從電機到機械臂,所有汽車組件均可利用強大的算法進行分析。
作為寶馬汽車新興技術團隊的成員,Lukas Müller的工作是對各種新技術進行深入分析,以評估寶馬是否有望應用這些技術。他并非首位考慮量子計算如何影響寶馬集團造車方式的內部員工,但他向SAE透露,寶馬正計劃駕馭更強大的計算機以打造更優(yōu)質的未來汽車,而他所做的工作只是冰山一角。
今年6月,寶馬宣布將與Classiq和英偉達達成合作,共同研究最適合未來汽車的電氣和機械系統(tǒng)的架構設計。他們的合作思路是利用量子計算開發(fā)出一種實時解決方案,將傳動系看作線性方程組來分析動力總成中可能包含的電機、電池、冷卻系統(tǒng)等一系列組件,從而提高汽車架構的效率。
Classiq的技術市場經理Erik Garcell告訴SAE:“電氣和機械系統(tǒng)非常復雜。在分析系統(tǒng)數(shù)據時,必須考慮輸入這些系統(tǒng)的電力相位問題,以及開啟或關閉設備的時序問題,這樣才能確保系統(tǒng)正常運行。”
Garcell表示該項目最終將開發(fā)出一款車載設備,可根據實時數(shù)據計算出需要開啟或關閉的設備,以及相應的執(zhí)行順序。但由于目前尚未開發(fā)出可擴展的量子系統(tǒng),因此目前首先由量子計算機完成汽車生產前的復雜分析工作。接著,精簡后的車載系統(tǒng)會利用這些分析結果,根據量子系統(tǒng)生成的規(guī)則來控制動力總成的各個設備。Garcell表示,接下來就需要利用量子近似優(yōu)化算法(QAOA)來解決問題了。
他表示:“我們一直在嘗試優(yōu)化這套線性方程組系統(tǒng),也就是由電氣組件構成的傳動系。如何優(yōu)化這個如神經網絡般的龐大且復雜的系統(tǒng)?這并非一次性就能解決的問題,而是需要不斷向系統(tǒng)本身反饋數(shù)據。我們可將傳動系看作一個圖形理論問題。通過使用QAOA算法進行優(yōu)化,研究人員便能構想出一個更高效的線性方程組系統(tǒng),并將其與原系統(tǒng)進行回溯比較,從而判斷哪個傳動系更優(yōu)。具體而言就是分析哪些設備以怎樣的方式相連,數(shù)據以怎樣的方式互相傳輸,才能獲得最高效的線性方程。更高效的電子傳動系通常能夠節(jié)省更多的能源。”
不過,據寶馬集團的Müller介紹,必須首先了解量子計算機對優(yōu)化傳動系的作用才能執(zhí)行整個優(yōu)化過程。了解機器人如何能將工作時間縮短幾毫秒(這是寶馬集團的另一個量子計算項目)與通過仿真來研究最優(yōu)冷卻液管道厚度或如何優(yōu)化車內的冷卻液流動路徑皆然不同。
Müller表示:“我們決定從一個非?;A的問題入手,選取汽車中的四個組件,并向算法提出問題,例如,‘這些組件之間采取怎樣的熱量傳遞和連接方式才能達到最高效率?’也許你會向算法提供一些解決方案,例如功率各不相同的冷卻設備,當然這些方案可能會非常昂貴。而算法最終可能會輸出這樣的結果:‘最高效的解決方案是將這三個組件以A方式連接起來’,或是‘選擇這四個組件,將其中三個相連,另一個僅與B組件相連’。”
Classiq曾與羅羅公司(Rolls Royce)在噴氣發(fā)動機領域進行過合作,也曾與其他汽車OEM有過合作經歷。不過與寶馬的合作是Classiq能夠公開分享的首個汽車領域的項目。
Garcell稱:“量子計算機本質上應成為優(yōu)化系統(tǒng)架構的顛覆性技術。許多公司都在考慮使用這項技術,不僅是為了提高電動車的效率,更是希望利用量子仿真技術研發(fā)性能更高的電池,創(chuàng)造出能夠加快充電速度或提高整體容量的全新電池化合物。汽車行業(yè)正在多個領域探究量子計算技術的應用。”
Müller透露,寶馬確實有一小部分人從事量子計算工作。他們進行自主研究、撰寫論文并與外部公司合作。在2024年初,寶馬聯(lián)合空中客車(Airbus)舉辦了空客—寶馬集團量子計算挑戰(zhàn)賽(ABQCC),旨在“將量子技術應用于現(xiàn)實工業(yè)場景”。
在利用量子計算機研究電氣架構之前,寶馬曾使用該技術檢驗工廠的改進措施。確切地說,檢驗內容是機器人在制造工廠中的移動路徑規(guī)劃,而Müller當時參與了其中的量子計算工作。寶馬并未直接在工廠中應用量子計算生成的解決方案,即尚未在實際生產線上投入優(yōu)化后的機器人,而是正在研究哪些問題適合采用量子計算機解決,并評估可以提高多少速度和效率。
Müller稱:“優(yōu)化問題是我們關注的關鍵領域之一,因為我們面對許多需要優(yōu)化的任務。生產和設計汽車是其中最復雜的任務之一。如今,工廠對機械臂的應用日漸增多,而我們總希望縮短其完成特定任務的時間,因為這會直接影響到生產汽車的速度。”
Müller以寶馬利用量子計算研究PVC的應用策略為例向我們進一步解釋:“我們面臨的主要問題是‘一個或多個機器人完成這項工作的最佳順序是什么?’隨著處理的接縫數(shù)增加,可執(zhí)行的順序數(shù)量可能會呈指數(shù)級增長。如果機器人可使用不同類型的工具,比如角式噴嘴或雙噴嘴,那么問題就會變得更加復雜。即使只涉及到幾秒鐘,甚至不到一秒的用時變化,也會對后續(xù)流程產生影響。”
Everything from electric motors to robot arms can be looked at through the lens of a powerful algorithm.
As part of BMW’s Emerging Technologies team, Lukas Müller picks apart new ideas to evaluate if they are relevant for the automaker. He wasn’t the first at BMW to look into how quantum computing might change the way the German automaker builds cars, but he told SAE Media that the work he’s doing is just the tip of the iceberg when it comes to harnessing more powerful computers that will help build the better cars of tomorrow.
In June, BMW announced it would collaborate with Classiq and NVIDIA to find optimal architecture designs for the electrical and mechanical systems in future vehicles. The idea was to use quantum computing to develop a real-time solution that would make a vehicle’s architecture more efficient by analyzing a series of potential motors, batteries, cooling systems and other components that might be used in the powertrain by looking at a drive train as a series of linear equations.
“It’s a really complex system,” Classiq’s technical marketing manager, Erik Garcell, told SAE Media. “When you get into the data, you have to worry about the phase of the power going into these systems, too, and timing that and making sure it’s all good.”
Garcell said any eventual product that comes out of this project would be an on-board device that calculates what to turn on and what to turn off in which sequence based on real-time data. Since scalable quantum systems are not yet available, the difficult analysis would be done by the quantum computer before the vehicles are built. Then, simpler, on-board systems would use the learnings to control powertrain devices based on the rules the quantum system came up with. Garcell said the next step was to apply a quantum approximate optimization algorithm (QAOA) to the problem.
“We were trying to optimize the system of linear equations, that drivetrain we’re talking about of electrical components,” he said. “How do you optimize this huge and complex neural network? It’s not just the one pass. It’s feeding data into itself. You could think of it almost like a graph theory problem. By optimizing this using this QAOA algorithm, they’re able to figure out a more efficient system of linear equations which they can then backtrack out to the original system and say, this is the more optimal drive train. If this is connected to this, connected to this, connected to this, in this way, and the data is feeding to each other in this way, that would be the most efficient linear equation, the more efficient electronic drive train that would more often than not save energy.”
Step one of the whole process though, was understanding what quantum computers can and can’t do to solve this problem, according to BMW’s Müller. Understanding how robots might shave a few milliseconds off of their job time – another of BMW’s quantum computing projects – is different than running simulations to discover which pipe thickness will work best or how to optimize the flow of the cooling liquid in a vehicle.
“We decided on, as a very basic problem, taking four components that would exist in a car and asking, for example, how do we transfer heat between them and how do we connect them together,” Müller said. “Maybe you could put into the solution space a range of different coolers which have different powers but, of course, they might be more expensive. In the end, the algorithm would spit out, ‘the most efficient one is taking these three and connecting them together in that way,’ or ‘take these four, but connect these three and this one only to this component.”
Classiq previously worked with Rolls Royce on jet engines and has done work with other automotive OEMs. The work with BMW, though, is the first automotive endeavor that Classiq can talk about publicly.
“Quantum computers are supposed to be, essentially, the game changer for optimization,” Garcell said. “A lot of companies are looking at them, not just to make their electric cars more efficient, but to build better batteries themselves, through quantum simulation, to create kind of new compounds for the battery that can either charge faster or hold more charge overall. There are a lot of different places people are looking into quantum computing for the automotive space.”
BMW does have a small number of people working on quantum computing, doing their own research, working on papers and working with external companies, Müller said. In early 2024, BMW partnered with Airbus for the Airbus-BMW Group Quantum Computing Challenge (ABQCC) which was designed to “harness quantum technologies for real-world industrial applications.”
Before investigating electrical architectures using quantum computers, BMW used the technology to test out factory improvements. Specifically, Müller was involved in quantum work on robot path planning in a manufacturing facility. BMW hasn’t yet putting quantum computing’s solutions to work in its plants, with optimized robots cruising around actual production lines. Instead, BMW is investigating which problems are amenable to quantum computer solutions and how much speed and efficiency might be gained.
“Optimization problems are one of our key areas because we have quite a lot of them,” Müller said. “Producing and designing vehicles is one of the most complex tasks there is. Nowadays, there are more and more robot arms that work in the factory. And you always want to decrease the time they need to finish a certain task, because this can have big influences on how quick we are able to produce the cars.”
BMW used quantum computing to investigate its PVC application strategy, Müller said. “The main question is, ‘what’s the best order for one or maybe multiple robots to do this?’ The number of possibilities of the order that you can do increases exponentially with the number of seams that you have. It gets more complex if you have different kinds of tools that the robot can use, maybe corner nozzles or ones with two nozzles. Even if you [just] get a couple of seconds, or less than a second improvement, this can have influences down the line.”