MODEL TESTING. The Contractor shall prepare a Model Test Program Plan (DI-031). The Contractor shall conduct the following model tests (including construction of all hull and propeller models):
MODEL TESTING. The scripting approach can also be useful for model testing and testcase generation. For model testing, we can save and replay execution traces. Saving a trace is done by generating a groovy script and replaying is done by executing that script. In contrast to previous approaches [Elb09] this scripting approach is not limited to simple linear traces because we have the full expressivity of the host language at our hands. The new ProB plug-in has a built in JUnit4 test runner. There is also built in support for the Spock test framework1. This means that users can write JUnit4 or Spock tests using the ProB abstractions. Tests are written within groovy scripts. In order to use the BUnit test runner, the user has to write a test class that is compatible with JUnit4 or Spock and then create an new instance of the test class in the last line of the script. Then, when the script is run, the groovy engine will recognize the class and run it through the test runner. An example of a model test for the watertank example from chapter 6 written as a Spock specification could look like this import spock.lang.Specification import de.prob.scripting.Api class WaterTankTest extends Specification { static Api api def ctrl def setup() { ctrl = api.b_load("waterTankCtrl.mch") as History ctrl = ctrl.anyEvent(); // setup constants ctrl = ctrl.anyEvent(); // initialize machine 1xxxx://xxx.xxxxxxxxxxxxxx.xxx } def "Initially the pump is turned on"() { expect: // This test will fail! The value is FALSE assert ctrl.getCurrentState().value("pump") == "TRUE" } def "If we get below low level, pump is turned on"() { when: ctrl = ctrl.fmiReadInputs("l=799") // set level < Low ctrl = ctrl.readLevel().decide().writePump() // decide then: assert ctrl.getCurrentState().value("pump") == "TRUE" } } WaterTankTest.api = api new WaterTankTest()
MODEL TESTING. The model was tested with a dataset containing empirical data from which the ART score was derived for each observation (e.g. weight fraction) along with the measured output (e.g. inhalation exposure). The dataset used to test the model contained more than 600 data points. Where this was the case, the missing inputs were imputed by empirical sampling from existing values in the available data. To summarize the output, the geometric mean, median, 2.5th percentile and 97.5th percentile were calculated. Thereafter, the geometric mean of the output against the true measured values with 95% confidence intervals, medians and geometric means of the output against the true values (all on log10 scale) were plotted for comparison. For each sub-scenario, 10,000 iterations were used. The models were only compared or tested with available exposure data. Thus no calibration or transforming/fitting was performed. Results from the model testing are presented in figures 1 to 4. For all the sub-scenario, The y-axis represents the measured potential inhalation exposure. The x-axis represents the geometric mean of the model outputs. Blue dots indicate imputed values, red dots indicate that art scores were available. Figures 1 to 4 show the result of the model testing. Art default values for HHA air concentration (Table 1) were tested against measured data representative of HHA application activities.
