or slower,” says Noah Paulson, an argon computing scientist and author of the study, in an email to Popular Mechanics. Another difference between batteries was in the chemical additives of the battery electrolyte.

Scientists have allowed computers to perform work to determine exactly how long different batteries will last. In this study, researchers studied lithium-ion batteries, which can be “charged and discharged thousands of times, depending on how they are used,” Paulson says. This work reflects what will happen to most batteries, such as lithium-ion, nickel-metal-hydride, or lead-acid. “This is important because batteries for cars, planes, network storage, electronics and more need to be rechargeable,” he explains. On the other hand, alkaline batteries, like the ones you use in the TV remote control, do not usually recharge.

The authors based their research on the fact that machine learning can predict the life of a lithium-ion battery in just a few weeks, with a maximum of 100 cycles. According to a study published in the February 25 issue of the Journal of Power Sources, the researchers’ method of machine learning took just one previous cycle to make a useful prediction.

Batteries store chemical energy in the form of chemical compounds. For example, an alkaline battery, or cell, contains zinc, manganese dioxide and potassium hydroxide. When connected at both ends to a circuit such as a light bulb, the zinc inside reacts with manganese dioxide and loses electrons. Electrons flow through the metal rod in the cell from the negative terminal to the lamp and cause it to ignite. They then continue to flow and enter the positive terminal of the cell. In this type of battery without recharging, once the zinc electrons are used, the cell is dead.

Batteries power our tools, instruments, toys and vehicles. No one knows how long a new battery will last, because its potential lifespan depends on many factors – what we use them for, their internal chemistry and overall design.

“For every different type of battery application, from cell phones to electric vehicles and network storage, battery life is fundamental to every consumer,” Paulson said in a release. Battery testing means carrying it through thousands of cycles when it converts stored chemical energy into electrical energy. The machine learning method “creates a kind of computational test kitchen” that quickly shows how the battery will work, he explains.

To set up their experiment, the researchers provided computers with vast amounts of raw data, identifying 397 different features of batteries that they thought would be useful for machine learning algorithms. Computers used a specific set of rules or algorithms to statistically analyze these raw “training data,” which were to familiarize computers with current and voltage as a function of time throughout the life of different batteries.

Based on computer training, machines have learned to recognize patterns among different battery functions and build a model that could be used to predict new data such as average battery charging time. “Then we used an algorithm to find which subset of functions would give the best predictions,” Paulson says.

By repeating this process, computers ’predictions about the new battery design became more accurate. “In this study, we only use this information from the first to 100 charge-discharge cycles so that, given the new battery, we can estimate its life without an experimental cycle of several months or years,” says Paulson.

Researchers have established a machine learning algorithm based on a lithium battery that is well understood. They “taught” computers on this algorithm to make predictions about the longevity of the unknown chemical composition of the battery. “In essence, the algorithm can help steer us toward new and improved chemicals that provide longer life,” Paulson says. Researchers believe the machine could accelerate the development of potential materials for batteries because scientists from the lab will be able to test the material faster.

Different batteries degrade and fail differently, Paulson says. “The value of this study is that it gave us signals specific to how different batteries work.” The most aggressive way to shorten battery life is to charge very quickly. As a result, lithium metal covers the electrode particles. “The interaction between the electrolyte and the electrode particles creates a film that can protect the particles, but if it becomes too thick, it can become a barrier to [lithium] ions that move in and out of particles, ”says Paulson. Another way to reduce battery life is to split the electrode particles. These are just a couple of examples.

“Say you have new material and you’ll roll it over a few times. You can use our algorithm to predict its longevity and then decide whether you want to continue using it experimentally or not, ”says Paulson.












Source by [author_name]

Previous articleTwo-time World Cup winner Andrew Simmonds killed in car crash – SABC News
Next articleJUST IN | The load reduction will resume at 5pm and move on to Stage 3 on Monday night