What is the core technology of BMS?
Recently, I saw a company’s promotional board, which claimed to fully master the software and hardware technology of the power lithium battery management system (BMS), reach the world’s advanced level, and adopt multiple balance control capabilities because of the use of the underlying software such as the AUTOSAR software architecture. Very eye-catching. Are these things the core technology of BMS?
Usually the BMS system usually includes a detection module and an operation control module.
Detection refers to measuring the voltage, current and temperature of the battery cell and the voltage of the battery pack, and then sending these signals to the computing module for processing and issuing instructions. So the computing control module is the brain of the BMS. Control modules generally include hardware, basic software, runtime environment (RTE) and application software. One of the core part of the application software. The environment developed with Simulink is generally divided into two parts: the estimation algorithm of the battery state and fault diagnosis and protection. State estimation includes SOC (StateOfCharge), SOP (StateOfPower), SOH (StateofHealth), and balance and thermal management.
Battery state estimation is usually to estimate SOC, SOP and SOH. SOC (state of charge) simply means how much power is left in the battery; SOC is the most important parameter in BMS, because everything else is based on SOC, so its accuracy and robustness (also called error correction ability) is extremely important. If there is no accurate SOC, adding more protection functions will not make the BMS work normally, because the battery will often be in a protected state, and the life of the battery cannot be extended.
In addition, the estimation accuracy of SOC is also very important. The higher the accuracy, the higher the cruising range for the battery with the same capacity. Therefore, high-precision SOC estimation can effectively reduce the required battery cost. For example, Chrysler's Fiat 500eBEV can always discharge SOC=5%. It became the electric car with the longest cruising range at that time. The figure below is an example of the robustness of the algorithm. The battery is a lithium iron phosphate battery. Its SOCvsOCV curve only changes about 2-3mV from 70% to 95% of SOC. The measurement error of the voltage sensor is 3-4mV. In this case, we deliberately let the initial SOC have a 20% error, and see if the algorithm can correct the 20% error. If there is no error correction function, SOC will follow the curve of SOCI. The SOC output by the algorithm is CombinedSOC, which is the blue solid line in the figure. CalculatedSOC is the real SOC backed up according to the final verification result.
Accurate estimation of SOP can maximize battery utilization efficiency. For example, when braking, it can absorb as much feedback energy as possible without damaging the battery. When accelerating, more power can be supplied to obtain greater acceleration without damaging the battery. At the same time, it can also ensure that the car will not lose power due to undervoltage or overcurrent protection during driving even when the SOC is very low. In this way, the so-called first-level protection and second-level protection are all passing in front of the precise SOP. That's not to say protection isn't important. Protection is always needed. But it cannot be the core technology of BMS. Accurate SOP estimation is especially important for low temperature, old batteries, and very low SOC. For example, regarding a group of well-balanced battery packs, when the SOC is relatively high, the SOC difference between them may be very small, such as 1-2%. But when the SOC is very low, the voltage of a cell will drop rapidly. The voltage of this cell is even more than 1V lower than other battery voltages. To ensure that the voltage of each battery cell is always not lower than the minimum voltage given by the battery supplier, the SOP must accurately estimate the maximum output power of the battery cell whose voltage drops rapidly at the next moment to limit the use of the battery and protect the battery. The core of estimating SOP is to estimate each equivalent impedance of the battery online in real time.
SOH refers to the state of health of the battery. It includes two parts: ampere-hour capacity and power changes. It is generally believed that when the ampere-hour capacity decreases by 20% or the output power decreases by 25%, the battery life is over. However, this does not mean that the car cannot be driven. For pure electric vehicles EV, the estimation of the ampere-hour capacity is more important because it has a direct relationship with the cruising range and the power limit is only important when the SOC is low. For HEV or PHEV, the power change is more important because the battery's ampere-hour capacity is relatively small, and the power that can be supplied is limited, especially at low temperatures. The requirements for SOH are both high precision and robustness. And SOH without robustness is meaningless. Accuracy below 20% is meaningless. The estimation of SOH is also based on the estimation of SOC. So the SOC algorithm is the core of the algorithm. The battery state estimation algorithm is the core of BMS. Everything else is at the service of this algorithm. So when someone claims to have broken through or mastered the core technology of BMS, you should ask him what he has done in BMS? Is it an algorithm or active equalization or only the hardware and underlying software of the BMS? Or just propose a BMS structure?
Some people say that Tesla is awesome because its BMS can manage 7104 batteries. Is this where it's awesome? Is it really managing 7104 cells? Tesla model S does use 7104 batteries, but there are only 96 batteries in series, and only one battery in parallel, no matter how many batteries you connect in parallel. why? Because the battery packs of other companies only count the number of series connection instead of the number of parallel connection. Why should Tesla be special? In fact, if you understand Tesla's algorithm, you will understand that Tesla's algorithm not only requires a large amount of working condition data to be calibrated, but also cannot guarantee the estimation accuracy under any circumstances, especially after the battery ages. Of course, Tesla's algorithm is much better than almost all domestic BMS algorithms. Almost all domestic BMS algorithms use current integration plus open circuit voltage to calculate the initial SOC with the open circuit voltage, and then use current integration to calculate the change in SOC. The problem is that if the voltage at the starting point is wrong, or the ampere-hour capacity is inaccurate, wouldn’t it be corrected until the battery is fully charged again? Will the voltage at the starting point be wrong? Relevant experience tells us that it will, although the probability is very low. If you want to be foolproof, you can't just rely on the accurate voltage at the starting point to ensure the correctness of the starting SOC. An active equalization technology selected by experts last year won a Lithium Battery Golden Globe Award. The reason is that its core technology - active equalization technology can prolong battery life by 30% and cruising range by 20%. This looks unreliable. Because it cannot be quantified at all. Compared with whom can you extend your life by 30%? Does it make sense to compare yourself to yourself? Compared with no balance? Then your level is far behind. Compared with others, it should be meaningful to compare with the best. Not to mention the best BMS in the world, at least not bad, have no balance problem. How do you prolong life by 30%? The same goes for extending the cruising range. Such as Chrysler's Fiat500e, its SOC can be kept at 5%. How can you extend the cruising range by 20%? Going one step further, is active balancing difficult? Hardware In 2008, TI sold its active equalization IC to the company I worked for at the time. The algorithm is nothing more than equalizing the battery from the same module and balancing the batteries between different modules. General Motors had completed the simulation verification as early as 6-7 years ago. There are even articles. From an algorithmic point of view, there is no difficulty at all. And the active equalization is not at all what is said on the Internet. The active equalization function has always been the killer feature of foreign products. Why do foreign countries basically not use active balancing? It is important to consider cost issues. If passive balance can be done, why use automatic balance? Why does the country vigorously advocate active balance? The author thinks that the important thing is that the passive balance cannot be fixed. Speaking of passive equalization, most people told the author that it is because the quality of domestic batteries is too poor and the consistency is not good. However, through conversations, the author found that the root cause lies in unclear concepts and wrong methods. Otherwise, how could the balance become more balanced and worse when driving? The effect of equalization can be calculated. The so-called multiple equalization technology clearly means that there is no single method that can achieve equalization. Some people say that passive equalization wastes a lot of electricity. So not good. Taking a 96-cell battery pack as an example, we can calculate how much power is wasted by passive equalization in the worst case. If the equalization current is 0.1A, a battery will waste about 0.4W when it is equalized. The worst case is that 95 batteries have to be discharged, so the worst case is 0.4X95=38W. It's not as expensive as a car's headlight (about 45 watts). If it is not the worst case, perhaps only a dozen or even a few watts is enough. Therefore, although passive equalization wastes a little electricity, if it can greatly extend the life of the battery, why not do it? Others say that the 0.1A current is too small for a battery with a relatively large Ah capacity. If imbalances can be nipped in the bud, there will be no powerlessness. If the battery itself is not working properly, neither active equalization nor passive equalization can do anything. Therefore, the poor consistency of the battery cannot be entirely blamed. You have to find the reason yourself. There are two PHEV cars in the car that the author has made, and the difference in SOC in the battery pack is as high as 45% after only a few months of driving. Moreover, due to SOC and SOP problems, the car often breaks down on the road. The company agreed that it was a battery quality problem and agreed to replace the battery supplier. But I just changed the algorithm and solved the balance problem. And it was done when the company clearly stipulated that charging was not allowed. Because one car has already been in an accident due to a battery problem. The difference in cell SOC in the battery pack is reduced from 45% to 3%. The car has now driven more than 100,000 kilometers. The problem of breaking down never happened again. What kind of algorithm is the core technology?
From a control point of view, a good algorithm should have two criteria: accuracy and robustness (error correction capability). The higher the accuracy, the better the reason is not much to say here. The aforementioned current integration plus open-circuit voltage actually uses open-circuit voltage to correct errors, but this method is obviously far less robust than online real-time error correction. This is why large foreign companies are using online real-time estimation of the open circuit voltage to achieve online real-time error correction.
Why the emphasis on real-time online estimation here? What are its benefits? All equivalent parameters of the battery are estimated through real-time online estimation, thereby accurately estimating the state of the battery pack. Real-time online estimation greatly simplifies the calibration work of the battery. It makes the precise control of the state of the battery pack that is not very consistent become a reality.
Some people in China often don't know what other people's algorithms are. It is inappropriate to think that a certain manufacturer has mastered the core technology of BMS when they see some parts of BMS for a certain factory. Those large-scale publications that cost tens of thousands of dollars to buy comment on the pros and cons of BMS from various manufacturers, regardless of the differences in BMS algorithms or core technologies, have little practical significance. Just looking at whether it is supplying BMS for a well-known OEM is considered awesome, and I don’t know what is in the BMS. I don't know if there is a kind of psychology of admiration for foreigners.
What are the characteristics of the best BMS in the world at present? It can estimate the battery parameters of the battery pack in real time online to accurately estimate the SOC, SOP, and SOH of the battery pack, and can correct the error of the initial SOC exceeding 10% and the error or percentage of the ampere-hour capacity exceeding 20% in a short time How many current measurement errors. General Motors of the United States did an experiment to test the robustness of the algorithm when it developed Volt 6 years ago: Remove one string from the battery packs connected in parallel with 3 strings, and then increase the internal resistance by 1/3, Ah capacity is reduced by 1/3. But BMS doesn't understand. The result is that SOC, SOP are all corrected in less than 1 minute and SOH is then accurately estimated. This not only shows the strong error correction ability of the algorithm, but also shows that the algorithm can keep the estimation accuracy unchanged throughout the battery life cycle.
As far as the computer is concerned, if a blue screen appears, we usually just need to restart the computer. However, for a car, even if the probability of breaking down is only one in ten thousand, it is intolerable. Therefore, unlike publishing articles, automotive electronics must be guaranteed to work under any circumstances. To make a good algorithm requires a lot of energy to solve those situations where the probability of occurrence is only one in a thousand or one in ten thousand. This is the only way to ensure nothing goes wrong. For example, when a car is driving on a winding mountain road at high speed, the battery model that everyone knows will fail. This is because the continuous high current will quickly consume the charged ions on the electrode surface, and the internal ions will not have time to diffuse out, and the battery voltage will drop sharply. The estimated SOC will have a large error or even an error of more than 10%. The precise mathematical model is the diffusion equation mentioned in the textbook of mathematical physics methods. But it cannot be used in the car because the computational complexity of the numerical solution is too large. The CPU computing power of the BMS is not enough. This is not only an engineering problem, but also a math and physics problem. Solving such technical problems can resolve almost all known polarization problems that affect battery state estimation.
The state estimation technology of BMS is the core technology of BMS. Although 6 years have passed, there is still no supplier in the world that can achieve such a high level of precision and robustness to ensure that the battery works foolproof. Even Tesla, which is now red and purple, is far behind. This is not bragging. Tesla fans must have heard of Tesla being dragged away on the streets of Beijing. Tesla's algorithm also cannot guarantee accuracy and robustness as batteries age. Only an algorithm that can guarantee high precision and high robustness is the killer! Without such technology, how to overtake on a curve?