IN THE PIPELINE: Using Software To Predict The Future
The Web site offers to "Change the Future - To Your Advantage" and greets visitors with a crystal ball flashing images of hospital patients, astronauts and nuclear reactors.
This is not some Internet hoax. Nor is it the domain of a dubious fortune teller. It's the homepage of a Chicago area startup called SmartSignal Corp.
The privately held company says it has developed software that can predict when machines - everything from airplane engines to power plant turbines - will break down, so they can be fixed before they fail.
The fault-detection software was created a decade ago by researchers at Argonne National Laboratory in Argonne, Ill., to monitor nuclear facilities. SmartSignal has spent years refining and testing the software with the goal of expanding beyond heavy machinery into automobiles, home appliances and medical equipment.
"There will come a day when your refrigerator or car...has a sense of self awareness and tells you a problem is developing so you can take care of it," says Gary Conkright, SmartSignal's chairman and chief executive. "It may sound like science fiction but it's not that far off."
There's plenty of science, but not much fiction behind the Lisle, Ill., company, whose 40-person staff includes the team of Argonne scientists that created the mathematical equations and analytical process at the core of the software.
SmartSignal's software culls data collected from existing sensors to find signs of impending equipment failure. It can identify problems hours or days before they would trigger typical sensor alarms. It also notifies technicians of the expected failure date and time.
It works by comparing real-time sensor readings to a model of how the machine should be operating. The models are created using performance data from each machine and updated regularly. The software looks for subtle changes in the "noise" between the model and actual readings, and then distinguishes between normal variations and incipient faults.
"We look at (a piece of equipment) in a holistic way rather than sensor by sensor," Conkright says. "Therefore, we see things much earlier than other approaches with a much greater degree of confidence."
The startup has faced considerable skepticism from the engineers and technicians at large companies, which often have homegrown fault-detection systems. But it has signed up 15 customers by proving that its software works in pilot projects.
After a 12-months of testing, Delta Air Lines Inc. (DAL) decided earlier this year to use the software to monitor its fleet of jet engines. "So far, 100% of the time that SmartSignal has altered us to a potential problem that problem has come true," says Walter Taylor, director of process and technology engineering at Delta in Atlanta.
Another customer is a division of General Motors Corp. (GM) that builds diesel locomotives. GM Electro-Motive has been testing SmartSignal for two years and uses it on 100 freight and passenger trains. It has found some problems more than a week before they occurred. "We're convinced we have saved locomotives from road failure several times," says Curt Swenson, director of marketing at the GM unit in LaGrange, Ill.
From Power Plants To Pacemakers
Although SmartSignal has focused on industrial uses of its software, the company is starting to explore other markets, such as healthcare. Executives say the software could be adapted to predict problems with medical devices, like pacemakers and home dialysis kits, that are retrofitted to wirelessly transmit data.
"It's a technique that's completely generic...It could work on anything from jet engines to a toaster oven to your watch," says Alan Thomas, director of technology commercialization at the University of Chicago, which runs Argonne and bankrolled SmartSignal's early years.
Stephan Wegerich, a senior scientist at SmartSignal and a former Argonne researcher, says the software could even predict when hospital patients in intensive care wards will crash by tracking data such as pulse rates, blood pressure and oxygen levels.
"When you're looking at someone in the ICU, they are already bad. You are looking to see if they are getting worse," Wegerich says. "You collect your initial data, just like a jet engine monitor, and look for deviations from the initial dynamics."
SmartSignal is also conducting trials with Questra Corp., a Redwood City, Calif., firm that makes software to collect data over the Internet from machines in diagnostic laboratories and hospitals.
The software not only predicts when lab equipment, like a robotic arm, will fail, but it can also measure the accuracy of test results, says Vesna Swartz, Questra's vice president of marketing. It can analyze a batch of test scores in real time to determine if there was "a problem with the way the test was run," Swartz says.
Fault-detection software is making inroads with manufacturers of heavy equipment and operators of large fleets of machines, but it will be some time before the technology finds its way into medical devices and consumer products, says Marc McCluskey, an analyst with AMR Research in Boston.
SmartSignal's software is best at analyzing rotating machines, like jet engines and gas turbines, McCluskey says. "Once you get into medical devices and toasters, there are different nuances...as to how they operate, where they fail, why they fail," he adds.
A Needle In The Haystack
The technology has already come a long way. It was born in the early 1990s when the U.S. Department of Energy asked researchers to develop a better tool for predicting problems at nuclear facilities after the failure of a coolant pump forced an emergency reactor shutdown at Idaho Falls, Idaho.
The Energy Department held a contest. It buried 10 faults in 18-months worth of data, recorded at one minute intervals, from a Volkswagon-sized coolant pump and asked researchers to find the problems.
"It was like finding a needle in the haystack," says Wegerich. The Argonne team crunched gigabytes of data and not only found the 10 hidden faults, but their software uncovered problems that were previously unknown.
The researchers later realized they had stumbled upon a technique that could be used on other pieces of equipment. The University of Chicago, which secured patents on the early Argonne work, and Alan Wilks , a former AlliedSignal Corp. researcher, created SmartSignal to improve and commercialize the software.
Investors were unwilling to fund the startup and it went through several management teams until Conkright, an engineer turned businessman, joined in 1998. It has since raised more than $23 million from venture capitalists, expanded its patent portfolio and begun to win over the skeptics.
用软件预测未来
这个网站表示要"改变未来，造福于你"，迎接访客的是一个水晶球，医院患者、宇航员和核反应堆的图像在上面交替闪现。
这不是互联网上的恶作剧，也不是含糊其辞的算命先生的领地。这是芝加哥一家新兴公司的主页。这家公司叫做SmartSignal Corp。
这家由私人控股的公司称，该公司开发了一种软件，从飞机引擎到发电厂的涡轮，无论什么机器，这种软件都可以预测它们出故障的时间，从而能够在出故障前及时修理。
这种查错软件是10年前由总部位于伊利诺伊州阿尔贡的阿尔贡国家实验室(Argonne National Laboratory)的研究人员开发的，其目的是为了监控核设施。SmartSignal多年来不断对这种软件进行改进和测试，以期将软件适用范围从重型机械扩展至汽车、家用电器和医疗设备。
SmartSignal的董事长兼首席执行长加里?康克莱特(Gary Conkright)说："总有一天，你的冰箱或是汽车会有自我意识，能够告诉你就要出问题了，让你能及时解决问题。这听上去也许像科幻小说，但已经为时不远了。"
在这家公司有很多科学，而不是幻想。公司40名职员中包括阿尔贡的科学家，后者建立了该软件核心部分的数学等式和分析程序。
SmartSignal的软件对从现有传感器中收集的数据精挑细选，寻找设备即将出现故障的信号。这个软件可以在一般传感器发出警告前数小时或是数天找出问题。它还可以通知技术人员故障可能出现的日期和时间。
软件的工作原理是把传感器的实时数据与机器正常运作的模型相比较。模型以每台机器的表现数据为基础，而且会定期更新。软件寻找模型和实际数据之间的细微变化，以确定究竟是正常的变化，还是故障的先兆。
康克莱特说："我们是对整个设备进行全盘考虑，而不是一个传感器、一个传感器地比较。因此，与其他方式相比，我们发现问题更早，准确性也更高。"
这家新兴公司遭到了大公司的工程技术人员的广泛质疑，这些公司通常有自己的检错系统。但是，这种软件通过了实验性项目的考验，已签得了15个客户。
经过12个月的测试，达美航空公司(Delta Air Lines Inc., DAL)今年早些时候决定使用这种软件来监测其喷气式飞机引擎的表现。达美航空程序和技术工程部门的负责人沃尔特?泰勒(Walter Taylor)称："迄今为止，每次SmartSignal向我们指出可能会出问题的时候，问题最后都确实发生了。"
另一个客户是通用汽车公司(General Motors Corp., GM)生产柴油机车的子公司。通用汽车的Electro-Motive两年来一直在测试SmartSignal，并在100辆货车和客车上使用了这种软件。在一些问题发生的一周以前，这种软件就已经发现了问题。这家通用汽车子公司负责营销的库尔特?斯文松(Curt Swenson)说："我们相信，有几次我们避免了机车在路上出故障。"
从发电厂到起博器
尽管SmartSignal的软件主要用于工业用途，但该公司正在开发其他市场，如医疗保健。公司管理人士称，这种软件可用于预测医疗设备的问题，例如起博器和家用透析设备等经过改进可以以无线方式传输数据的设备。
芝加哥大学(University of Chicago)负责技术商业化的艾伦?托马斯(Alan Thomas)说："这完全是一种通用性技术，在什么东西上都能用，从喷气式飞机的引擎，到烤面包机，到手表，都行。"芝加哥大学是阿尔贡国家实验室的主办人，在SmartSignal成立的最初几年也为该公司提供资金。
SmartSignal的资深学者、前阿尔贡国家实验室的研究人员斯特凡?韦格利奇(Stephan Wegerich)称，通过追踪脉搏、血压和氧含量等数据，这种软件甚至可以预测医院重症监护室患者的病情什么时候会突然变化。
韦格利奇说："患者在进重症监护室的时候，情况已经很糟糕了，你需要注意的是他们的病情是否会恶化。你收集初始数据，就像喷气式飞机引擎的监控器一样，然后寻找与初始状态背离之处。"
SmartSignal还在与Questra Corp.共同进行试验。后者总部位于加州雷德伍德城，开发通过互联网从诊断实验室和医院的机器设备上采集数据的软件。
Questra负责营销部门的副总裁韦斯娜?斯沃茨(Vesna Swartz)称，这种软件不仅可以预测实验室的设备（如机器手臂）何时会出故障，也可以测量试验结果的准确性。它可以实时分析大量测试分数，以确定测试方式是否存在问题。
AMR Research驻波士顿的分析师马克?麦克拉斯基(Marc McCluskey)说，在重型设备制造商和拥有大量机器的营运商之间，查错软件已经得到了很大程度上的认可，但是这种技术要进入医疗设备和消费品领域尚须时日。 麦克拉斯基称，SmartSignal的软件在分析喷气式飞机引擎和燃气轮机等旋转的机器方面表现最好，而医疗设备和烤面包机等，则存在很多细微的差别，如工作原理、故障发生的地方和发生的原因等。
大海捞针
这种技术已经取得了很大进展。它最初是在20世纪90年代初开发出来的。当时，由于冷却泵故障迫使爱达荷州爱达荷福尔斯的核反应堆紧急关闭，美国能源部(Department of Energy)要求研究人员开发一种能够更好地预测核设施故障的工具。
能源部举办了一次竞赛：在一个冷却泵18个月的数据资料里，隐藏了10个错误，研究人员要找出这些错误。数据以分钟为间隔录制。
韦格利奇说："这就像大海捞针。"阿尔贡研究小组对数十亿字节的数据进行了处理，不仅发现了隐藏的10个错误，还找出了以前不知道的其他问题。
这些研究人员后来意识到，他们无意之中发现了一种可以用于其他设备的技术。获得了阿尔贡前期工作的专利的芝加哥大学和前AlliedSignal Corp.的研究人员艾伦?韦尔克斯(Alan Wilks)，联手创建了SmartSignal，以改进这种软件，并将其投入商业化使用。
投资者不愿为新兴公司提供资金，在工程师出身的实业家康克莱特于1998年加盟之前，公司已历经数任领导团队。但自康克莱特加盟以来，公司已经从风险投资者手中筹集了2,300多万美元资金，增加了产品专利，并且在与怀疑论者的斗争中开始节节获胜。
