Aspen Technology Inc.Cambridge, Mass.

Problems usually come in packs, one small mishap reinforcing another until a major disaster occurs. Nowhere is this more true than in modern manufacturing systems, where dozens of processes may be underway at the same time. A sixth sense is needed to allow process managers to easily examine data from a multitude of sources and see not just what each process is doing, but how individual processes are affecting each other. Aspen Multivariate attempts to do just that, turning information from multiple sources into simple graphs or readouts. The system acts as a lens for information, enabling operators to focus on potential problems and resolve them before difficulties become disasters. Created by Cambridge, Mass.-based Aspen Technology Inc., Multivariate comes from a long line of statistical process control products designed to allow managers to observe what's happening in their facilities. Most of these products, however, take days or even months to spit out results, or look at just one variable at a time. Multivariate provides information on many variables in real time. "It makes complex processes easier for operations managers to understand," says Aspen vice president Brian Harkins. "A typical operator has hundreds of things he's got to keep an eye on." Multivariate uses a multi-step process that involves several different levels of data collection and information processing. The product isn't standalone. Facilities must first be users of the company's InfoPlus.21 information management system database. Adding Multivariate can cost $2,000 or more, depending on a facility's size and complexity. InfoPlus collects ongoing information about how systems operate. "Once we start to see alarm limits hit, we can go back through that component model," Harkins says. It's then possible to drill down into the data and unearth the specific elements that went wrong. These problem components are called "latent" variables by Aspen. A latent variable can have anywhere from two to a multitude of components -- one user is distilling 70 different pieces of information into nine or 10 understandable displays. What's crucial is to find out the latent variable's exact makeup. This is done using an offline module of Multivariate. Once the latent variable's specific parameters are determined, the information is exported to Multivariate's on-line module, which watches processes as they're actually happening. There still isn't any substitute for experience. Operators must know their equipment and its limits. With Multivariate, though, they can intervene when necessary. "It's almost like the traffic light application that everybody wants -- they want to see green, yellow, or red," Harkins says. And as everybody knows, traffic lights prevent wrecks.