Research > Scheduling > SimYard

SimYard
SimYard is a stochastic shipyard simulation tool that can model a wide variety of shipyard production conditions, including problems that arise and how the shipyard reacts to those problems.  It was built to evaluate ARGOS under actual production conditions to validate theoretical labor savings estimates.

The main challenge in SimYard to take performance data describing how a particular shipyard performs and tune SimYard to behave accordingly; this requires a model of SimYard itself.  This multivariate calibration problem is tackled using inverse regression methods.

There are two main inputs to SimYard:
  • Historical data: A completed project as well as performance data describing how things went (how often constraints were broken, tasks were delayed, etc.).
  • New project and schedule:  These two pieces of data represent the particular situation of interest to be simulated.
SimYard involves two main steps.  In the first, hundreds of simulations are done with the historical project, using a wide range of possible input parameters (how often workers are sick, how often durations are under-estimated, etc).  Using inverse regression, the historical performance data and the SimYard performance data (how things went during these simulations) can be combined to determine appropriate input parameters for the shipyard in question.

In the second step, the appropriate input parameters discovered in the first step are used to tune SimYard; simulations are then done with the new project and schedule.  The output data from these simulations describe how the shipyard can be expected to behave on that project with the given schedule.  For example, outputs include the likelihood that deadlines will be missed and the fraction of tasks likely to be paused.  Output data also includes cost information; therefore, by simulating both a shipyard schedule and an ARGOS schedule and compariing the resulting costs, labor savings likely to result from ARGOS can be obtained.

The approach used by SimYard has been validated in a number of ways, including internal consistency tests (comparing it with itself on simple problems), external  consistency tests (running SimYard against production shipyard data), and evaluation by outside experts.

Results

For a large shipyard project (7000 tasks), while an ARGOS schedule saves 14% percent of labor costs in theory, the following SimYard graph shows that, in reality, that schedule can be expected to save between 5 and 15%, with savings greater than 9% on average (more than 50% of the time): 

ARGOS savings predicted by SimYard

SimYard provides a number of ways to present the data to help understand how the simulations behave and how that behavior is affected by various parameters.  The following is a simple example:

How savings are affected by resource surprises

This graph illustrates that, as the mean manpower paremeter (how many people are needed to do a task, on average, relative to the number of people predicted) increases, the average savings of the ARGOS schedule decrease (but only slightly).