加拿大家园论坛

周末技术宅男兴趣帖: 太空叉猎鹰火箭为什么能够省钱还可靠 (ZT)

原文链接:https://forum.iask.ca/threads/907524/

金角大王 : 2020-06-26#1
在航天圈里,有一种器件如贵族般存在:宇航级器件。

一个二极管只要上天验证成功,就可以从一个工业级十八线小明星跃升为宇航级一线大
明星,身价可以倍增上百倍甚至上万倍!

以现有载人飞船搭载的星载计算机和控制器举例,单个控制器价格为500万人民币左右
,一共14个系统,为了追求高可靠性,每个系统1+1备份,一共28个控制器,成本总计
约1.4亿人民币!

而SpaceX的龙飞船主控系统的芯片组,仅用了2.6万人民币,成本相差5384倍!

Elon Musk到底是如何做到的?

我们看以下几条重要的知识点:

SpaceX 猎鹰九号和龙飞船用的都是Intel双核的x86处理器;
操作系统用的是Linux,还有LabView和Matlab;
软件工程用的是C++,有些时候也用Python;
整个主控程序只有几十万行代码。

工业级器件小屌丝的困境:粒子翻转

航天器所有的器件要经历很苛刻很苛刻的环境。

首先发射时要禁得住剧烈的抖动和很高的温度,才能走出地球。

而真正的炼狱在入轨后才刚刚开始,面对太阳面的时候,温度迅速提升,最高到120°C
;背离太阳面的时候,温度骤减,最低到-150°C。

就这样90分钟一圈又一圈,周而复始,每圈都是270°C的温差。

而对于电子器件来说,温度不是最难熬的,最难熬的是太空中的辐射:

这些辐射有来自地球的召唤:地球磁场
也有来自太阳的问候:高能粒子
还可能有来自三体文明的问候:其他太阳系以外的粒子

而这些粒子,将引发电子器件的神经紊乱,专业名词是:粒子翻转。

它将很Surprise地告诉星载计算机和星载存储器:

“下面将是见证奇迹发生的一刻!”

“我要把1变成0,然后再把0变成1。”

有些人问了,多大点事啊,不就差个1吗?!

但是在比特界,差一位就可差之千里。

举个栗子:

如果指令20是向上爬升,指令24是停止推进,后果是难以想象的。

所以如果发生了1和0不分的情况,整个飞行器的运算结果曾导致非常大的灾难。

在1996年,阿里安501火箭,虽然没有粒子翻转,但是系统试图将一个64位的数字,放
到一个16位的地址里面去,随即发生了1/0错乱的现象。

结果在点火37秒后,火箭开始侧翻,随之爆炸,因为这个“小”问题,那次发射损失高
达3.7亿美金!

回到主题,既然粒子翻转这么恐怖,那SpaceX如何做到发现问题和解决问题的呢?

很简单:民主决策

技术名词叫:parity bits

同位位元

既然判断不了一个是否翻转,那就多放几个一样的设备,通过比较,把不一样的结果给
踢出去。

攒火箭硬件选择

上文提到,SpaceX没有选择用贵族宇航级器件,而是选择了经典厂牌Intel的X86双核处
理器,京东售价仅478元人民币(参考价为奔腾系列,赛扬更便宜):


而SpaceX也没有用双核做一件事,而是把双核拆成了两个单核,分别计算同样的数据。

每个系统配置3块芯片做冗余,也就是6个核做计算。


如果其中1个核的数据和其他5个核不同,那么主控系统会告诉这个核重新启动,再把其
他5个核的数据拷贝给重启的核,从而达到数据一直同步。

周而复始,不让一个核掉队。


据SpaceX前火箭总师John Muratore透露,龙飞船一共有18个系统,每个系统配置了3块
X86芯片,龙飞船一共有54块。

所以龙飞船主控芯片的总价约:2.6万人民币,3600美元!

而猎鹰九号一共有9个分立式发动机,每个发动机配置了3块X86芯片,加上主控系统配
置了3块,猎鹰九号一共有30块这样的芯片。

猎鹰九号主控芯片的总价约:1.4万人民币,2000美元!

我差点砸了手里的X1,是它阻拦了我攒火箭的大计!

更让SpaceX开心的事情,是Intel X86的程序员一抓一大把啊,而专业宇航级器件的程
序用的基本都是特定语言,程序员比元器件还难找。

而且硬件工程师压力也小,X86芯片随便造,烧坏了?再来1个。不不,再买一打!

可是宇航级器件仅仅是测试费,就都够再买一车X86芯片的。

攒火箭软件选择

SpaceX就用的开源Linux写的操作系统,而Linux用随便一台电脑就可以编写。

同样的,SpaceX程序员最爱的还是C++,用开源的GCC或者GDB做火箭的主控程序。

SpaceX还用LabView,一款图形化编辑语言,对于火箭程序来讲,它更容易实现可视化
和流程化,更容易做复杂的算法设计和数据分析。

SpaceX也用Matlab,在仿真和矩阵计算上,真的很好用。

而且,龙飞船,猎鹰九号,猎鹰重型,分享着同一款代码,分享着同一类迭代,分享着
同一种喜悦,多么的模块化,多么的互联网…

大数据监控和测试

2018年,SpaceX一共发射21次,一个公司占全球发射数量约20%,而SpaceX的工程师和
分析师,手里有大量的测试数据和实际数据,而且他们也被鼓励用不同的维度,去检验
飞行器的安全性,形成最新的也最实用的测试程序,从而降低实测成本。

同时,Continuous Integration,持续集成也被应用在了程序测试上。

注:持续集成

为了配合敏捷开发(相对于瀑布开发)的速度和效率而产生的一个用于编译、测试、发
布、部署的工具。

通过这种办法,可以让团队每时每刻在持续的基础上,收到反馈并进行改进,不必等到
开发周期后期才寻找和修复缺陷。

而且火箭程序不同于其他,会进行“断弦式”测试,突然关闭一台电脑,来看看发动机
到底有什么反应。

总结

航天已经经历了60年的历史,每一次阶跃其实都伴随着各类器件技术革新,比如:

1950年代的晶体管技术;
1970年代的微控制器技术;
1980年代的数字信号处理技术;
1990年代的高性能存储技术。

现在,芯片工艺从28nm,16nm,10nm到7nm,工艺的提升也增加了芯片在太空中的抗辐
射性能,让商业器件在太空中应用可行性大大提升!

伴随航天成长的是经典的:摩尔定律。

但是摩尔定律到现在在地面侧都快失效了,而在航天侧还没有开始。

比如Greg Wyler在2019年1月6号,Twitter的Oneweb的新型相控阵天线,目标定价15美
金。

比如AWS与Lockheed Martin在2018年11月发布的超小型地面站,可降低地面站80%的成
本。

北京九天微星正在研制200mW卫星物联网终端模组,目标定价5美金。

因此,航天缺少的仅仅是大胆的商业器件验证,缺少的仅仅是采用MVP快速迭代的环境
,而逐年降低的发射成本正在迅速降低试验成本,因此:

属于航天的摩尔定律才刚刚开始!

属于航天的互联网思维才刚刚开始!

属于航天的大时代才刚刚开始!

大大哥 : 2020-06-26#2
感觉老有道理。
“航天摸耳定律”如果是大王首创,不如改为“金角定律”。
其余类推,金角思维、金角时代

Timmocha : 2020-06-26#3
看到敏捷开发,我不厚道地笑了。

金角大王 : 2020-06-26#4
很遗憾,不是
君子不能夺人之美
感觉老有道理。
“航天摸耳定律”如果是大王首创,不如改为“金角定律”。
其余类推,金角思维、金角时代

小暑 : 2020-06-26#5
不明觉厉

风雪夜 : 2020-06-26#6
文科女看了十行就开始蒙圈。。。。不明觉厉
每个时代都会出几个马斯克这样的大头蒜,加油啊帅哥。

shw019 : 2020-06-26#7
文科女看了十行就开始蒙圈。。。。不明觉厉
每个时代都会出几个马斯克这样的大头蒜,加油啊帅哥。
前10行我也蒙了, 放弃顺序用倒序, 看明白后7行就行。

风雪夜 : 2020-06-26#8
前10行我也蒙了, 放弃顺序用倒序, 明白看后7行就行。
你这个办法还真不错!最后一段一般是总结性发言。

金角大王 : 2020-06-26#9
说不定是:
待续 :ROFLMAO:
你这个办法还真不错!最后一段一般是总结性发言。

geomarb : 2020-06-26#10

geomarb : 2020-06-26#11
Hardware

Launch vehicles[edit]
Main article: SpaceX launch vehicles

The landing of a Falcon 9 Block 5 first stage at Cape Canaveral in July 2019 – VTVL technologies are utilized in many of SpaceX's launch vehicles.
Falcon 1 was a small rocket capable of placing several hundred kilograms into low Earth orbit.[115] It functioned as an early test-bed for developing concepts and components for the larger Falcon 9.[115] Falcon 1 attempted five flights between 2006 and 2009. With Falcon I, when Musk announced his plans for it before a subcommittee in the Senate in 2004, he discussed that Falcon I would be the 'worlds only semi-reusable orbital rocket' apart from the Space Shuttle.[116] On September 28, 2008, on its fourth attempt, the Falcon 1 successfully reached orbit, becoming the first privately funded, liquid-fueled rocket to do so.[117]

Falcon 9 is an EELV-class medium-lift vehicle capable of delivering up to 22,800 kilograms (50,265 lb) to orbit, and is intended to compete with the Delta IV and the Atlas V rockets, as well as other launch providers around the world. It has nine Merlin engines in its first stage.[118] The Falcon 9 v1.0 rocket successfully reached orbit on its first attempt on 2010-06-04. Its third flight, COTS Demo Flight 2, launched on 2012-05-22, and was the first commercial spacecraft to reach and dock with the International Space Station.[119] The vehicle was upgraded to Falcon 9 v1.1 in 2013, Falcon 9 Full Thrust in 2015, and finally to Falcon 9 Block 5 in 2018. As of 23 March 2020, the Falcon 9 family have flown 84 successful missions with one failure, one partial success, and one vehicle destroyed during a routine test several days prior to a scheduled launch.

In 2011, SpaceX began development of the Falcon Heavy, a heavy-lift rocket configured using a cluster of three Falcon 9 first stage cores with a total 27 Merlin 1D engines and propellant crossfeed.[120][121] The Falcon Heavy successfully flew on its inaugural mission on February 6, 2018 with a payload consisting of Musk's personal Tesla Roadster into heliocentric orbit[122] The first stage would be capable of lifting 63,800 kilograms (140,660 lb) to LEO with the 27 Merlin 1D engines producing 22,819 kN of thrust at sea level, and 24,681 kN in space. At the time of its first launch, SpaceX described their Falcon Heavy as "the world's most powerful rocket in operation".[123]

geomarb : 2020-06-26#12
Rocket engines[edit]
Main article: SpaceX rocket engines

The Merlin 1D engine, SpaceX's most numerous engine, undergoing testing at SpaceX's Rocket Development and Test Facility in McGregor, Texas.
Since the founding of SpaceX in 2002, the company has developed three families of rocket enginesMerlin and the retired Kestrel for launch vehicle propulsion, and the Draco control thrusters. SpaceX is currently developing two further rocket engines: SuperDraco and Raptor. SpaceX is currently the world's most prolific producer of liquid fuel rocket engines.[124] Merlin is a family of rocket engines developed by SpaceX for use on their launch vehicles. Merlin engines use LOX and RP-1 as propellants in a gas-generator power cycle. The Merlin engine was originally designed for sea recovery and reuse. The injector at the heart of Merlin is of the pintle type that was first used in the Apollo Program for the lunar module landing engine. Propellants are fed via a single shaft, dual impeller turbo-pump. Kestrel is a LOX/RP-1 pressure-fed rocket engine, and was used as the Falcon 1 rocket's second stage main engine. It is built around the same pintle architecture as SpaceX's Merlin engine but does not have a turbo-pump, and is fed only by tank pressure. Its nozzle is ablatively cooled in the chamber and throat, is also radiatively cooled, and is fabricated from a high strength niobium alloy. Both names for the Merlin and Kestrel engines are derived from species of North American falcons: the kestrel and the merlin.[125]

Draco engines are hypergolic liquid-propellant rocket engines that utilize monomethyl hydrazine fuel and nitrogen tetroxide oxidizer. Each Draco thruster generates 400 newtons (90 lbf) of thrust.[126] They are used as reaction control system (RCS) thrusters on the Dragon spacecraft.[127] SuperDraco engines are a much more powerful version of the Draco thrusters, which were initially meant to be used as landing and launch escape system engines on Dragon 2. The concept of using retro-rockets for landing was scrapped in 2017 when it was decided to perform a traditional parachute descent and splashdown at sea.[128] Raptor is a new family of methane-fueled full-flow staged combustion cycle engines to be used in its future Starship launch system.[129] Development versions were test fired in late 2016.[130] On April 3, 2019, SpaceX conducted a successful static fire test in Texas on its Starhopper vehicle, which ignited the engine while the vehicle remained tethered to the ground.[131] On July 24, 2019, SpaceX conducted a successful test hop of 20 meters of its Starhopper.[132] On August 28, 2019, Starhopper conducted a successful test hop of 150 meters.[133]

geomarb : 2020-06-26#13
Dragon spacecraft[edit]

The SpaceX's Crew Dragon spacecraft, designed to deliver crew to and from the International Space Station as part of the Commercial Crew Program.
In 2005, SpaceX announced plans to pursue a human-rated commercial space program through the end of the decade.[134] The Dragon is a conventional blunt-cone ballistic capsule which is capable of carrying cargo or up to seven astronauts into orbit and beyond.[135][135] In 2006, NASA announced that the company was one of two selected to provide crew and cargo resupply demonstration contracts to the ISS under the COTS program.[136] SpaceX demonstrated cargo resupply and eventually crew transportation services using the Dragon.[119] The first flight of a Dragon structural test article took place in June 2010, from Launch Complex 40 at Cape Canaveral Air Force Station during the maiden flight of the Falcon 9 launch vehicle; the mock-up Dragon lacked avionics, heat shield, and other key elements normally required of a fully operational spacecraft but contained all the necessary characteristics to validate the flight performance of the launch vehicle.[137] An operational Dragon spacecraft was launched in December 2010 aboard COTS Demo Flight 1, the Falcon 9's second flight, and safely returned to Earth after two orbits, completing all its mission objectives.[138] In 2012, Dragon became the first commercial spacecraft to deliver cargo to the International Space Station,[119] and has since been conducting regular resupply services to the ISS.[139]

In April 2011, NASA issued a $75 million contract, as part of its second-round commercial crew development (CCDev) program, for SpaceX to develop an integrated launch escape system for Dragon in preparation for human-rating it as a crew transport vehicle to the ISS.[140] In August 2012, NASA awarded SpaceX a firm, fixed-price SAA with the objective of producing a detailed design of the entire crew transportation system. This contract includes numerous key technical and certification milestones, an uncrewed flight test, a crewed flight test, and six operational missions following system certification.[141] The fully autonomous Crew Dragon spacecraft is expected to be one of the safest crewed spacecraft systems. Reusable in nature, the Crew Dragon will offer savings to NASA.[141] SpaceX conducted a test of an empty Crew Dragon to ISS in early 2019, and later in the year they plan to launch a crewed Dragon which will send US astronauts to the ISS for the first time since the retirement of the Space Shuttle.[142][143] In February 2017, SpaceX announced that two would-be space tourists had put down "significant deposits" for a mission which would see the two tourists fly on board a Dragon capsule around the Moon and back again.

In addition to SpaceX's privately funded plans for an eventual Mars mission, NASA Ames Research Center had developed a concept called Red Dragon: a low-cost Mars mission that would use Falcon Heavy as the launch vehicle and trans-Martian injection vehicle, and the Dragon capsule to enter the Martian atmosphere. The concept was originally envisioned for launch in 2018 as a NASA Discovery mission, then alternatively for 2022.[144] The objectives of the mission would be return the samples from Mars to Earth at a fraction of the cost of the NASA own return-sample mission now projected at 6 billion dollars.[144][145] In September 2017, Elon Musk released first prototype images of their space suits to be used in future missions. The suit is in testing phase and it is designed to cope with 2 atm (200 kPa; 29 psi) pressure in vacuum.[146][147] The Crew Dragon spacecraft was first sent to space on March 2, 2019.

On March 27, 2020, SpaceX revealed the Dragon XL resupply spacecraft to carry pressurized and unpressurized cargo, experiments and other supplies to NASA's planned Lunar Gateway under a Gateway Logistics Services (GLS) contract.[148] The equipment delivered by Dragon XL missions could include sample collection materials, spacesuits and other items astronauts may need on the Gateway and on the surface of the Moon, according to NASA. It will launch on SpaceX Falcon Heavy rockets from pad 39A at the Kennedy Space Center in Florida. The Dragon XL will stay at the Gateway for six to 12 months at a time, when research payloads inside and outside the cargo vessel could be operated remotely, even when crews are not present.[149] Its payload capacity is expected to be more than 5,000 kilograms (11,000 lb) to lunar orbit.[150]

geomarb : 2020-06-26#14
Research and development[edit]

First test firing of a scale Raptor development engine in September 2016 in McGregor, Texas.
SpaceX is actively pursuing several different research and development programs. Most notable are those intended to develop a fully reusable launch vehicle called Starship and a global telecommunications network called Starlink.

Reusable launch system[edit]

An autonomous spaceport drone ship in position prior to Falcon 9 Flight 17 carrying CRS-6.
Main articles: SpaceX reusable launch system development program and Falcon 9 first-stage landing tests
SpaceX's reusable launcher program was publicly announced in 2011 and the design phase was completed in February 2012. The system returns the first stage of a Falcon 9 rocket to a predetermined landing site using only its own propulsion systems.[151]

SpaceX's active test program began in late 2012 with testing low-altitude, low-speed aspects of the landing technology. The prototypes of Falcon 9 performed vertical takeoffs and landings.

High-velocity, high-altitude aspects of the booster atmospheric return technology began testing in late 2013 and have continued through 2018, with a 98% success rate to date. As a result of Elon Musk's goal of crafting more cost-effective launch vehicles, SpaceX conceived a method to reuse the first stage of their primary rocket, the Falcon 9,[152] by attempting propulsive vertical landings on solid surfaces. Once the company determined that soft landings were feasible by touching down over the Atlantic and Pacific Ocean, they began landing attempts on a solid platform. SpaceX leased and modified several barges to sit out at sea as a target for the returning first stage, converting them to autonomous spaceport drone ships (ASDS). SpaceX first achieved a successful landing and recovery of a first stage in December 2015,[153] and in April 2016, the first stage booster first successfully landed on the ASDS Of Course I Still Love You.[154][155]

SpaceX continues to carry out first stage landings on every orbital launch that fuel margins allow. By October 2016, following the successful landings, SpaceX indicated they were offering their customers a ten percent price discount if they choose to fly their payload on a reused Falcon 9 first stage.[156] On March 30, 2017, SpaceX launched a "flight-proven" Falcon 9 for the SES-10 mission. This was the first time a re-launch of a payload-carrying orbital rocket went back to space.[78][157] The first stage was recovered and landed on the ASDS Of Course I Still Love You in the Atlantic Ocean, also making it the first landing of a reused orbital class rocket. Elon Musk called the achievement an "incredible milestone in the history of space."[158][159]

The autonomous spaceport drone ships are named after giant starships from the Culture series stories by science fiction author Iain M. Banks.[

geomarb : 2020-06-26#15
Starship[edit]
Main articles: SpaceX Mars transportation infrastructure, SpaceX Starship, and Starship development history

Static fire of Starship SN4.
SpaceX is developing a super-heavy lift launch system, Starship. Starship is a fully reusable second stage and space vehicle intended to replace all of the company's existing launch vehicle hardware by the early 2020s; plus ground infrastructure for rapid launch and relaunch and zero-gravity propellant transfer technology in low Earth orbit (LEO).

SpaceX initially envisioned a 12-meter-diameter ITS concept in 2016 which was solely aimed at Mars transit and other interplanetary uses. In 2017, SpaceX articulated a smaller 9-meter-diameter BFR to replace all of SpaceX launch service provider capabilities—Earth-orbit, lunar-orbit, interplanetary missions, and potentially, even intercontinental passenger transport on Earth—but do so on a fully reusable set of vehicles with a markedly lower cost structure.[161] A large portion of the components on Starship are made of 301 stainless steel. Private passenger Yusaku Maezawa has contracted to fly around the Moon in Starship in 2023.[162][163][164]

Musk's long-term vision for the company is the development of technology and resources suitable for human colonization on Mars. He has expressed his interest in someday traveling to the planet, stating "I'd like to die on Mars, just not on impact."[165] A rocket every two years or so could provide a base for the people arriving in 2025 after a launch in 2024.[166][167] According to Steve Jurvetson, Musk believes that by 2035 at the latest, there will be thousands of rockets flying a million people to Mars, in order to enable a self-sustaining human colony.[16

geomarb : 2020-06-26#16
Other projects[edit]
Main articles: Starlink and Hyperloop pod competition
In January 2015, SpaceX CEO Elon Musk announced the development of a new satellite constellation, called Starlink, to provide global broadband internet service. In June 2015, the company asked the federal government for permission to begin testing for a project that aims to build a constellation of 4,425 satellites capable of beaming the Internet to the entire globe, including remote regions which currently do not have Internet access.[169][170] The Internet service would use a constellation of 4,425 cross-linked communications satellites in 1,100 km orbits. Owned and operated by SpaceX, the goal of the business is to increase profitability and cashflow, to allow SpaceX to build its Mars colony.[171] Development began in 2015, initial prototype test-flight satellites were launched on the SpaceX PAZ mission in 2017. Initial operation of the constellation could begin as early as 2020. As of March 2017, SpaceX filed with the US regulatory authorities plans to field a constellation of an additional 7,518 "V-band satellites in non-geosynchronous orbits to provide communications services" in an electromagnetic spectrum that had not previously been "heavily employed for commercial communications services". Called the "V-band low-Earth-orbit (VLEO) constellation", it would consist of "7,518 satellites to follow the [earlier] proposed 4,425 satellites that would function in Ka- and Ku-band".[172] In February 2019, SpaceX formed a sibling company, SpaceX Services, Inc., to license the manufacture and deployment of up to 1,000,000 fixed satellite earth stations that will communicate with its Starlink system.[173] In May 2019, SpaceX launched the first batch of 60 satellites aboard a Falcon 9 from Cape Canaveral, FL.[174]

In June 2015, SpaceX announced that they would sponsor a Hyperloop competition, and would build a 1-mile-long (1.6 km) subscale test track near SpaceX's headquarters for the competitive events.[175][176] The first competitive event was held at the track in January 2017, the second in August 2017 and the third in December 2018.[177][178][179

金角大王 : 2020-06-26#17
谢谢geomarb的具体信息

geomarb : 2020-06-26#18
谢谢geomarb的具体信息
(y)不客气, 我对航天业了解的不多, 贴出来一起看看。:LOL:。。

luf_7_6 : 2020-06-26#19
控制系统出故障的远不如执行机构故障的多,比如挑战者o型圈失效,哥伦比亚的密封层损坏

金角大王 : 2020-06-26#20
有道理,但这篇文章应该还有个重点,控制系统的成本大大降低了
控制系统出故障的远不如执行机构故障的多,比如挑战者o型圈失效,哥伦比亚的密封层损坏

zunhuhu : 2020-06-26#21
"系统试图将一个64位的数字,放到一个16位的地址里面去 ",跟787Max一样,一个软件小错误导致大灾难。硬件错误可以通过冗余屏蔽,软件错误才是防不胜防。
既然SpaceX已经证明非宇航级器件可用于航天,那连摩尔定律都不需要了,直接划时代。

geomarb : 2020-06-26#22
有道理,但这篇文章应该还有个重点,控制系统的成本大大降低了
控制系统的一部分信息:
Most launch vehicles are not controlled manually, if that is what you are asking. Conventionally, companies that develop launch vehicles have all physical "steering" done in a combination of two methods:
  1. Electronic sensor systems that provide awareness of the vehicle's position, speed and etc. This information is then used with guidance and navigation software, which is more akin to where the vehicle knows that it "needs" to go. Now it compares the two and if the vehicle is off trajectory it will use the below.
  2. Engines on gimbals and an attitude control system. Engines on gimbals essentially means the vehicle can alter the angle of their engines in relation to the rocket for directional changes. An attitude control system (ACS) is essentially a system of "spouts" around the vehicle (in every direction) that expel high-pressure gasses usually to make it spin in certain ways or turn harder
SpaceX has stated that the vehicle must first "turn-around" in space after the first stage engine cuts off, therefore it is unlikely that "gimbaling" an engine is their methodology.

However, if they were to have their software know that the vehicle must essentially pull a 180, it can use its ACS to flip the rocket around before engaging the engine to begin travelling back towards where it launched (or to eliminate its horizontal velocity).

Now the rocket is hurtling back to Earth, face first. In order to survive through the atmosphere at high-speeds it must do so in a way that the rocket can take the force and temperature. So now we need the vehicle to enter "bottom first".

It now likely cuts its engines once again, repeats maneuver once again to "back into" the atmosphere. From here, steering is done in combination of gimbal-ing engines when they are reactivated, using the ACS as well as using these bad boys:

SpaceX uses optimal control scheme for its landing. The goal of this scheme is to reach zero altitude with zero vertical velocity given tight constraints of landing area and fuel. This makes it a non-linear optimal control problem.

The algorithm running behind the landing is known as G-FOLD (guidance for optimal large diverts) which uses loseless convexification for converging globally optimal solution. This algorithm was co-developed by Lars Blackmore (SpaceX principal landing engineer) in JPL and was originally designed for mars rover for powered descent.

Falcon 9 uses high end GPUs for its math processing.

Aidemengdun : 2020-06-27#23
绝对赞同

就是太短。技术宅不够看呀
在航天圈里,有一种器件如贵族般存在:宇航级器件。

一个二极管只要上天验证成功,就可以从一个工业级十八线小明星跃升为宇航级一线大
明星,身价可以倍增上百倍甚至上万倍!

以现有载人飞船搭载的星载计算机和控制器举例,单个控制器价格为500万人民币左右
,一共14个系统,为了追求高可靠性,每个系统1+1备份,一共28个控制器,成本总计
约1.4亿人民币!

而SpaceX的龙飞船主控系统的芯片组,仅用了2.6万人民币,成本相差5384倍!

Elon Musk到底是如何做到的?

我们看以下几条重要的知识点:

SpaceX 猎鹰九号和龙飞船用的都是Intel双核的x86处理器;
操作系统用的是Linux,还有LabView和Matlab;
软件工程用的是C++,有些时候也用Python;
整个主控程序只有几十万行代码。

工业级器件小屌丝的困境:粒子翻转

航天器所有的器件要经历很苛刻很苛刻的环境。

首先发射时要禁得住剧烈的抖动和很高的温度,才能走出地球。

而真正的炼狱在入轨后才刚刚开始,面对太阳面的时候,温度迅速提升,最高到120°C
;背离太阳面的时候,温度骤减,最低到-150°C。

就这样90分钟一圈又一圈,周而复始,每圈都是270°C的温差。

而对于电子器件来说,温度不是最难熬的,最难熬的是太空中的辐射:

这些辐射有来自地球的召唤:地球磁场
也有来自太阳的问候:高能粒子
还可能有来自三体文明的问候:其他太阳系以外的粒子

而这些粒子,将引发电子器件的神经紊乱,专业名词是:粒子翻转。

它将很Surprise地告诉星载计算机和星载存储器:

“下面将是见证奇迹发生的一刻!”

“我要把1变成0,然后再把0变成1。”

有些人问了,多大点事啊,不就差个1吗?!

但是在比特界,差一位就可差之千里。

举个栗子:

如果指令20是向上爬升,指令24是停止推进,后果是难以想象的。

所以如果发生了1和0不分的情况,整个飞行器的运算结果曾导致非常大的灾难。

在1996年,阿里安501火箭,虽然没有粒子翻转,但是系统试图将一个64位的数字,放
到一个16位的地址里面去,随即发生了1/0错乱的现象。

结果在点火37秒后,火箭开始侧翻,随之爆炸,因为这个“小”问题,那次发射损失高
达3.7亿美金!

回到主题,既然粒子翻转这么恐怖,那SpaceX如何做到发现问题和解决问题的呢?

很简单:民主决策

技术名词叫:parity bits

同位位元

既然判断不了一个是否翻转,那就多放几个一样的设备,通过比较,把不一样的结果给
踢出去。

攒火箭硬件选择

上文提到,SpaceX没有选择用贵族宇航级器件,而是选择了经典厂牌Intel的X86双核处
理器,京东售价仅478元人民币(参考价为奔腾系列,赛扬更便宜):


而SpaceX也没有用双核做一件事,而是把双核拆成了两个单核,分别计算同样的数据。

每个系统配置3块芯片做冗余,也就是6个核做计算。


如果其中1个核的数据和其他5个核不同,那么主控系统会告诉这个核重新启动,再把其
他5个核的数据拷贝给重启的核,从而达到数据一直同步。

周而复始,不让一个核掉队。


据SpaceX前火箭总师John Muratore透露,龙飞船一共有18个系统,每个系统配置了3块
X86芯片,龙飞船一共有54块。

所以龙飞船主控芯片的总价约:2.6万人民币,3600美元!

而猎鹰九号一共有9个分立式发动机,每个发动机配置了3块X86芯片,加上主控系统配
置了3块,猎鹰九号一共有30块这样的芯片。

猎鹰九号主控芯片的总价约:1.4万人民币,2000美元!

我差点砸了手里的X1,是它阻拦了我攒火箭的大计!

更让SpaceX开心的事情,是Intel X86的程序员一抓一大把啊,而专业宇航级器件的程
序用的基本都是特定语言,程序员比元器件还难找。

而且硬件工程师压力也小,X86芯片随便造,烧坏了?再来1个。不不,再买一打!

可是宇航级器件仅仅是测试费,就都够再买一车X86芯片的。

攒火箭软件选择

SpaceX就用的开源Linux写的操作系统,而Linux用随便一台电脑就可以编写。

同样的,SpaceX程序员最爱的还是C++,用开源的GCC或者GDB做火箭的主控程序。

SpaceX还用LabView,一款图形化编辑语言,对于火箭程序来讲,它更容易实现可视化
和流程化,更容易做复杂的算法设计和数据分析。

SpaceX也用Matlab,在仿真和矩阵计算上,真的很好用。

而且,龙飞船,猎鹰九号,猎鹰重型,分享着同一款代码,分享着同一类迭代,分享着
同一种喜悦,多么的模块化,多么的互联网…

大数据监控和测试

2018年,SpaceX一共发射21次,一个公司占全球发射数量约20%,而SpaceX的工程师和
分析师,手里有大量的测试数据和实际数据,而且他们也被鼓励用不同的维度,去检验
飞行器的安全性,形成最新的也最实用的测试程序,从而降低实测成本。

同时,Continuous Integration,持续集成也被应用在了程序测试上。

注:持续集成

为了配合敏捷开发(相对于瀑布开发)的速度和效率而产生的一个用于编译、测试、发
布、部署的工具。

通过这种办法,可以让团队每时每刻在持续的基础上,收到反馈并进行改进,不必等到
开发周期后期才寻找和修复缺陷。

而且火箭程序不同于其他,会进行“断弦式”测试,突然关闭一台电脑,来看看发动机
到底有什么反应。

总结

航天已经经历了60年的历史,每一次阶跃其实都伴随着各类器件技术革新,比如:

1950年代的晶体管技术;
1970年代的微控制器技术;
1980年代的数字信号处理技术;
1990年代的高性能存储技术。

现在,芯片工艺从28nm,16nm,10nm到7nm,工艺的提升也增加了芯片在太空中的抗辐
射性能,让商业器件在太空中应用可行性大大提升!

伴随航天成长的是经典的:摩尔定律。

但是摩尔定律到现在在地面侧都快失效了,而在航天侧还没有开始。

比如Greg Wyler在2019年1月6号,Twitter的Oneweb的新型相控阵天线,目标定价15美
金。

比如AWS与Lockheed Martin在2018年11月发布的超小型地面站,可降低地面站80%的成
本。

北京九天微星正在研制200mW卫星物联网终端模组,目标定价5美金。

因此,航天缺少的仅仅是大胆的商业器件验证,缺少的仅仅是采用MVP快速迭代的环境
,而逐年降低的发射成本正在迅速降低试验成本,因此:

属于航天的摩尔定律才刚刚开始!

属于航天的互联网思维才刚刚开始!

属于航天的大时代才刚刚开始!

zunhuhu : 2020-06-27#24
绝对赞同

就是太短。技术宅不够看呀
确实好文章。尤其还是中文的,不费眼神 ✌

Aidemengdun : 2020-06-27#25
确实好文章。尤其还是中文的,不费眼神 ✌
没错。后面一个网友转的英文只能一目十行地看了一些?

zunhuhu : 2020-06-27#26
没错。后面一个网友转的英文只能一目十行地看了一些?
我只看到了控制系统几个字。大部分的评论性文章,只要是英文的,就觉得读起来费劲。

zunhuhu : 2020-06-27#27
控制系统的一部分信息:
Most launch vehicles are not controlled manually, if that is what you are asking. Conventionally, companies that develop launch vehicles have all physical "steering" done in a combination of two methods:

Electronic sensor systems that provide awareness of the vehicle's position, speed and etc. This information is then used with guidance and navigation software, which is more akin to where the vehicle knows that it "needs" to go. Now it compares the two and if the vehicle is off trajectory it will use the below.
Engines on gimbals and an attitude control system. Engines on gimbals essentially means the vehicle can alter the angle of their engines in relation to the rocket for directional changes. An attitude control system (ACS) is essentially a system of "spouts" around the vehicle (in every direction) that expel high-pressure gasses usually to make it spin in certain ways or turn harder

SpaceX has stated that the vehicle must first "turn-around" in space after the first stage engine cuts off, therefore it is unlikely that "gimbaling" an engine is their methodology.

However, if they were to have their software know that the vehicle must essentially pull a 180, it can use its ACS to flip the rocket around before engaging the engine to begin travelling back towards where it launched (or to eliminate its horizontal velocity).

Now the rocket is hurtling back to Earth, face first. In order to survive through the atmosphere at high-speeds it must do so in a way that the rocket can take the force and temperature. So now we need the vehicle to enter "bottom first".

It now likely cuts its engines once again, repeats maneuver once again to "back into" the atmosphere. From here, steering is done in combination of gimbal-ing engines when they are reactivated, using the ACS as well as using these bad boys:

SpaceX uses optimal control scheme for its landing. The goal of this scheme is to reach zero altitude with zero vertical velocity given tight constraints of landing area and fuel. This makes it a non-linear optimal control problem.

The algorithm running behind the landing is known as G-FOLD (guidance for optimal large diverts) which uses loseless convexification for converging globally optimal solution. This algorithm was co-developed by Lars Blackmore (SpaceX principal landing engineer) in JPL and was originally designed for mars rover for powered descent.

Falcon 9 uses high end GPUs for its math processing.
没有觉得你贴的不好的意思哈,就是我自己比较怕看英文评论

周雅 : 2020-06-27#28
大王好贴要支持,老马精打细算作风也要支持。。。

MySunflower327 : 2020-06-27#29
感觉这跟传统的航天业思路是两个极端。马斯克的方案是低成本中迭代取优,成本低可以多造,失败了更换代价也低;传统航天业是软硬件都先严格择优(选型,软件哪敢用开源的?),不愿失败,害怕变更造成的代价,造成每个环节成本都高。

现在的硬件技术更好了,用低成本的硬件多做冗余设计可换取可靠性,Linux也出来快三十年了,工程师们对它的可掌控性更大了。马斯克确实是个天才,不得不佩服其系统工程能力!敢于实践这种新路子,将总的成本大大降低,传统航空业确实应该反思了。

卡城西北 : 2020-06-27#30
马斯克和那个理查德布兰森爵士,前者显然档次高很多。

0706 : 2020-06-27#31
好棒呀,难道从我们这一代终于要开始正式走向宇宙了么?川陀在群星深处等我呢

今年夏天 : 2020-06-27#32
耐性越来越好了,这么多字居然都看完了
虽然看不懂,感觉自己更有知识了?

0706 : 2020-06-27#33
耐性越来越好了,这么多字居然都看完了
虽然看不懂,感觉自己更有知识了?
楼主的意思就是说,之前的火箭相当一芯专政,必须要很靠谱的芯才行,靠谱的芯成本很高。现在的火箭相当于民主政府,需要比较靠谱的团队,比较靠谱的一个一个芯成本低得多,通过民主的方式,他们可以纠正错误的情况。(最终飞到川陀去)

今年夏天 : 2020-06-27#34
楼主的意思就是说,之前的火箭相当一芯专政,必须要很靠谱的芯才行,靠谱的芯成本很高。现在的火箭相当于民主政府,需要比较靠谱的团队,比较靠谱的一个一个芯成本低得多,通过民主的方式,他们可以纠正错误的情况。(最终飞到川陀去)
谢谢解说
大王最好再首贴放一个通俗易懂版☺️

卡城西北 : 2020-06-27#35
好棒呀,难道从我们这一代终于要开始正式走向宇宙了么?川陀在群星深处等我呢
第一代银河帝国一万年以后成立。

金角大王 : 2020-06-27#36
这个说到通过民主的方式,应该是指通过像Voting这样的方法决策,这在人工智能中用得很多
楼主的意思就是说,之前的火箭相当一芯专政,必须要很靠谱的芯才行,靠谱的芯成本很高。现在的火箭相当于民主政府,需要比较靠谱的团队,比较靠谱的一个一个芯成本低得多,通过民主的方式,他们可以纠正错误的情况。(最终飞到川陀去)

金角大王 : 2020-06-27#37
就是: 白菜化后,又好又便宜
谢谢解说
大王最好再首贴放一个通俗易懂版☺

浅茉 : 2020-06-27#38
文科女看了十行就开始蒙圈。。。。不明觉厉
每个时代都会出几个马斯克这样的大头蒜,加油啊帅哥。

俺看楼主的大作 不是懵圈,而是看"天书"的feeling.

1593287623035.png

浅茉 : 2020-06-27#39
在航天圈里,有一种器件如贵族般存在:宇航级器件。

一个二极管只要上天验证成功,就可以从一个工业级十八线小明星跃升为宇航级一线大
明星,身价可以倍增上百倍甚至上万倍!

以现有载人飞船搭载的星载计算机和控制器举例,单个控制器价格为500万人民币左右
,一共14个系统,为了追求高可靠性,每个系统1+1备份,一共28个控制器,成本总计
约1.4亿人民币!

而SpaceX的龙飞船主控系统的芯片组,仅用了2.6万人民币,成本相差5384倍!

Elon Musk到底是如何做到的?

我们看以下几条重要的知识点:

SpaceX 猎鹰九号和龙飞船用的都是Intel双核的x86处理器;
操作系统用的是Linux,还有LabView和Matlab;
软件工程用的是C++,有些时候也用Python;
整个主控程序只有几十万行代码。

工业级器件小屌丝的困境:粒子翻转

航天器所有的器件要经历很苛刻很苛刻的环境。

首先发射时要禁得住剧烈的抖动和很高的温度,才能走出地球。

而真正的炼狱在入轨后才刚刚开始,面对太阳面的时候,温度迅速提升,最高到120°C
;背离太阳面的时候,温度骤减,最低到-150°C。

就这样90分钟一圈又一圈,周而复始,每圈都是270°C的温差。

而对于电子器件来说,温度不是最难熬的,最难熬的是太空中的辐射:

这些辐射有来自地球的召唤:地球磁场
也有来自太阳的问候:高能粒子
还可能有来自三体文明的问候:其他太阳系以外的粒子

而这些粒子,将引发电子器件的神经紊乱,专业名词是:粒子翻转。

它将很Surprise地告诉星载计算机和星载存储器:

“下面将是见证奇迹发生的一刻!”

“我要把1变成0,然后再把0变成1。”

有些人问了,多大点事啊,不就差个1吗?!

但是在比特界,差一位就可差之千里。

举个栗子:

如果指令20是向上爬升,指令24是停止推进,后果是难以想象的。

所以如果发生了1和0不分的情况,整个飞行器的运算结果曾导致非常大的灾难。

在1996年,阿里安501火箭,虽然没有粒子翻转,但是系统试图将一个64位的数字,放
到一个16位的地址里面去,随即发生了1/0错乱的现象。

结果在点火37秒后,火箭开始侧翻,随之爆炸,因为这个“小”问题,那次发射损失高
达3.7亿美金!

回到主题,既然粒子翻转这么恐怖,那SpaceX如何做到发现问题和解决问题的呢?

很简单:民主决策

技术名词叫:parity bits

同位位元

既然判断不了一个是否翻转,那就多放几个一样的设备,通过比较,把不一样的结果给
踢出去。

攒火箭硬件选择

上文提到,SpaceX没有选择用贵族宇航级器件,而是选择了经典厂牌Intel的X86双核处
理器,京东售价仅478元人民币(参考价为奔腾系列,赛扬更便宜):


而SpaceX也没有用双核做一件事,而是把双核拆成了两个单核,分别计算同样的数据。

每个系统配置3块芯片做冗余,也就是6个核做计算。


如果其中1个核的数据和其他5个核不同,那么主控系统会告诉这个核重新启动,再把其
他5个核的数据拷贝给重启的核,从而达到数据一直同步。

周而复始,不让一个核掉队。


据SpaceX前火箭总师John Muratore透露,龙飞船一共有18个系统,每个系统配置了3块
X86芯片,龙飞船一共有54块。

所以龙飞船主控芯片的总价约:2.6万人民币,3600美元!

而猎鹰九号一共有9个分立式发动机,每个发动机配置了3块X86芯片,加上主控系统配
置了3块,猎鹰九号一共有30块这样的芯片。

猎鹰九号主控芯片的总价约:1.4万人民币,2000美元!

我差点砸了手里的X1,是它阻拦了我攒火箭的大计!

更让SpaceX开心的事情,是Intel X86的程序员一抓一大把啊,而专业宇航级器件的程
序用的基本都是特定语言,程序员比元器件还难找。

而且硬件工程师压力也小,X86芯片随便造,烧坏了?再来1个。不不,再买一打!

可是宇航级器件仅仅是测试费,就都够再买一车X86芯片的。

攒火箭软件选择

SpaceX就用的开源Linux写的操作系统,而Linux用随便一台电脑就可以编写。

同样的,SpaceX程序员最爱的还是C++,用开源的GCC或者GDB做火箭的主控程序。

SpaceX还用LabView,一款图形化编辑语言,对于火箭程序来讲,它更容易实现可视化
和流程化,更容易做复杂的算法设计和数据分析。

SpaceX也用Matlab,在仿真和矩阵计算上,真的很好用。

而且,龙飞船,猎鹰九号,猎鹰重型,分享着同一款代码,分享着同一类迭代,分享着
同一种喜悦,多么的模块化,多么的互联网…

大数据监控和测试

2018年,SpaceX一共发射21次,一个公司占全球发射数量约20%,而SpaceX的工程师和
分析师,手里有大量的测试数据和实际数据,而且他们也被鼓励用不同的维度,去检验
飞行器的安全性,形成最新的也最实用的测试程序,从而降低实测成本。

同时,Continuous Integration,持续集成也被应用在了程序测试上。

注:持续集成

为了配合敏捷开发(相对于瀑布开发)的速度和效率而产生的一个用于编译、测试、发
布、部署的工具。

通过这种办法,可以让团队每时每刻在持续的基础上,收到反馈并进行改进,不必等到
开发周期后期才寻找和修复缺陷。

而且火箭程序不同于其他,会进行“断弦式”测试,突然关闭一台电脑,来看看发动机
到底有什么反应。

总结

航天已经经历了60年的历史,每一次阶跃其实都伴随着各类器件技术革新,比如:

1950年代的晶体管技术;
1970年代的微控制器技术;
1980年代的数字信号处理技术;
1990年代的高性能存储技术。

现在,芯片工艺从28nm,16nm,10nm到7nm,工艺的提升也增加了芯片在太空中的抗辐
射性能,让商业器件在太空中应用可行性大大提升!

伴随航天成长的是经典的:摩尔定律。

但是摩尔定律到现在在地面侧都快失效了,而在航天侧还没有开始。

比如Greg Wyler在2019年1月6号,Twitter的Oneweb的新型相控阵天线,目标定价15美
金。

比如AWS与Lockheed Martin在2018年11月发布的超小型地面站,可降低地面站80%的成
本。

北京九天微星正在研制200mW卫星物联网终端模组,目标定价5美金。

因此,航天缺少的仅仅是大胆的商业器件验证,缺少的仅仅是采用MVP快速迭代的环境
,而逐年降低的发射成本正在迅速降低试验成本,因此:

属于航天的摩尔定律才刚刚开始!

属于航天的互联网思维才刚刚开始!

属于航天的大时代才刚刚开始!

1593287676550.png 1593287694702.png

KDA2018 : 2020-06-27#40
真希望我能看得懂,可打横打竖看了两遍,还是不懂,只好弯腰给你们顶贴。

浅茉 : 2020-06-27#41
就是: 白菜化后,又好又便宜

大王,刚注意到你有70个粉丝,如果你想变成像小猫那样的69的内什么,我可以帮你的。

1593288409795.png 1593288792279.png

金角大王 : 2020-06-27#42
谢谢浅茉MM, 没事,70也不错,69在心里即可 :ROFLMAO:
大王,刚注意到你有70个粉丝,如果你想变成像小猫那样的69的内什么,我可以帮你的。