Hidden Semi-Markov Models for Predictive Maintenance

Figure 3

Akaike Information Criterion (AIC) applied to continuous and discrete observations data. AIC is effective for automatic model selection, since its minimum value provides the same number of states and duration model used to generate the data.

(a) AIC values for continuous data and Gaussian duration distribution

(b) AIC values for continuous data and Gamma duration distribution

(c) AIC values for continuous data and Weibull duration distribution

(d) AIC values for discrete data and Gaussian duration distribution

(e) AIC values for discrete data and Gamma duration distribution

(f) AIC values for discrete data and Weibull duration distribution