Each tab in the Constant Stress ALT Design platform Test Plans section summarizes a specified plan. Sections include plan comparisons and diagnostics.
Test Settings and Acceleration Model
The table provides a summary of the plan specifications that define the test plans that are contained in the tab.
For each set of Constant Stress ALT plan specifications, three test plans are generated. An alert appears when a plan is not capable of estimating the model parameters. The three test plans to consider for your test are the optimal, compromise, and balanced plans.
Optimal Plan
A plan built to optimize the large-sample approximate variance of the optimality criterion specified in the plan settings. Compared to the compromise and balanced plans, the optimal plan typically has the fewest test settings. However, it is also the most dependent on the assumptions of the underlying model.
Note: Optimal designs are provided as a reference against which to compare the compromise and balanced plans. The optimal design might have too few test settings for estimating all of the acceleration model parameters.
Compromise Plan
A plan built to optimize the large-sample variance of the optimality criterion specified in the plan settings with the requirement of a middle test setting. This setting is between the lowest and highest settings with a minimum proportion of test units allocated to the setting. The default for the minimum proportion is 10%. Use the Set Minimum Proportion option in the Plan Comparisons red triangle menu to change the proportion.
Balanced Plan
A plan built to provide every combination of test settings, based on the number of levels per factor. Compared to the optimal and compromise plans, the balanced plan typically has the most test settings with an equal distribution of test units across settings.
The Constant Stress ALT Design platform Diagnostics section has a table of metrics for each plan type.
Approx. Std Dev of Log Quantile
The approximate standard deviation of the logarithm of the quantile estimate. It is computed using a large-sample approximation based on the Fisher information.
Approx. Std Dev of Log Failure Probability
The approximate standard deviation of the logarithm of the failure probability. It is computed using a large-sample approximation based on the Fisher information.
R Precision Factor
The R Precision Factor (for a 95% CI) is a measure of the precision of the 95% confidence interval for the quantile of interest. A value less than or equal one is optimal.
Use the simulator in the Constant Stress ALT Design platform to explore your accelerated model before collecting data. Use simulation to assess how well the asymptotic criteria represent the actual uncertainty in the quantile or failure time estimates. The results include a table of summary statistics and a plot of simulated model fits for each factor. When there is more than one factor, each plot is based on the levels of the other factors being fixed to their specified use conditions.
The initial simulation parameter values are set to the plan specification values. Adjust the settings to explore differences in plans, the probability of interest, or use conditions.
Plan Type
The plan type to use for the simulation.
Number of Simulations
The number of model fits to simulate.
Simulation Probability of Interest
The failure probability of interest.
Simulation Usage
The usage value for the specified factor.
Once you have specified your simulation parameters, click Run Simulations to run the simulation and obtain simulated model fits.
In each model fit plot, the red curve represents the model fit that is calculated from the planning values. The black curves are the simulated model fits. The curves demonstrate the variability one could expect for the given plan under the specified simulation parameters.
The simulated values for the quantile or failure probability and R precision should be close to the diagnostic values based on the planning values. The simulations are intended to evaluate the asymptotic theory of the accelerated test. If the simulated performance values differ greatly from the performance values that are based on the planing values, then one might question the asymptotic theory assumptions.
The CSALT diagnostics section provides three diagnostic plots.
Distribution Profiler
A profiler to evaluate the impact of plan type and factor(s) on the failure probabilities. The probabilities are based on the model assumptions.
R Precision Factor for a 95% CI
A profiler to evaluate the impact of plan type, number of units on test, level of the factor(s), and the quantile of interest on the R Precision Factor.
Model Sensitivity Profiler
A profiler to evaluate the impact of plan type, number of units on test, and failure time estimates for the settings specified in the Acceleration Model Settings on the large-sample approximate variance of the metric of interest. Above this profiler, there is a reference table summarizing the factor levels and quantile specified in the Acceleration Model Settings for each setting. Use this profiler to evaluate how sensitive the approximate variance is to failure time assumptions.