20 多年來，電動汽車電池開發(fā)取得了顯著進步，各類創(chuàng)新技術(shù)不斷涌現(xiàn)，電池性能不斷提升，充電更快速、更安全，電池也更加長壽耐用。在這些技術(shù)中，使用人工智能技術(shù)（AI）優(yōu)化電池管理系統(tǒng)（BMS）是最近的熱門話題之一。Eatron Technologies 是一家總部位于英國的公司，專門從事電動和自動駕駛汽車的智能軟件開發(fā)。公司總經(jīng)理 Umut Genc 博士表示，這是人工智能的一個創(chuàng)新應(yīng)用領(lǐng)域。
“我們認為這是一種借助選擇性人工智能技術(shù)的基于物理建模的方法。” Genc 博士說，“對此，了解人工智能的局限性至關(guān)重要。首先，AI 系統(tǒng)有很高的內(nèi)存和處理能力需求，這無疑會增加成本和能耗。其次，AI 技術(shù)的應(yīng)用仍處于早期階段，因此通常被認為并非萬無一失。目前，任何完全依賴 AI 技術(shù)的關(guān)鍵功能系統(tǒng)都很難獨立通過汽車行業(yè)的標準驗證。”
Genc 回答說，這兩個問題的答案是一致的，那就是首先要正確設(shè)定對人工智能技術(shù)的期待。Eatron 公司的技術(shù)總監(jiān) Can Kurtulus 強調(diào)稱，OEM 必須認識到，電池化學技術(shù)不會在未來短時間內(nèi)出現(xiàn)長足進步。這種情況下，只能通過優(yōu)化電池管理來提升電池的耐久性、有效能量密度和充電速度。
Kurtulus 進一步認為，優(yōu)化電池管理的關(guān)鍵是應(yīng)用基于設(shè)備實體的軟件，即采用基于嵌入式系統(tǒng)模型的控制系統(tǒng)，這種系統(tǒng)使用嵌入式軟件模型取代簡單的查找表。
“在車輛設(shè)計的其他領(lǐng)域，數(shù)字模擬有助于避免發(fā)生過度工程的情況，形成一種‘良性循環(huán)’，從而降低成本和重量。”Kurtulus 說，“在電池管理系統(tǒng)設(shè)計中應(yīng)用人工智能技術(shù)的情況很類似，不同之處是我們必須把模型嵌入每一個電池組的控制系統(tǒng)中，并將其應(yīng)用至車輛的完整使用壽命中，而非僅應(yīng)用在設(shè)計階段。因此，可以說這不僅僅是常規(guī)的 AI 技術(shù)。”
在一家未公布名稱的德國一級供應(yīng)商以及英國和土耳其的工程團隊的支持下，Eatron 開展了與汽車底盤管理系統(tǒng)和 SAE L2 級自動駕駛輔助系統(tǒng)相關(guān)的初始生產(chǎn)項目。不過，如今電池管理系統(tǒng)已經(jīng)成為了Eatron 的重點研發(fā)項目之一。Eatron 聲稱，公司應(yīng)用選擇性 AI 技術(shù)的直接好處是無需再對電池進行過度管理，但同樣可以保護電池耐用性，從而降低保修成本。
“目前，對電池壽命影響最大的因素是荷電狀態(tài)變化窗口，也就是電池過沖和過放的極限，其次就是充電速率。” Kurtulus 表示，“正是如此，如今的大多數(shù)量產(chǎn)電池管理系統(tǒng)均經(jīng)過專門校準，會堅決避免電池出現(xiàn)過沖或過放的情況，即使車輛幾公里外就有充電站也無法多行駛一公里，并始終嚴格限制快充功能的使用，從而最大限度地避免發(fā)生保修索賠的情況。”
Eatron 的另一項創(chuàng)新是使用其基于物理的建模技術(shù)，預(yù)測電池的剩余使用壽命（RUL）。這家公司有望成為首家可為量產(chǎn)鋰離子電池提供剩余使用壽命預(yù)測功能的供應(yīng)商。公司還計劃率先為汽車 OEM 的車輛診斷系統(tǒng)增加剩余使用壽命的指標。
Genc 指出，將車輛的電池管理系統(tǒng)接入云可以讓公司的選擇性 AI 技術(shù)充分發(fā)揮其主要優(yōu)勢。“電池剩余使用壽命預(yù)測功能允許車輛 OEM 及時發(fā)布 OTA 校準更新，以最大限度地保護客戶利益，例如放開快速充電和提升續(xù)航里程等，但同時有效控制保修索賠成本，保證電池的耐用性。”
隨著分析數(shù)據(jù)的不斷累計，人工智能技術(shù)也會進行持續(xù)學習，從而提供更加準確的預(yù)測，進而作出更加全面的決策。車輛 OEM 也有機會借此推出新功能，為車主帶來額外收益。
舉個例子，如今車主在購買二手車時很難確定車輛的電池剩余使用壽命。Genc 說，“在處理轉(zhuǎn)讓車輛時，能否查看并驗證電池剩余使用壽命非常重要，特別是對于車隊來說。我們可以在車輛儀表盤上增加顯示電池的剩余使用壽命。” Eatron 的業(yè)務(wù)發(fā)展總監(jiān) Amedeo Bianchimano 補充說，“人工智能和云連接的結(jié)合可以讓電池管理軟件發(fā)揮更重要的作用，不僅可以最大限度地優(yōu)化電動汽車的性能，還有助于提供主動保修管理等更多附加軟件服務(wù)。”
Eatron 公司與英國華威大學下屬華威大學制造工程學院（WMG）開展了一個聯(lián)合研究項目，探索電池管理系統(tǒng)可以在車輛故障預(yù)測方面發(fā)揮的更大作用?，F(xiàn)階段，電池管理系統(tǒng)只能在電芯已經(jīng)出現(xiàn)故障或即將發(fā)生故障時（提前幾秒）發(fā)揮作用，即關(guān)閉電池組或讓汽車進入“緩行回家”模式。
與華威大學制造工程學院的聯(lián)合研究主要是為了改進一項汽車故障預(yù)測方面的創(chuàng)新，幫助 Eatron 的系統(tǒng)在故障發(fā)生前有效發(fā)現(xiàn)電池相關(guān)問題，但不是像目前這樣僅提前幾秒鐘發(fā)現(xiàn)，而是提前幾個月前就能發(fā)現(xiàn)問題。
“通過將這項技術(shù)集成至電池組中，可以幫助電池組發(fā)揮更大的化學潛能。”Kurtulus 強調(diào)說，“充分發(fā)揮電池的化學潛力一直是電池領(lǐng)域的努力方向之一，但之前只能靠增加很多非常復雜的硬件設(shè)備實現(xiàn)。我們的思路是將機器學習技術(shù)和成熟的信號處理理論結(jié)合應(yīng)用，避免信號雜音對汽車故障診斷的影響，而且無需增加任何額外組件。”
按照 Eatron 的說法，車主還可以借助電池管理系統(tǒng)校準技術(shù)，享受到更長的續(xù)航里程和更快的充電速度，“并同時降低車輛臨時故障拋錨的風險。”Genc 表示，“我們的方法是使用經(jīng)過驗證的基于模型的控制技術(shù)，創(chuàng)建一個穩(wěn)健的系統(tǒng)，然后根據(jù)具體需求有針對性的應(yīng)用人工智能技術(shù)。人工智能技術(shù)在電池健康管理中的應(yīng)用只是一個起點，真正令人興奮的是之后即將發(fā)生的事情！”
作者：Stuart Birch
來源：SAE《汽車工程》雜志
After two decades of significant advances in electric vehicle (EV) battery development, innovations continue to bring promises of greater battery performance, safety, faster charging, longevity and durability. One area of fresh thinking concerns battery management systems (BMS) using artificial intelligence (AI) as an enabler. Or more specifically, a novel dimension of AI, according to Dr. Umut Genc, managing director of Eatron Technologies, a U.K.-based company specializing in intelligent software for electric and autonomous vehicles.
“We define [it] as physics-based modeling with selective AI,” he said. “It is vital to understand AI’s limitations. First, the amount of memory and processing power required by an AI system is substantial, increasing cost and energy consumption. Second, AI is still in the early stages of application, so it is often seen to be not without risks. Any mission-critical system that is solely dependent on AI is currently difficult to validate to automotive standards,” Genc explained.
The solution, he noted, is the same for both challenges, starting with understanding the value that AI brings to objectives. Eatron’s technical director, Can Kurtulus, underlined the importance of OEMs recognizing that there will not be a huge step in battery chemistry for some time. Therefore, many improvements in durability, usable energy density and charging speed must come from improved battery management.
Kurtulus said he believes the key to be the application of physics-based software – control systems with embedded system models. This is a control technique in which look-up tables are replaced by an embedded software model of the system.
“In other areas of vehicle design, accurate digital simulations have helped to eliminate the need for over-engineering. This has brought down cost and weight in a ‘virtuous circle,’” he said. “Our approach to BMS is similar, except we will be embedding the model in the control system of every battery pack and applying it in real time throughout the vehicle’s life, not just during the design stage, so rather more than regular AI technology.”
First for EV diagnostics
With the backing of an unnamed German Tier 1 supplier and engineering teams in the U.K. and Turkey, Eatron’s initial production programs concerned chassis management systems and SAE Level 2 automated driver-assist systems. But BMS has now become a major R&D program. The immediate benefits claimed for Eatron’s selective AI applications are the reduced need to over-specify the battery to protect durability and hence lower warranty costs.
“The biggest influences on battery life are currently the state of charge window: how close to full charge and full discharge the battery is allowed to go, and then the rate of charge,” Kurtulus said. “So, in today’s production systems, warranty claims are minimized by calibrating the BMS to prevent the battery being charged to full capacity, to never approach deep discharge even if the vehicle is only a few miles from a charging station, and to always restrict the fast-charge capability.”
An Eatron innovation is to use its physics-based modeling techniques to predict the remaining useful life (RUL) of the battery. The company expects to be the first supplier of production technologies offering this feature for lithium-ion EV batteries. It also aims to be first to offer vehicle OEMs an opportunity to include a RUL indicator in the vehicle diagnostics.
Genc noted selective AI’s major benefits will be fully gained when the BMS is connected to the cloud. “Prediction of the RUL will allow the vehicle manufacturer to issue over-the-air calibration updates that maximize customer benefits, such as fast charging and range,” he noted, “while remaining confident that warranty and durability targets will be met.”
As more data is analyzed, the AI learns; predictions become more accurate and the benefits are automatically incorporated in the decision-making. There is also potential for the vehicle OEM to add new features that bring additional benefits for owners. For example, there is currently no way for a used vehicle buyer to determine the RUL of an EV’s battery.
“For fleets especially, an ability to manage and then validate the RUL will add significant value when they come to dispose of their vehicles. A dashboard RUL indicator could be fitted,” Genc stated. Added Eatron’s business development director, Amedeo Bianchimano: “The combination of AI and a connection to the cloud will give battery management software a much more important role, not just as a tool for maximizing e-vehicle attributes but also in offering warranty management and additional software services.”
More detailed prognostics
Eatron has established a joint research project with the U.K.’s Warwick Manufacturing Group (WMG), a department of The University of Warwick, to extend their BMS development into more detailed prognostics. At present, the system can only identify problems with battery cells as they occur (typically no more than a few seconds before failure), shutting down the battery pack or putting it into limp-home mode.
The objective of the research with WMG is to refine an innovation in diagnostics that allows Eatron to identify cell-related problems usefully before failure – not seconds before as is typically the situation today, but up to several months ahead of a failure.
“With this technology embedded in a battery pack, the chemistry can be driven much harder,” stressed Kurtulus. “This has been a goal for many years but previously it required considerable added hardware complexity. Our approach applies machine learning to established signal processing theory to isolate the diagnostic characteristic from the noise, without any additional components.”
For the vehicle owner, the benefits claimed by Eatron will also include BMS calibrations that deliver more range and faster charging, “while simultaneously reducing the risk of being left at the side of the road with a disabled vehicle,” Genc said. “Our approach is to create a robust system using proven model-based control technologies, then to carefully add in AI where it is needed for specific calculations. Battery health management is just the starting point. The real excitement is what comes next!”
Eatron Technologies 基于云的網(wǎng)聯(lián)電池架構(gòu)。（圖片來源：Eatron Technologies)
Eatron Technologies 電池管理系統(tǒng)的 AI 電芯診斷。（圖片來源：Eatron Technologies）
Eatron Technologies 公司高管（左起）Can Kurtulus、Umut Genc博士和 Amedeo Bianchimano 介紹了選擇性 AI 技術(shù)的潛力。（圖片來源：Eatron Technologies）