This article discusses that data acquisition techniques give researchers insights into fields outside the realm of machines. The methods that mechanical engineers most frequently call upon in their work have never been strictly confined to the province of machinery. Patterns may indicate the need for preventive maintenance or they may warn that a key piece of equipment is on the verge of failure. Time-series analysis has found uses in the important and potentially lucrative-intersection of engineering and medicine. Engineers commonly look at the relationship between time and response. But in biology that time relationship is not well characterized because it is hard to isolate and look at living systems over time. Currently, cell response is often studied on the macroscopic level.
The methods that mechanical engineers most frequently call upon in their work have never been strictly confined to the province of machinery. Today, perhaps more than ever before, engineers are applying their methods and technology to make advances in other fields.
Time-series analysis, for example, measures a system at discrete time intervals, then plugs all those readings-which can number in the tens of thousands-into a database. Specialized software that sits atop the database searches for patterns to reveal significant details about how a system operates. Engineers typically rely on time-series analyses to study operating parameters of a machine like a turbine.
In any plants, sensors on turbines take millisecond measurements. Some very finely tuned machines need to have their workings followed minutely-almost in real time-so that operators are immediately alerted to abnormal operating patterns.
Patterns may indicate the need for preventive maintenance or they may warn that a key piece of equipment" is on the verge of failure.
Time-series analysis has found uses in the important and potentially lucrative-intersection of engineering and medicine. Two mechanical engineers at Carnegie Mellon University are currently studying how cells move and change through time.
And at the University of Toronto, Willy Wong, an associate professor of electrical and computer engineering, is investigating the nature of pattern recognition itself. Better understanding of the process could lead to ever smarter machines.
Wong calls upon his background, combining engineering music, and psychology, to study how the human brain-the original time-series analysis software-intuitively organizes patterns from vast amounts of data.
We find patterns without thinking about it. For us, a series of separate musical notes comes together to form a melody, which we can remember. Conversation isn't an incoherent mess of babbling back and forth. We not only recognize words and follow the conversation, but we also pick up on nuances within those conversations.
"Maybe aliens don't hear or see a pattern there, but we do," Wong said. "From an evolutionary point of view, pattern recognition is one of the most important skills we learn from birth-to recognize friend from foe."
Many research challenges have pattern recognition at their heart, Wong said. Software systems such as those being developed by Ben Shneiderman's team at the University of Maryland are used by researchers across many fields. Shneiderman, a computer science professor at the school in College Park, Md., focuses on systems that visualize patterns.
His team's PatternFinder can search for and see patterns of events over time in patient histories, Web logs, news stories, and criminal activities. The group's TimeSearcher software, which visualizes trends in time-series data, is used by meteorologists, business analysts, and mechanical engineers, among others, to look for trends in weather patterns, the stock market, and turbine operation.
"Typically, the way computers analyze time-series data is through a statistical method, by looking for a spike or decline in numbers," Shneiderman said. "But when they can see these on the screen, it's so compelling."
Humans are great at recognizing such patterns. Beyond analysis software, machines aren't so great at pattern recognition. "Getting a machine to recognize differences ·is a challenging problem right now," Wong said.
But if they could be programmed to behave in this manner like humans, think of the applications, he said.
"The dishwasher hasn't changed in 40 years. So what if I want to build one that can not only wash dishes, but put them away?" he said. "You'd need to design a program that knew one dish was different than another; one is a saucer and it goes in a different place than the plate." Robots don't intuitively recognize speech. Wong's research could lead to programs of machine intelligence with pattern recognition to enhance capabilities for various kinds of human-machine communication.
To understand how humans recognize familiar patterns, Wong's team has formulated a human model that's based on the way software systems run pattern matching and time-series analysis. In his work, Wong looks at how engineers might intuitively model biological systems without even knowing they're doing so-when creating engineered systems.
For instance, Wong was interested in the role vision plays in our ability to analyze motion. He theorized that vision might work like the Hough transform, a technique used in digital image processing. A transform unit allows processes to identify lines and the position of shapes in the image.
"We said, 'If our work has resemblance to biology, is the Hough transform implemented somewhere in the eye?' " Wong said.
To his surprise, he found that scientists David Hubel and Torsten Wiesel had won the 1981 Nobel Prize in Physiology or Medicine for their work describing how signals from the eye are processed by the brain to generate edge detectors, motion detectors, stereoscopic depth detectors, and color detectors, which are all building blocks of vision.
"What surprised me was how similar their model was to the Hough transform," Wong said. "We had no idea of this when working with Hough transform."
Engineers today also look consciously to biological systems for ideas, but they have much to learn from the methods biologists use to investigate their world.
Take search engines, which require computer engineers to try to mimic human thought when designing them.
"Imagine you're trying to look for a movie but you don't know the name of it," Wong said. "You don't want to go through the database, looking at one film after another. But to get a search engine to work for you, I have to understand how you're clustering and grouping things together. That's the act of pattern recognition. If the engineer learns how you do it, he can teach a machine how you want it done."
For his part, Wong is presently marrying biology with engineering in the study of human gait recognition. He's looking at how we use our innate human ability to recognize patterns to identify a friend approaching at 20 paces. A system based on what he's learned could have a role in Hollywood and in security systems. In Hollywood, Wong's findings could help create software that allows animated characters to move naturally.
"In Lord of the Rings when you see these animation sequences for characters that don't exist, like Gollum, they use a real human actor to find out where the shoulders and joints move, and project some sort of image on top of that," Wong said. "It's time-consuming and expensive, so to automate that task in movies and speed it along is welcome."
An application based on Wong's findings could be used within a security system to identify people from a distance based on how they move.
To Get A Search Engine To Work For You, I Have To Understand How You're Clustering And Grouping Things Together. That's The Act Of Pattern Recognition.
A team at Carnegie Mellon University in Pittsburgh has applied time-series analysis in a device that may unlock some of the secrets of disease on the cellular level.
One of the team members is William Messner, a professor of mechanical engineering at Carnegie Mellon. According to Messner, working with cells isn't easy. Biomechanical researchers face less of a challenge calling upon engineering methods when they're working with large systems-like a pacemaker-that they can actually see and touch.
"It's a lot easier to apply forces and measure the results of something when it's on the macroscopic scale," Messner said.
But looking at systems on the cellular level is much more difficult.
"Cells are a whole different beast because they're very small and they're really complicated. If you think about a cell as a machine, you have on the order of 100 million proteins or parts," Messner said. "They're far more complicated than even the largest engineered system, than the Space Shuttle."
He and Philip LeDuc, an associate professor of mechanical engineering at Carnegie Mellon, have built a small system that Messner describes as similar to lab-on-a-chip technology. Their patent-pending device analyzes how cells behave in response to certain stimulations and forces with the help of time-series analysis.
"Cells are always in this state of feedback and communication with each other and with something flowing over them like hormones. It's extremely complicated to get a grip on," Messner said.
The device includes two narrow microfluidic channels that merge as they move toward one outlet. The fluids contain chemicals that stimulate a cell located within the outlet channel. By varying inlet pressure at the channels' head, the researchers vary the pressure on the cell. They observe cell response under a microscope.
"The microscope lets us watch the response and takes pictures of how the cell responds as a function of time," Messner said. "From an engineering point of view, it's hard to measure inputs and outputs of a biological syssystem. But if you use time-series and you get this value at this time and another at another time, you can compare them."
By varying fluid flow, researchers can also get fluid to stimulate the cell at different locations or flow entirely over the cell. They then measure cell response. The device has many applications, including study of the early days of human or animal life.
"The development of an organism starts from a single cell, which becomes a number of cells that organize themselves," Messner said. "How does it do this? By the signaling between cells and chemical gradients, and we hope to look into it using this system."
The device could also help researchers understand why a tumor can go dormant for a time and then suddenly reactivate to send cancer cells into the bloodstream.
"Why is it you have this tumor made up of millions of cells and then for some reason some of them decide, 'I'm not happy in this tumor; I need to get out and disperse,' which is bad," Messner said. "There are chemical cues that these cells get from neighboring cells that say, 'Maybe you should get out of here and go elsewhere.'
His device might help isolate and find those chemical cues and determine why cells respond to them the way they do.
Make The Machine
Engineers commonly look at the relationship between time and response. But in biology that time relationship isn't well characterized because it's hard to isolate and look at living systems over time. Currently, cell response is often studied on the macroscopic level.
"You can say at this point, we applied the chemical-or gave a drug-and then you say the patient got sleepy, so I guess the cells started to morph," Messner said. "But it's kind of vague the time it happened in."
Although cell reaction can be studied under a microscope, tracking response over time is another story. In those instances, time-series information is often the result of a grad student watching the cells, then writing down information at evenly spaced intervals. But even the best graduate students can't look through a microscope for more than about an hour. And they certainly can't take measurements by the second, Messner said.
Much research today is devoted to trying to isolate what happens inside a cell, and by marrying engineering and biological techniques, such research can move forward quickly, Messner said.
"You see more and more engineers having an impact on biology. A lot of stuff isn't happening now because there's no instrument to measure it. The human genome project was going to take 50 years until engineers automated a whole bunch of it," he said.
"Engineers can contribute here because we're the people that make the machine," he said