From Diagnostics to Prognostics with eXpress

Good Diagnostics as a Baseline

There's a general concensus that has finally been reached (one that DSI has long supported), which is that good prognostics requires good diagnostics. That is, a good diagnostic approach is essential before trying to migrate to prognostics. While this may not sound like a big deal, failure to recognize that prognostics needs diagnostics can result in the failure to either.

Of course, "talking the talk" and "walking the walk" are entirely different and one is left with the question of how to accomplish the transition. How does one begin with a good diagnostic approach and add prognostics without disturbing the diagnostic foundation? Our solution will focus on eXpress, specifically on two key features--its hybrid model and its test capability.

First, it is worth reiterating how good diagnostics are achieved. As is true of nearly all System Engineering processes, the process of developing a diagnostic solution is iterative. eXpress uses tests to capture diagnostic knowledge as part of the iterative process. With each iteration, the design is modified (more test points, new tests, etc) and the tests provide the mechanism by which those design changes can be assessed. Ultimately, and especially in the hands of skilled engineer, the diagnostics of the system can be improved to meet requirements (e.g. 98% detection, 95% isolation to 1 LRU).

What also happens throughout design capture, as well as test definition to a certain degree, is the addition of functions and failure modes. eXpress' ability to support both modes simultaneously through its hybrid diagnostic model provides the engineer with a powerful testing capability. Since functions and/or failure modes can be tested, the engineer can decide which approach best captures the diagnostic knowledge. Functional testing tends to be used the higher in the system one tests, or the earlier in the design process one is capturing diagnostic knowledge (that is, before failure modes are known).

Trade-offs for Diagnostics and Prognostics

During the maturation of the diagnostics, tests are best created in sets that represents types of testing. In this way, a test set will typically contain a group of similar tests like Initiated BIT. This grouping of tests is what enables simple diagnostic trade studies.

Incorporating Prognostics into the diagnostic approach requires understanding of a few basic principles:

Prognostics can't increase Detection Coverage
Prognostics can reduce ambiguity
Prognostics can detect failures before they occur
Prognostic Tests can detect failure before Diagnostic Tests

Therefore, we can conclude that as we add Prognostics, our detection percentage should remain the same. What we will hope for is detection before a critical failure, or fault isolation to a smaller fault group.

To add Prognostic knowledge to the design, the problem is broken into two parts:

Trend Analysis and Prediction
Detection of Impending Failure

Accounting for Prognostics Trend Analysis and Prediction

Trend analysis and prediction represents a processing capability that is added to the system. Most often, it runs on-board and supplies enough information to remediate before a critical failure. It can also be used for condition-based maintenance (CBM) and a host of other situations where detection prior to actual failure provides cost and safety benefits.

Accounting for this capability in the design itself is done by acknowledging the percentage of time that failures can be detected prior to their failure. That is, to what degree can insipient failures be recognized.

In eXpress, there are two approaches to adding this knowledge. First, let's consider the case where only functions exist, and there are no failure modes on the component of interest. In this case, eXpress' hybrid diagnostic model provides a quick and simple solution. Add a failure mode for each prognostic capability with a rate equal to the percentage of time that the insipient failure is identified before actual failure.

In the second case, both functions and failure modes already exist. Adding the prognostic prediction in this case means breaking existing failure modes into two or more pieces. One piece represents the percentage of time that the failure occurs suddenly without prior detection through trending, while other pieces represent those percentages of time that a trend is identified first.

Next, we will discuss the implications of these changes to the existing testing and diagnostic knowledge.

Accounting for Prognostic Detection Capabilities

If we added failure modes to a part that didn't previously have any failure modes, then there is no reconciliation that is needed. However, if failure modes already existed, we must consider whether any tests were already using those failure modes.

Failure mode tests that called out the failure modes we changed must be updated to now test for both failure modes. Testing for both failure modes ensures that the original detection percentage will remain the same. Having reconciled this split in failure modes with our original tests, we can now focus on building prognostic tests.

Prognostic tests identify impending failure. This only differs from standard diagnostic tests in regards to "when." As a result, one only need be concerned with two aspects of the Prognostic test:

When is the test used
What failures does the test cover

First, the test must be used prior to diagnostic tests (in our strategy) if we wish see the true impact of prognostics. This is accomplished by adding a test prerequisite from diagnostic tests to each respective prognostic test equivalent. This will ensure that if applicable, the prognostic test will have a chance to detect and/or isolate failure ahead of the diagnostic tests. Any benefits will be clearly identified in the fault groups that result.

Second, the failures that the test covers should be typically confined to those additional failure modes that were added to account for the capabilities of trend analysis and prediction. Although there may be unusual circumstances where a trend is only examined when another unrelated failure occurs, it is certainly not the norm.

Having accounting for both the capabilities of trend analysis (by adding failure modes) and the prognostic decision-making itself (by adding tests), eXpress now has the information necessary to accurately assess the benefits of prognostics. Keeping prognostic tests in their own test set also allows simple trade-off studies and reporting.