The Power of Prediction

In a storage facility in southeastern Idaho, a cluster of potatoes infected with late blight disease is beginning to decompose. As the hours go by, this small core of diseased potatoes will affect the healthy tubers surrounding them. The gasses emitted as the decomposition process takes hold will eventually reach a level detectable by workers at the facility but not until a portion of the stored potato crop is lost.

Thousands of miles away, a maintenance supervisor for a 200-engine fleet of commuter trains serving a large metropolitan area is about to face her worst nightmare: the traction system on a locomotive traveling at more than 50 miles per hour and carrying hundreds of commuters is about to fail just 10 hours before that system is scheduled for inspection!

Both of these scenarios, as different as they may seem, are alike in a very significant way: they illustrate the inherent shortcomings of conventional monitoring and preventive maintenance programs and the benefits of predicting failure in a process before that failure manifests itself.

Imagine instead two different scenarios -- ones in which problems are predicted long before they become evident. In these scenarios, maintenance and prevention resources are allocated based on information gathered and interpreted by a system of sensors and software connected through a wireless communications network that predicts where, and when, a problem is likely to occur.

In this scenario, a potato storage worker miles away from the storage facilities he oversees logs on to an Internet site. A screen displays icons for each location, followed by colored indicators that show the overall status of the facilities based on the presence of gasses associated with disease and decay. By clicking on any of these individual locations, the worker zooms in on that particular site and checks various "zones" within the storage area. Graphical displays of data show changes in levels of ammonia, alcohol and hydrogen sulfide. Trend data for one zone in Cellar Number 4 suggest the presence of late blight, and the worker sends an urgent e-mail to a maintenance crew giving them the location information and instructions for removal of the infected potatoes from the pile. The waste of a larger number of potatoes is averted.

In the example of the commuter train system, maintenance inspection likewise takes place not in the "field," but at a Web site. The maintenance supervisor logs on to the site and gets a graphical display of the conditions of each locomotive as it is being transmitted from advanced sensors that monitor conditions such as temperature and vibration in a variety of critical systems. Specialized algorithms associated with the sensors compare the real time data with historical data and operational specs from the engine manufacturer and look for variance patterns that would indicate potential failure.

Welcome to the world of predictive resource maintenance a world in which periodic physical inspection and scheduled maintenance is replaced by advanced monitoring technology and the "power of prediction." In this world, periodic inspection is replaced by continuous sampling of functional parameters often affecting aspects of equipment or processes that would otherwise have to be shut down in order to be inspected on a less frequent basis. And rather than assigning maintenance resources to preventive tasks, a single resource manager can receive reports on a daily or hourly basis that convert rate of change data into a statistical analysis of when maintenance resources may actually be required.

Welcome to the World of Insightek!

In the above scenarios, maintenance technicians or supervisors log on to a Web site maintained by Insightek. The user interface displays a graphic representation of the systems or processes being monitored. For clarity, the information is color coded according to their prediction status:

In the example of the commuter train systems, the maintenance technician can zoom in on a specific locomotive and view representations of each monitored system (e.g., drive train, power plant, brakes, etc.). As the maintenance technician passes the mouse pointer over a specific system, maintenance data flashes on the screen showing date of last service and the number of hours of operation since that service. With a further mouse click, the technician drills down to view more detailed information relating to the cause of the yellow or red maintenance alert. Perhaps measured vibration in a lubricant pump has exceeded normal parameters, and based on historical patterns and manufacturers specifications presents an 80 percent probability of failure within the next 72 hours, and a 100 percent probability within the next 100 hours!

Armed with this information, the maintenance technician can check inventory and order the appropriate parts, schedule the personnel staff to take the locomotive off-line the next morning and replace the pump, or perform some other required service related to the cause of the vibration. The end result is a potential interruption of commuter service averted, a more efficient use of personnel resources...and a more efficient (and therefore less costly) operation.

Predictive resource management is equally beneficial in monitoring processes, as well as physical assets. For example, an Insightek system installed in a potato storage facility looks for trends in the release of degradation chemical by-products of spoilage such as ammonia and hydrogen sulfide. As in the railroad example, the ability to deliver easy-to-interpret predictive analysis over the Internet means that the farmer of agribusiness monitoring the potato storage facilities can do so from anywhere, and can take preventive actions before any loss of assets occurs. There are virtually no applications involving equipment or processes in which some form of predictive resource management is not applicable.

What Is Predictive Resource Management?

The two most common approaches to resource management are post-fixial and preventive. Of the two, post-fixial represents the worst possible approach, in that a remedy is applied after a failure has occurred. At the very least, a post-fixial response implies a loss of assets, measured either in: time (lost production), money (the repair or replacement of an asset), or productivity (a decrease in process or production yield).

Ultimately, these losses represent increased costs and reduced profit. These costs may be inconsequential in the face of environmental and/or human disasters that may occur as the result of a post-fixial situation. Imagine what the benefits would have been had plant operators been able to predict the failure of a simple valve at Chernobyl.

Today, scheduled maintenance based on expected time-to-failure is considered the highest form of resource management. The problem with this form of preventive maintenance is that equipment and processes do not always behave according to expectations. When they do not, one is in the post-fixial response mode. Although better than the wait for failure mode, preventive maintenance represents a less than optimal use of resources since it necessitates equipment downtime and allocation of maintenance resources when nothing is actually wrong.

The technological underpinning of both post-fixial and preventive resource management approaches are SCADA-based monitoring systems. While SCADA technology is very sophisticated at communicating status, predictive resource management represents an entirely new paradigm based on measured time-to-failure vs. expected time-to-failure.

The technology and service model introduced by Insightek embody this new paradigm in resource management: predictive resource management. In contrast to preventive maintenance, which is a scheduled remedy based on expected time-to-failure, predictive maintenance is a scheduled remedy based on measured time-to-failure.

Benefits of Predictive Resource Management

Aside from reducing the inestimable risks of environmental damage and loss of human life, depending on the applications involved, predictive resource management lowers costs based on a number of different factors, including:

   Reduced unscheduled downtime

   Reduced spare parts inventory

   Reduced repair time/increased labor efficiency

   Fewer unnecessary repairs

   Less product loss

Today's most sophisticated SCADA (System Control And Data Acquisition) based monitoring systems are still predicated on data associated with historical experience, rather than on predictive analysis. Their value lies chiefly in their ability to monitor and report a condition, rather than predict when a condition may occur.

The key to predictive resource management is the ability to measure time-to-failure. And to accomplish measured time-to-failure requires a unique blending of state-of-the-art sensor technology, algorithms capable of supporting distributed analysis, and a communications and reporting infrastructure that is easily installed, cost-efficient, and user-friendly.

Measuring Time-To-Failure: The Key to Predictive Resource Management

Sensors
The computer industry has long acknowledged the power of Moore's Law in the exponential expansion of processing power at ever-decreasing costs. What is less recognized is that a similar revolution is taking place in sensor technology. Today, sensors are not only transforming real-world phenomenon into digital information, but are incorporating data processing capabilities that, in essence, are making sensors "aware" of their surroundings.

The impact of "smart sensors" in the realm of predictive resource management is that these devices are now able to evaluate the phenomena they measure, rather than simply serving as indiscriminate observers of some physical state. This capability is critical to the concept of measured time-to-failure, which implies an analysis of rate-of-change patterns. As with processing technology, the further good news associated with today's sensors is that these capabilities are being delivered is increasingly less expensive packages. The ability to replace a thousand dollar system with a $20 component is making the concept of "self-aware appliances", on the Internet, a commercially viable reality, and is technological force behind Insightek's use of distributed analysis.

Distributed Analysis
The simple, but powerful, premise behind the concept of distributed analysis is the removal of unneeded information from an otherwise constant flow of sensor data. The combination of sensor technology, wireless communications and software algorithms embedded in Insightek installations yield an information service that tells an end-user only what they need to know about a monitored process.

In an Insightek installation, equipment-monitoring sensors interact with remote processing nodes to interpret measured data against an established set of process parameters. Data processed at these nodes is passed along via a wireless network only when rate-of-change patterns are detected that predict an impending problem at the sensor location. Rather than forcing end users to pore over strip chart recordings to come to their own predictive conclusions, Insightek installations convert data into useful information before it leaves the sensor. The result is "event reporting" vs. data communications.

Reporting
The distributed process nodes in an Insightek installation pass their information to a gateway PC that in turn provides a wireless link to an Insightek Web server/data warehouse. It is through the Internet that Insightek customers can access their Web-based reports. It is these reports, and not capital equipment, that are the heart of the Insightek business model an Internet-based information service.

Communications
Distributed analysis means that each remote node in an Insightek installation handles its own processing, which essentially removes the network from the analysis. The practical benefit of this approach is to reduce the need for high-bandwidth networks, which enables Insightek to utilize low-cost wireless networks.

Insightek: The Service Model Approach to Predictive Resource Management

Insightek takes advantage of a number of converging technologies: smart sensors, wireless communications networks, high-speed/low-cost data processing, and the ubiquitous power of the Internet. While conceivably any company could put these components together to create their own predictive resource management system, Insightek is unique in two very important ways:

This last point is critical to the success of Insightek customers, in that Insightek's compensation is based on the delivery of useful information, rather than on the delivery of assets that require the customer to derive some return-on-investment. As a service provider, Insightek removes the ROI risk that customers would incur in the attempt to create their own predictive resource management system. In addition, Insightek's service agreement places the responsibility for installation and maintenance on Insightek, rather than on the customer's information systems resources.