Piston Simulation Function

import numpy as np
import matplotlib.pyplot as plt
import uqtestfuns as uqtf

The Piston simulation test function is a seven-dimensional scalar-valued function. The function computes the cycle time of a piston system.

This function has been used as a test function in metamodeling exercises [BAS07]. A 20-dimensional variant was used in the context of sensitivity analysis [Moo10] by introducing 13 additional inert input variables.

Test function instance

To create a default instance of the piston simulation test function:

my_testfun = uqtf.Piston()

Check if it has been correctly instantiated:

print(my_testfun)

Function ID      : Piston
Input Dimension  : 7 (fixed)
Output Dimension : 1
Parameterized    : False
Description      : Piston simulation model from Ben-Ari and Steinberg (2007)
Applications     : metamodeling, sensitivity

Description

The Piston simulation computes the cycle time of a piston moving inside a cylinder using the following analytical expression:

\[\begin{split} \begin{align} \mathcal{M}(\boldsymbol{x}) & = 2 \pi \left( \frac{M}{k + S^2 \frac{P_0 V_0}{T_0} \frac{T_a}{V^2}} \right)^{0.5}, \\ V & = \frac{S}{2 k} \left[ \left(A^2 + 4 k \frac{P_0 V_0}{T_0} T_a \right)^{0.5} - A \right]^{0.5}, \\ A & = P_0 S + 19.62 M - \frac{k V_0}{S}, \end{align} \end{split}\]

where \(\boldsymbol{x} = \{ M, S, V_0, k, P_0, T_a, T_0 \}\) is the seven-dimensional vector of input variables further defined below.

Probabilistic input

Two probabilistic input model specifications for the OTL circuit function are available as shown in the table below.

No.	Keyword	Source
1.	BenAri2007 (default)	[BAS07]
2.	Moon2010	[Moo10]

The default selection, based on [BAS07], contains seven input variables given as independent uniform random variables with specified ranges shown in the table below.

Show code cell source Hide code cell source

print(my_testfun.prob_input)

Function ID     : Piston
Input ID        : BenAri2007
Input Dimension : 7
Description     : Probabilistic input model for the Piston simulation model
                  from Ben-Ari and Steinberg (2007).
Marginals       :

 No.    Name    Distribution      Parameters              Description
-----  ------  --------------  -----------------  ----------------------------
  1      M        uniform          [30. 60.]           Piston weight [kg]
  2      S        uniform        [0.005 0.02 ]     Piston surface area [m^2]
  3      V0       uniform        [0.002 0.01 ]      Initial gas volume [m^3]
  4      k        uniform        [1000. 5000.]      Spring coefficient [N/m]
  5      P0       uniform      [ 90000. 110000.]  Atmospheric pressure [N/m^2]
  6      Ta       uniform         [290. 296.]       Ambient temperature [K]
  7      T0       uniform         [340. 360.]     Filling gas temperature [K]

Copulas         : Independence

Note

In [Moo10], 13 additional inert independent input variables are introduced (totaling 20 input variables); these input variables, being inert, do not affect the output of the function.

To create an instance of the piston simulation test function with the probabilistic input specified in [Moo10], pass the corresponding keyword ("Moon2010") to the parameter input_id):

my_testfun = uqtf.Piston(input_id="Moon2010")

Reference results

This section provides several reference results of typical UQ analyses involving the test function.

Sample histogram

Shown below is the histogram of the output based on \(100'000\) random points:

Show code cell source Hide code cell source

np.random.seed(42)
xx_test = my_testfun.prob_input.get_sample(100000)
yy_test = my_testfun(xx_test)

plt.hist(yy_test, bins="auto", color="#8da0cb");
plt.grid();
plt.ylabel("Counts [-]");
plt.xlabel("$\mathcal{M}(\mathbf{X})$");
plt.gcf().set_dpi(150);

Moments estimation

Shown below is the convergence of a direct Monte-Carlo estimation of the output mean and variance with increasing sample sizes.

Show code cell source Hide code cell source

# --- Compute the mean and variance estimate
np.random.seed(42)
sample_sizes = np.array([1e1, 1e2, 1e3, 1e4, 1e5, 1e6, 1e7], dtype=int)
mean_estimates = np.empty(len(sample_sizes))
var_estimates = np.empty(len(sample_sizes))

for i, sample_size in enumerate(sample_sizes):
    xx_test = my_testfun.prob_input.get_sample(sample_size)
    yy_test = my_testfun(xx_test)
    mean_estimates[i] = np.mean(yy_test)
    var_estimates[i] = np.var(yy_test)

# --- Compute the error associated with the estimates
mean_estimates_errors = np.sqrt(var_estimates) / np.sqrt(np.array(sample_sizes))
var_estimates_errors = var_estimates * np.sqrt(2 / (np.array(sample_sizes) - 1))

# --- Do the plot
fig, ax_1 = plt.subplots(figsize=(6,4))

ax_1.errorbar(
    sample_sizes,
    mean_estimates,
    yerr=mean_estimates_errors,
    marker="o",
    color="#66c2a5",
    label="Mean",
)
ax_1.set_xlabel("Sample size")
ax_1.set_ylabel("Output mean estimate")
ax_1.set_xscale("log");
ax_2 = ax_1.twinx()
ax_2.errorbar(
    sample_sizes + 1,
    var_estimates,
    yerr=var_estimates_errors,
    marker="o",
    color="#fc8d62",
    label="Variance",
)
ax_2.set_ylabel("Output variance estimate")

# Add the two plots together to have a common legend
ln_1, labels_1 = ax_1.get_legend_handles_labels()
ln_2, labels_2 = ax_2.get_legend_handles_labels()
ax_2.legend(ln_1 + ln_2, labels_1 + labels_2, loc=0)

plt.grid()
fig.set_dpi(150)

The tabulated results for each sample size is shown below.

Show code cell source Hide code cell source

from tabulate import tabulate

# --- Compile data row-wise
outputs = []
for (
    sample_size,
    mean_estimate,
    mean_estimate_error,
    var_estimate,
    var_estimate_error,
) in zip(
    sample_sizes,
    mean_estimates,
    mean_estimates_errors,
    var_estimates,
    var_estimates_errors,
):
    outputs += [
        [
            sample_size,
            mean_estimate,
            mean_estimate_error,
            var_estimate,
            var_estimate_error,
            "Monte-Carlo",
        ],
    ]

header_names = [
    "Sample size",
    "Mean",
    "Mean error",
    "Variance",
    "Variance error",
    "Remark",
]

tabulate(
    outputs,
    headers=header_names,
    floatfmt=(".1e", ".4e", ".4e", ".4e", ".4e", "s"),
    tablefmt="html",
    stralign="center",
    numalign="center",
)

Sample size	Mean	Mean error	Variance	Variance error	Remark
1.0e+01	5.2537e-01	4.1089e-02	1.6883e-02	7.9588e-03	Monte-Carlo
1.0e+02	4.9583e-01	1.4869e-02	2.2110e-02	3.1426e-03	Monte-Carlo
1.0e+03	4.5979e-01	4.5237e-03	2.0464e-02	9.1563e-04	Monte-Carlo
1.0e+04	4.6528e-01	1.3772e-03	1.8966e-02	2.6824e-04	Monte-Carlo
1.0e+05	4.6324e-01	4.4313e-04	1.9637e-02	8.7819e-05	Monte-Carlo
1.0e+06	4.6249e-01	1.3957e-04	1.9480e-02	2.7549e-05	Monte-Carlo
1.0e+07	4.6248e-01	4.4112e-05	1.9459e-02	8.7024e-06	Monte-Carlo

References

[Moo10] (1,2,3,4)

Hyejung Moon. Design and analysis of computer experiments for screening input variables. PhD thesis, Ohio State University, Ohio, 2010. URL: http://rave.ohiolink.edu/etdc/view?acc_num=osu1275422248.

[BAS07] (1,2,3)

Einat Neumann Ben-Ari and David M. Steinberg. Modeling data from computer experiments: an empirical comparison of kriging with MARS and projection pursuit regression. Quality Engineering, 19(4):327–338, 2007. doi:10.1080/08982110701580930.