degradr (version 1.0.1)

fit_healthindex: Fitting a Health Index Model from Multivariate Signals

Description

Fits a health index–based degradation model by projecting multivariate sensor signals into a univariate health index and modeling its evolution using a linear mixed-effects model.

Usage

fit_healthindex(data,
  type = "exponential",
  method = "lm",
  degree = 2,
  phi = NULL,
  r = 0.5)

Value

Returns an object of class "healthindex", which contains:

index

A list with components: weights w, offset phi, and raw projections.

model

A fitted mixed-effects model of the health index over time.

Arguments

data

A data frame containing sensor readings over time. Must include the columns t (time), unit (unit identifier), and multiple degradation signals. All degradation signals must have an upward trend.

type

Model type. Either "linear" or "exponential". The exponential model applies a logarithmic transformation to x - phi. Default is "exponential".

method

Estimation method. Either "nlme" to fit a nonlinear mixed-effects model using nlme::lme(), or "lm" to fit separate linear models per unit and estimate fixed and random effects from the set of coefficients. Default is "lm".

degree

Degree of the polynomial model. Default is 2. The fixed and random effects will include powers of time up to the specified degree.

phi

Initial degradation level for non-defective units. Used in the exponential model as a fixed offset to ensure that the logarithmic transformation is valid and interpretable. If NULL, it is automatically estimated as a value slightly below the minimum observed degradation level. Ignored when type = "linear".

r

parameter that controls the relative importance of the threshold variance and the weighted residual sum of squares in the index-fitted degradation model.

Details

This function implements a two-stage modeling strategy. In the first stage, a univariate health index is constructed as a weighted linear combination of the input signals, using correlation-based shrinkage. In the second stage, the resulting health index is modeled over time with a linear mixed-effects model (on the log scale for exponential models).

The exponential model uses a log transformation of x - phi, where phi ensures positivity and interpretability. The phi parameter can be supplied by the user or estimated automatically.

The resulting object stores both the projection (health index definition) and the fitted model used for RUL prediction.

References

Liu, K. and Huang, S. (2016). Integration of Data Fusion Methodology and Degradation Modeling Process to Improve Prognostics. IEEE Transactions on Automation Science and Engineering, 13(1), 344--354.tools:::Rd_expr_doi("10.1109/TASE.2014.2349733")

Examples

library(degradr)
library(dplyr)
# Load example data
data(train_FD001)
data(test_FD001)
data <- train_FD001 %>%
  select(unit,t,T24,T50,P30,
         Nf,Ps30,phi, NRf,
         BPR,htBleed,
         W31, W32) %>%
  mutate(across(c(P30,phi,W31,W32), ~ . * -1))

test <- test_FD001 %>%
  select(unit,t,T24,T50,P30,
         Nf,Ps30,phi, NRf,
         BPR,htBleed,
         W31, W32) %>%
  mutate(across(c(P30,phi,W31,W32), ~ . * -1))

# Fit a health index model (exponential trajectory of degree 2)
model <- fit_healthindex(data = data, type = "exponential",
                         degree = 2, r = 0.8)
rul <- predict_rul(data = test, model = model)
head(rul)