Package {PINNProgCens}


Type: Package
Title: Physics-Informed Neural Networks for Progressive Censoring
Version: 0.1.0
Description: Implementation of Physics-Informed Neural Networks ('PINN') for lifetime estimation under progressive Type-II censoring schemes. Combines parametric baseline hazards with physical differential degradation models.
License: MIT + file LICENSE
Encoding: UTF-8
Imports: deSolve, stats
Config/roxygen2/version: 8.0.0
NeedsCompilation: no
Packaged: 2026-08-05 07:02:56 UTC; Dr. O. J. Obulezi
Author: Okechukwu J. Obulezi [aut, cre]
Maintainer: Okechukwu J. Obulezi <oj.obulezi@unizik.edu.ng>
Repository: CRAN
Date/Publication: 2026-08-09 08:10:15 UTC

Physics Residuals at Collocation Points

Description

Physics Residuals at Collocation Points

Usage

compute_ode_residuals(sol_colloc, z_colloc, phi, ode_func)

Arguments

sol_colloc

Solved ODE states across collocation grid.

z_colloc

Collocation evaluation grid points.

phi

Physics differential parameters.

ode_func

Differential system function.

Value

Vector of physical model residuals.


Default Physical ODE System (Degradation Process)

Description

Default Physical ODE System (Degradation Process)

Usage

default_ode_system(t, h, phi)

Arguments

t

Time variable.

h

Current degradation state.

phi

Parameter vector governing the differential equation.

Value

List containing derivative dh/dt.


Maximum Likelihood Estimation via PINN Optimization

Description

Maximum Likelihood Estimation via PINN Optimization

Usage

fit_pinn_progcens_mle(
  times,
  R,
  z_colloc,
  ode_func = default_ode_system,
  max_epochs = 50
)

Arguments

times

Observed progressive censored failure times.

R

Progressive removal counts.

z_colloc

Collocation points for differential equation loss.

ode_func

Differential system function.

max_epochs

Maximum optimization iterations.

Value

List containing parameter estimates and standard errors.

Examples

times_obs <- c(0.85, 1.42, 2.10)
R_vec     <- c(1, 0, 1)
colloc    <- seq(0.1, 3.0, length.out = 5)
fit_pinn_progcens_mle(times_obs, R_vec, colloc, max_epochs = 5)

Interpolate Trajectory State

Description

Interpolate Trajectory State

Usage

interpolate_state(tau, sol_obs)

Arguments

tau

Intermediary integration time.

sol_obs

Trajectory solution matrix.

Value

Scalar estimated state value.


Physics-Regularized Hazard Rate Mapping

Description

Physics-Regularized Hazard Rate Mapping

Usage

pinn_hazard(t, alpha, beta, h_t, psi)

Arguments

t

Evaluation time point.

alpha

Weibull scale parameter.

beta

Weibull shape parameter.

h_t

Degradation state at time t.

psi

List containing network weights W2 and bias b2.

Value

Hazard rate value.

Examples

pinn_hazard(1.0, 1.0, 1.0, 0.5, list(W2 = matrix(1,1,1), b2 = 0))

Combined Structural Loss (Censored Likelihood + Physics Residual)

Description

Combined Structural Loss (Censored Likelihood + Physics Residual)

Usage

pinn_progcens_loss(params, times, R, z_colloc, ode_func, mu = 0.1)

Arguments

params

List of current candidate parameters (alpha, beta, phi, psi).

times

Vector of observed failure times.

R

Vector of progressive censorship removals.

z_colloc

Vector of physics collocation evaluation points.

ode_func

Physics ODE differential system function.

mu

Physics loss regularization hyperparameter penalty weight.

Value

Total scalar PINN structural loss value.


Softplus Activation Function

Description

Softplus Activation Function

Usage

softplus(x)

Arguments

x

Numeric vector or matrix input.

Value

Smooth approximation to ReLU.

Examples

softplus(c(-1, 0, 1))

Forward State Trajectory Integration

Description

Forward State Trajectory Integration

Usage

solve_ode_trajectory(times, phi, ode_func)

Arguments

times

Vector of evaluation time points.

phi

Differential parameters.

ode_func

Right-hand-side differential equation function.

Value

Matrix of interpolated trajectories at evaluated times.