Imagine unleashing a virus on your plant -- on purpose. It quickly penetrates every intelligent controller and sensor, spreading like wildfire to processing tanks, pipelines and even motors and pumps out in the field. What a nightmare. And you, the responsible party, are sure to be fired, right? Wrong. In about five years, this scenario may well become a dream come true for plant managers and CIOs.
Right now, engineers in R&D labs and scientists at major universities are designing technologies that experts are calling "intentional viruses." A more attractive term for the technology is "autonomous agents," a type of software that incorporates information and artificial intelligence rules. These software agents are intended to be spread around various devices on a network where they can assess what's really going on outside of an operator's view and even negotiate how devices should interact based on the conditions they detect.
Once they become low cost and ubiquitous, autonomous agents, in combination with secure industrial wireless mesh networks, new energy-reducing technologies and predictive diagnostic algorithms, will change the way manufacturers operate everything from the assembly line to recipe management.
By removing the need for high-cost labor from many manufacturing processes and enabling unattended data collection, such technologies will allow manufacturers to gain an efficiency edge. Ultimately, these technologies will create an information-rich, self-healing environment, enabling proactive behavior that will allow manufacturers to disrupt higher-cost, less-automated competitors.
Automation platform vendors, including Rockwell Automation (Milwaukee), Honeywell (Minneapolis), Invensys plc (London) and Emerson Process Management (St. Louis), have been investigating ways to use autonomous agents to add visibility into all corners of production -- be it on a rig in the Arctic or an assembly facility in Beijing.
LINKING AUTONOMOUS AGENTS
So far, most of the implementations are in the military. For instance, Honeywell is working with the Defense Advanced Research Projects Agency (DARPA) on a battlefield management application called Coordinators that uses autonomous agents. The idea is to equip soldiers and vehicles with sensors that communicate over a wireless network back to a battlefield commander. As soldiers enter an area, the software agents communicate conditions -- for example, that a truck has broken down -- which can be instantly relayed back to central command where a new plan can be launched.
Similar scenarios will play out in industrial settings. For the last seven years, Rockwell Automation, in conjunction with Cambridge University, has been designing the Manufacturing Agent Simulation Environment. It includes autonomous agents that talk to each other via a software program using technologies such as RFID, Rockwell's ControlLogix and even Web service communication technologies such as XML. Pilot projects are underway at a steel company in Australia using the agents to manage a water spraying process that cools the steel. Sometimes, due to dirt build-up, the sprayers are not fully open and, therefore, don't spray enough water. Rockwell paired software agents with each sprayer to gauge if it's fully open. Understanding the capacity of each sprayer, the agents can negotiate with each other to ensure the right amount of water is released.
According to Sujeet Chand, Rockwell's senior vice president and chief technology officer, autonomous agents are not new, but engineering the agents so that they respond to rules has been a real challenge. Even Rockwell, which has spent many years developing a simulation environment and working with the Navy on testing, is not going to release anything commercially for at least two to three years. When it does, however, the Manufacturing Agent Simulation Environment is sure to be disruptive while still allowing companies to use their existing manufacturing infrastructure. For instance, an automotive body manufacturer that runs sequential processes -- pulling the car through each welding station one work cell at a time, for example -- could use autonomous agent programs to automatically decide which cell the car needs to go to next to derive the most efficiency from the process. And that would save the manufacturer a lot of manual intervention, Chand says.
While the autonomous agent software can be layered onto existing devices, there is still an infrastructure issue to be addressed: How do you cull the information from equipment in far-flung places that may not be hooked into a network? To solve this, companies like Rockwell and Honeywell are investing in wireless mesh networks comprised of low-power sensors that communicate and self-heal, meaning a message may be rerouted in the event one sensor is not responding. There are many start-up companies delivering mesh networking solutions today. And mesh networks have a lot of value in many commercial areas. But when it comes to industrial applications, the technology needs a high level of security and reliability.
"The problem with general mesh networks is power management," says Dan Sheflin, chief technology officer for Honeywell's Automation and Controls Solutions Group. Honeywell is working with the Wireless Industrial Network Association (WINA) to design a mesh sensor network that can guarantee message delivery. "We're looking at a network that could have 50 or so sensors that communicate with powered nodes, and we can manage the energy of the sensor carefully," says Sheflin.
For its part, Rockwell is addressing the power issues associated with wireless sensor networks by working to create what are being called energy harvesting techniques. Rather than hook a battery (that will eventually die) to a sensor, the company's research arm, Rockwell Scientific, is developing ways to allow a sensor to tap into the vibration generated from a pump or motor it is monitoring, transforming that motion into its own energy source. It's still an evolving area, Chand says, but it has the potential to add the kind of reliability that industrial wireless applications require.
Emerson Process Management has a slightly different spin on self-healing technologies, a concept it is calling "continuous adaptive control." The controller, not yet released, applies to the proportional, integral and derivative (PID) control loop of process automation.
GAINING CONTROL
Typically, a large plant will have thousands of these loops, each with parameters that are determined at startup. But processes often change, due to weather, fluctuations in chemical compositions or degradation to equipment caused by factors such as power failure. Until now, there has been no way to adjust the PID parameters in real time, as these conditions change. To solve that problem, Emerson has created an adaptive control program that can gather information from the device to fine-tune the process and realign parameters based on external conditions.
The continuous adaptive controller technology was conceived in 1997, which is when Emerson developed the prototypes and engaged the University of California in Santa Barbara to develop proofs for the algorithms. In 2000, the technology was patented and has been in test beds since 2002. Some beta releases of continuous adaptive control will launch this summer, but general release is scheduled for a year and a half from now, corresponding with an upgrade to the company's DeltaV digital controller.
"What we've done is made a controller more intelligent so that it is able to acquire knowledge about a process and apply that for improved control," says John Caldwell, Emerson's product manager of advanced control products. "The ability to calculate a model for every loop in the control system, we believe, will be disruptive, because it provides understanding of the process that, before now, control systems haven't had." Adding intelligence to what has been a "dumb" device in the past will provide an added layer of productivity out in the field -- where typically companies have little control. Continuous adaptive control, for example, will not only calculate the best tuning for a specific mode of operation, it will remember what it has experienced in the past. "It not only learns, but it remembers," Caldwell says. Sounds hauntingly human.
THE HUMAN ELEMENT
But, while technology can mimic some characteristics of the human brain, it won't ever be able to replicate human behavior. That's why companies such as Invensys are taking their expertise into areas that directly affect how people -- not technology -- can change the rules of manufacturing on the plant floor.
"Technology has almost evolved to the point of being the biggest barrier our clients have in front of them to make their plants run well," says Peter Martin, Invensys' vice president and general manager of performance management. "Technology becomes the mediating factor in their thought process. With all of the rhetoric out there, my belief is that the single most disruptive technology that will impact the next two to five years is a true collaborative automation and information environment."
Martin says the barriers between operations need to be broken down, and Invensys has been researching ways to do that to the point that the company has patented a business model that describes, for example, what the maintenance department and the operational departments are supposed to do. It's not a product -- yet -- as the company is in the process of internal proof-of-concept work. But within the next few years manufacturers will have a model available to them that brings common sense back to operations that have been overrun by complex technology deployments, Martin says.
"We as an industry invest a tremendous amount of intellectual property, brainpower and money into things like intelligent agents and matrix analysis. It's fascinating science, but, for the most part, there's not two businesspeople out there running the plants that can tell you if that stuff made them any money," says Invensys' Martin. "The reason is because they are not accounting for it ... the accountants tell us if we are making or losing money, not the engineers." Martin is not discounting technology. He's just offering new ways to manage it meaningfully.
ARTIFICIAL INTELLIGENCE
In the meantime, manufacturers continue to look for ways to add efficiencies around their operations. And vendors realize they can't develop disruptive techniques in a vacuum. That's why many are supporting academic institutes like the University of Wisconsin at Milwaukee and the University of Michigan at Ann Arbor, which have created a cooperative research house called the Center for Intelligent Maintenance Systems (IMS). Working with over 45 different companies, the center focuses on infotronics (comprising the fields of industrial automation, integrated systems and information technology) for prognostic and smart-service systems. The goal: Achieve near-zero downtime.
"A machine should be able to assess itself," explains Jay Lee, director of IMS and professor at the College of Engineering and Applied Science at UW Milwaukee. "It should answer the question, 'how am I doing today?' to get to the point of zero breakdowns."
IMS has created a toolbox of algorithms it calls Watchdog Agents to help companies troubleshoot problems before they even happen. The tools enable quantitative assessment and performance-degradation predictions based on a fingerprint created from historical data. Currently, IMS has four test sites for Watchdog, but Lee predicts the technology, which can be embedded inside equipment, will be ubiquitous within three-to-five years.
While all of the vendors have varying strategies around next-generation technology for control platforms, the common thread comes down to embedding intelligence into equipment. The next-generation manufacturing plant has to be one step ahead of its operator, even if that means making decisions on its own.
"We have to make machines that learn and self-predict. I believe deeply that is the future of manufacturing," Lee says.