Neyer Software LLC announces the release of SenTest™ Version 1.0, Sensitivity Test and Analysis software. The software is a full function Windows® based test and analysis program for conducting and analyzing Sensitivity Tests. This software replaces a number of separate DOS programs which have been in use since 1988. The software allows the user to efficiently choose test levels using the Neyer D-Optimal test method and to analyze the results of the tests using the Likelihood Ratio Test method. These techniques are widely used in a number of industries. They are the methods of choice for testing explosive components for the aerospace, automotive airbag, and military markets. They are the prefered method of the OEDC Guidelines for testing of chemicals. Testing material for fracture toughness and analysis of the effects of brain pressure on stroke are other applications.
The experimenter specifies initial guesses for the parameters of the distribution. SenTest™ then guides the user by picking the stimulus levels. The experimenter tells the software whether the specimen responded or failed to respond to the stimulus, and SenTest™ then picks the next stimulus level.
In addition to picking the test levels, SenTest™ also contains a variety of analysis tools to analyze the results of the test. Any combination of these analysis views can be selected at any time, including during the performance of the test. The analysis is performed using the utilizing the likelihood ratio method. This method has been shown to produce more reliable confidence regions than the older methods. These analysis tools display the results on the analysis in a total of 6 different analysis views. These are the numerical analysis, history plot, probability plot, contour plot, multi contour plot, and different populations.
Description of Sensitivity Tests
SenTest™ uses the well established Neyer D-Optimal test procedure that utilizes knowledge of all of the past responses in order to more efficiently choose stimulus levels. This is the same method as the earlier DOS program Optimal. Because it continuously refines the knowledge of the parameters of the population, the procedure used by SenTest™ is much more efficient for estimating parameters of the population than a Probit Test, Bruceton Test, Robbins-Monro Test, Langlie Test, or any other standard sensitivity test. Also, unlike all other sensitivity tests, SenTest™ is rather independent of the initial guesses of parameters. A factor of ten uncertainty in guesses for the parameters still allows for a reasonably efficient test!
In addition to using the Neyer D-Optimal test, SenTest™ will also chose test levels according to the (Original) Neyer Test, Bruceton Test, Robbins-Monro Test, Langlie Test, and Adaptive Langlie Test methods, depending on the options ordered. These test methods are included for historical purposes only and should not be used if efficient testing is required. SenTest™ includes the functionality of the following DOS programs: Bruceton, Langlie, OldTest, Optimal, and Sensit.
The Bruceton test was developed during the 1940s. It was developed, not as an efficient test method, but rather a method which allowed easy analysis of the tests results. The Bruceton test uses a starting point and a fixed step size. The efficiency of the test is critically dependent on the step size chosen.
The Robbins-Monro test was developed during the 1950s. It was developed to "home in" on the mean value. The Robbins-Monro test uses a starting point and a decreasing step size. If the starting point is not chosen properly, or if the initial step size is too small, the test could waste many samples getting close to the mean.
The Langlie test was developed during the 1960s. It was developed to be a less parameter dependent test than the Bruceton test. With the advent of electronic computers, the restriction on ability to perform analysis that motivated the Bruceton test test was no longer an issue. The Langlie test uses a lower and upper limit and averages the last test with a previous level.
The Adaptive Langlie test, developed during the 1970s, is, as the name implies, an adaptation of the original Langlie test. It was developed to overcome the problem of specification of one test limit too close to the mean value. The test protocol is the same, except that the levels shift up if a test level near the upper limit results in a failure, with a similar shift down if a test level near the lower level results in a success.
The original Neyer test was developed during the late 1980s. The motivation was to design a test that would use knowledge of all of the test results to pick the test levels most efficiently. It was further refined and became known as the Neyer D-Optimal test (developed in the 1990's). With the advent of computers in the laboratory, it was finally possible to perform the difficult calculations needed to choose as the test level that one level which would provide the most information. The Neyer D-Optimal test uses a lower and upper guess for the mean and a guess for the standard deviation. Unlike the Bruceton test and Langlie test, the Neyer D-Optimal test is almost completely independent of the experimenters guess for the parameters.
Description of Sensitivity Analysis
The primary analysis method that SenTest™ uses is the likelihood ratio method. This method of analyzing sensitivity tests can be used to analyze the results of any sensitivity test. It is more general than the test specific methods that had been developed to analyze the results of Bruceton and other tests. Unlike any other known analysis method, it can also be used to analyze the results of tests where there is no unique estimate of the standard deviation (no overlap of the mixed test results).
In addition to performing the analysis using the likelihood ratio method, SenTest™ will also analyze tests using the ASENT (asymptotic), Bruceton Analysis, and Langlie analysis methods, depending on the options ordered. These analysis methods are included for historical purposes only and should not be used if reliable analysis is required. All of the analysis methods with the exception of different populations can be performed with these alternative analysis methods. SenTest™ includes the functionality of the following DOS programs: ASENT, BrucAnal, ComSen, LangAnal, MuSig, PlotSen, and ProbPlot.
The Numerical analysis computes the maximum likelihood estimates of several parameters (the mean, standard deviation, and requested response levels). It also calculates confidence regions for these parameters.
The History Plot is used to provide a graphical analysis of sensitivity tests. The data of the test are plotted as a function of stimulus level versus specimen number. Successes are plotted in one color with an "X" and failures in another color with a "0". The routine also prints the values of the parameters Mu and Sigma, as well as displaying a graph of both the probability curve and integrated curve.
The Probability Plot provides plots of the probability versus stress level. Plotted are the maximum likelihood estimates of probability as well as the specified confidence bands. The graph shows the expected probability of success versus stress level. Also plotted are various confidence bands for upper and lower limits of these stress values. The user also has the option of displaying the test data overlaid with the probability graph.
The Contour Plot is used to provide a plot of the confidence regions as a function of confidence. The confidence regions are those regions that contain the true parameters a certain fraction of the time. These confidence regions are displayed as a contour map. The contour
The Multi Contour Plot is used to provide a plot of the confidence regions as a function of confidence. This analysis displays the confidence regions of a number of tests simultaneously. The confidence regions are those regions that contain the true parameters a certain fraction of the time. These confidence regions are displayed as a contour map.
The Different Populations analysis is used to compare the results of two or more test sequences to see if the samples were chosen from similar or different populations. This analysis method uses the likelihood ratio method to compare the two sets of data. It can be used to compare results of two tests conducted with any test method. It can even compare the results of two tests conducted under different procedures with different sample sizes.