Related Topics:

Reliability Growth Planning

Reliability Growth Planning Examples

Crow-AMSAA (NHPP) Model

Crow Extended Model

Crow Extended - Continuous Evaluation Model

Crow Extended Model - Reliability Growth Planning

The Crow Extended model for reliability growth planning is a revised and improved version of the MIL-HDBK-189 growth curve [13]. MIL-HDBK-189 can be considered as the growth curve based on the Crow-AMSAA (NHPP) model. Using MIL-HDBK-189 for reliability growth planning assumes that the corrective actions for the observed failure modes are incorporated during the test and at the specific time of failure. However, in actual practice, fixes may be delayed until after the completion of the test or some fixes may be implemented during the test while others are delayed and some are not fixed at all. The Crow Extended model for reliability growth planning provides additional input to be able to account for a specific management strategy and delayed fixes with specified effectiveness factors.

Management Strategy Ratio & Initial Failure Intensity

When a system is tested and failure modes are observed, management can make one of two possible decisions, either to fix or not fix the failure mode. Therefore, the management strategy places failure modes into two categories: A modes and B modes. A modes are all failure modes such that, when seen during the test, no corrective action will be taken. This accounts for all modes for which management determines that it is not economically or otherwise justified to take a corrective action. B modes are either corrected during the test or the corrective action is delayed to a later time. The management strategy is defined by what portion of the failures will be fixed.

Let $\lambda _{I}$ be the initial failure intensity of the system in test. $\lambda _{A}$ is defined as the A mode initial failure intensity and $\lambda _{B}$ is defined as the B mode initial failure intensity. $\lambda _{A}$ is the failure intensity of the system that will not be addressed by corrective actions even if a failure mode is seen during test. $\lambda _{B}$ is the failure intensity of the system that will be addressed by corrective actions if a failure mode is seen during testing.

Then, the initial failure intensity of the system is:

MATH (1)

The initial system MTBF is:

MATH (2)

Based on the initial failure intensity definitions, the management strategy ratio is defined as:

MATH (3)

The $msr$ is the portion of the initial system failure intensity that will be addressed by corrective actions, if seen during the test.

The Type A and B failure mode initial failure mode intensity is:

MATH (4)
MATH (5)

Effectiveness Factor

When a delayed corrective action is implemented for a Type B failure mode, in other words a BD mode, the failure intensity for that mode is reduced if the corrective action is effective. Once a BD mode failure mode is discovered, it is rarely totally eliminated by a corrective action. After a BD mode has been found and fixed, a certain percentage of the failure intensity will be removed, but a certain percentage of the failure intensity will generally remain. The fraction decrease in the BD mode failure intensity due to corrective actions, $d$, MATH is called the effectiveness factor. A study on EFs showed that an average EF, $d,$, was about 70%. Therefore, typically about 30%, i.e. $100(1-d)\%$, of the BD mode failure intensity will remain in the system after all of the corrective actions have been implemented. However, individual EFs for the failure modes may be larger or smaller than the average. This average value of 70% can be used for planning purposes, or if such information is recorded, an average effectiveness factor from a previous reliability growth program can be used.

MTBF Goal

When putting together a reliability growth plan, a goal MTBF $M_{G}$ (or goal failure intensity $\lambda _{G}$) is defined as the requirement or target for the product at the end of the growth program.

Growth Potential

The failure intensity remaining in the system at the end of the test will depend on the management strategy given by the classification of the Type A and Type B failure modes. The engineering effort applied to the corrective actions determines the effectiveness factors. In addition, the failure intensity depends on $h(t)$, which is the rate at which problem failure modes are being discovered during testing. The rate of discovery drives the opportunity to take corrective actions based on the seen failure modes and it is an important factor in the overall reliability growth rate. The reliability growth potential is the limiting value of the failure intensity as time $T$ increases. This limit is the maximum MTBF that can be attained with the current management strategy. The maximum MTBF will be attained when all Type B modes have been observed and fixed.

If all seen Type B modes are corrected by time $T$, that is, no deferred corrective actions at time $T$, then the growth potential is the maximum attainable with the Type B designation of the failure modes and the corresponding assigned effectiveness factors. This is called the nominal growth potential. In other words, the nominal growth potential is the maximum attainable growth potential assuming corrective actions are implemented for every mode that is planned to be fixed. In reality, some fixes to modes might be implemented at a later time due to schedule, budget, engineering, etc.

If some seen Type B modes are not corrected at the end of the current test phase then the prevailing growth potential is below the maximum attainable with the Type B designation of the failure modes and the corresponding assigned effectiveness factors.

If all Type B failure modes are seen and corrected with an average effectiveness factor, $d$, then the maximum reduction in the initial system failure intensity is the growth potential failure intensity:

MATH (6)

The growth potential MTBF is:

MATH

Note that based Eqns. (6), (1) and (3), the initial failure intensity is equal to:

MATH (7)

Growth Potential Design Margin

The Growth Potential Design Margin ($GPDM$) can be considered as a safety margin when setting target MTBF values for the reliability growth plan. It is common for systems to degrade in terms of reliability when a prototype product is going into full manufacturing due to variation in material, processes, etc. Furthermore, the in-house reliability growth testing usually overestimates the actual product reliability, since the field usage conditions may not be perfectly simulated during growth testing. Typical values for the $GPDM$ are around 1.2. Higher values yield less risk for the program, but require a more rigorous reliability growth test plan. Lower values imply higher program risk, with less "safety margin."

During the planning stage, the growth potential MTBF, $M_{GP},$, can be calculated based on the goal MTBF, $M_{G},$, and the growth potential design margin, $GPDM$:

MATH (8)

or in terms of failure intensity:

MATH (9)

Nominal Idealized Growth Curve

During developmental testing, management should expect that certain levels of reliability will be attained at various points in the program in order to have assurance that reliability growth is progressing at a sufficient rate to meet the product reliability requirement. The idealized curve portrays an overall characteristic pattern, which is used to determine and evaluate intermediate levels of reliability and construct the program planned growth curve. Note that growth profiles on previously developed, similar systems provide significant insight into the reliability growth process and are valuable in the construction of idealized growth curves.

The nominal idealized growth curve portrays a general profile for reliability growth throughout system testing. The idealized curve has the baseline value $\lambda _{I}$ until an initialization time, $t_{0},$ when reliability growth occurs. From that time and until the end of testing, which can be one or, most commonly, multiple test phases, the idealized curve increases steadily according to a learning curve pattern until it reaches the final reliability requirement, $M_{F}$. The slope of this curve on a log-log plot is the growth rate of the Crow Extended model [13].

Nominal Failure Intensity Function

The nominal idealized growth curve failure intensity as a function of test time $t$ is:

MATH (10)

and:

MATH (11)

where $\lambda _{I}$ is the initial system failure intensity, $t$ is test time and $t_{0}$ is the initialization time, which is discussed in the next section.

It can be seen that Eqn. (10) is the failure intensity equation of the Crow Extended model.

Initialization Time

Reliability growth can only begin after a Type B failure mode occurs, which cannot be at a time equal to zero. Therefore, there is a need for an initialization time, different than zero, to be defined. The nominal idealized growth curve failure intensity is initially set equal to the initial failure intensity, $\lambda _{I},$, until the initialization time, $t_{0}$:

MATH

Therefore:

MATH (12)

Then:

MATH (13)

Using Eqn. (1) to substitute $\lambda _{I}$ we have:

MATH

Then:

MATH

The initialization time, $t_{0},$, allows for growth to start after a Type B failure mode has occurred.

Nominal Time to Reach Goal

Assuming that we have a target MTBF or failure intensity goal, we can solve Eqn. (10) to find out how much test time,$\ t_{N,G}$, is required, (based on the Crow Extended model and the nominal idealized growth curve) to reach that goal:

MATH (14)

Note that when $\lambda _{I}<r_{G}$ or, in other words, the initial failure intensity is lower than the goal failure intensity, there is no need to solve for the nominal time to reach the goal, because the goal is already met. In this case, no further reliability growth testing is needed.

Growth Rate for Nominal Idealized Curve

The growth rate for the nominal idealized curve is defined in the same context as the growth rate for the Duane Postulate [8]. The nominal idealized curve has the same functional form for the growth rate as the Duane Postulate and the Crow-AMSAA (NHPP) model.

For both the Duane Postulate and the Crow-AMSAA (NHPP) model, the average failure intensity is given by:

MATH

Also, for both the Duane Postulate and the Crow-AMSAA (NHPP) model, the instantaneous failure intensity is given by:

MATH

Taking the difference, $D(t),$ between the average failure intensity, $C(t)$, and the instantaneous failure intensity, $r(t)$, yields:

MATH

Then:

MATH

For reliability growth to occur, $D(t)$ must be decreasing.

The growth rate for both the Duane Postulate and the Crow-AMSAA (NHPP) model is the negative of the slope of $\log (D(t))$ as a function of $\log (t)$:

MATH

The slope is negative under reliability growth and equals:

MATH

The growth rate for both the Duane Postulate and the Crow-AMSAA (NHPP) model is equal to the negative of this slope:

MATH

The instantaneous failure intensity for the nominal idealized curve is:

MATH

The cumulative failure intensity for the nominal idealized curve is:

MATH

Therefore:

MATH

and:

MATH

Therefore, in accordance with the Duane Postulate and the Crow-AMSAA (NHPP) model, $a=1-\beta $ is the growth rate for the reliability growth plan.

Lambda - Beta Parameter Relationship

Under the Crow-AMSAA (NHPP) model, the time to first failure is a Weibull random variable. The MTTF of a Weibull distributed random variable with parameters $\beta $ and $\eta $ is:

MATH (15)

The parameter lambda is defined as:

MATH (16)

Using Eqn. (16), the MTTF relationship shown in Eqn. (15) becomes:

MATH (17)

Or, in terms of failure intensity:

MATH (18)

Actual Idealized Growth Curve

The actual idealized growth curve differs from the nominal idealized curve in that it takes into account the average fix delay that might occur in each test phase. The actual idealized growth curve is continuous and goes through each of the test phase target MTBFs.

Fix Delays and Test Phase Target MTBF

Fix delays reflect how long it takes from the time a problem failure mode is discovered in testing, to the time the corrective action is incorporated into the system and reliability growth is realized. The consideration of the fix delay is often in terms of how much calendar time it takes to incorporate a corrective action fix after the problem is first seem. However, the impact of the delay on reliability growth is reflected in the average test time it takes between finding a problem failure mode and incorporating a corrective action. The fix delay is reflected in the actual idealized growth curve in terms of test time.

In other words, the average fix delay is calendar time converted to test hours. For example, say that we expect an average fix delay of two weeks: if in two weeks the total test time is 1000 hours, the average fix delay is 1000 hours. If in the same two weeks the total test time is 2000 hours (maybe there are more units available or more shifts) then the average fix delay is 2000 hours.

There can be a constant fix delay across all test phases or, as a practical matter, each test phase can have a different fix delay time. In practice, the fix delay will generally be constant over a particular test phase. $L_{i}$ denotes the fix delay for phase $i=1,...,P,$ where $P$ is the total number of phases in the test. RGA 7 allows for a maximum of seven test phases.

Actual Failure Intensity Function

Consider a test plan consisting of $i$ phases. Taking into account the fix delay within each phase, we expect the actual failure intensity to be different (i.e. shifted) from the nominal failure intensity. This is because fixes are not incorporated instantaneously, thus growth is realized at a later time compared to the nominal case.

Specifically, the actual failure intensity will be estimated as follows:

Test Phase 1

For the first phase of a test plan, the actual idealized curve failure intensity, $r_{AI}(t)$, is$:$

MATH

Note that the end time of Phase 1, $T_{1},$ must be greater than $L_{1}+t_{0}$. That is, $T_{1}>L_{1}+t_{0}$.

The actual idealized curve initialization time for Phase 1, $T_{0}^{AIC},$ is calculated from:

MATH

Where MATH

Therefore, using Eqn. (12):

MATH (19)

Solving Eqn. (19) for $T_{0}^{AIC}$ we get:

MATH (20)

Test Phase $\QTR{bf}{i}$

For any test phase $i$, the actual idealized curve failure intensity is given by:

MATH (21)

where MATH and $T_{i}$ is the test time of each corresponding test phase.

The actual idealized curve MTBF is:

MATH (22)

Actual Time to Reach Goal

The actual time to reach the target MTBF or failure intensity goal, $t_{AC,G},$,$t_{AC,G},$, can be found by solving Eqn. (21):

MATH

Since the actual idealized growth curve depends on the phase durations and average fix delays, there are three different cases that need to be treated differently in order to determine the actual time to reach the MTBF goal. The cases depend on when the actual MTBF that can be reached within the specific phase durations and fix delays becomes equal to the MTBF goal. This can be determined by solving Eqn. (21) for phases $1$ through $i$, then solving in terms of actual MTBF using Eqn. (22) for each phase and finding the phase during which the actual MTBF becomes equal to the goal MTBF. The three cases are presented next.

Case 1: MTBF goal is met during the last phase

If $T_{F}$ indicates the cumulative end phase time for the last phase and $L_{F}$ indicates the fix delay for the last phase, then we have:

MATH

Starting to solve for $t_{AC,G}$ yields:

MATH

We can substitute the left term by using Eqn. (14), thus we have:

MATH

Therefore:

MATH (23)

Case 2: MTBF goal is met before the last phase

Eqn. (23) still applies, but in this case $T_{F}$ and $L_{F} $ are the time and fix delay of the phase during which the goal is met.

Case 3: MTBF goal is met after the final phase

If the goal MTBF, $M_{G},$, is met after the final test phase, then the actual time to reach the goal is not calculated, since additional phases have to be added with specific duration and fix delays. The reliability growth program needs to be re-evaluated with the following options:

Other alternative routes for consideration would be to investigate the rest of the inputs in the model:

Note that each change of input variables into the model can significantly influence the results. With that in mind, any alteration in the input parameters should be justified by actionable decisions that will influence the reliability growth program. For example, increasing the average effectiveness factor value should be done only when there is proof that the program will pursue a different, more effective path in terms of addressing fixes.