The productivity of most production and operating equipment could be greatly improved if parts could be replaced only when they are about to fail. Scheduled maintenance and replacement of parts on a regular basis can keep equipment running but generally results in replacing parts with considerable operational life left in them. Conversely, waiting until parts fail and stopping an assembly line or disrupting the normal function of critical equipment proves costly to production or other operations.

For many years industry has tried to detect imminent equipment failures by close monitoring of instrumentation. This has only marginally been successful because sensors were rarely placed for optimum measurement and failure characteristics of components were not known or measurable. Additionally, inexpensive computers were not powerful enough to process the data in a timely way.

New techniques are available to more accurately predict when equipment will degrade or fail and this can usually be done without adding additional expensive instrumentation. Model-based diagnostic engines connected to the system's instrumentation can make these predictions very quickly. They can not only predict the failures, but they can inform the operator what will happen to the rest of the system when the failure occurs and propagates. The best diagnostic engine can even detect and isolate failures that are not common or have not been observed before--as opposed to fault trees which must use a catalog of failure symptoms and histories. Likewise, rule-based systems rely on predefined symptoms and logic to detect and handle anomalies.

The RODON diagnostic engine is especially capable in this area. The detailed models of a functioning system it contains are the basis for determining the nominal, degraded and failure operation of a system. It continuously performs a "pre-mortem analysis" of all parts based on instrumentation data it receives. The process is made more accurate when the component models contain information about degraded or aging behavior. The RODON Diagnostic Engine functions by reading the sensor data and determining from its models what could be causing the observed performance under those circumstances, calculates a trend based on the modeled characteristics of all the system components and develops a prediction of time-to-component(s) failure. From this, it can be determined when the system can best undergo maintenance or when redundant equipment should be activated. Limitations of this methodology are few but random failures of reliable components, fast, hard failures of components which have no degraded operating characteristics and very poor instrumentation become the largest constituents of the uncertainty tolerance of such a method. But even as these failures occur and experience with them accumulates, learning will soon result in instrumentation or system improvements so these random failures will be driven back to predictable events.

There will be a wave of new system automation as new, powerful, less-expensive computers and workstations and some very sophisticated software are applied to cost reduction in industry. Prediction of failures and replacement of parts only when necessary at times of least interference can save a great of money. This new technology can be used on new and old equipment of any scale from home heating equipment, to automobiles, to spacecraft, to ships at sea. Integrated tools like Rodon can also be used in other parts of the product life cycle for design optimization and failure mode analysis and are therefore even more cost-effective.

September 1996
Bill Lokken

Return to White Paper Page