Abstract:Independent component analysis (ICA) is widely used for modeling and monitoring non-Gaussian process. However, traditional ICA lacks probabilistic representation of process uncertainties. In this study, a probabilistic ICA (PICA) model is proposed for non-Gaussian process modeling and monitoring. The independent latent spaces are specified with Student's formulation to account for both Gaussian and non-Gaussian data characteristics while the additional noise term is further served as a complement for explaining underlying process uncertainties. The Student's distribution with adjustable tails is essentially an infinite mixture of Gaussians with various scaling variances. In order to monitor retained variations, the noise space is further extracted and analyzed with probabilistic principal component analysis (PPCA). Simulation results show that compared with the deterministic ICA-based method, the proposed two-stage probabilistic extraction method is more effective for monitoring non-Gaussian industrial processes.Note to Practitioners-Multivariate modeling of non-Gaussian industrial process is of great importance. Traditional ICA extracts high-order statistics from process data and is very desirable for modeling non-Gaussian process. Based on that, PICA gives more insightful interpretations on industrial data. Apart from the probabilistic representation of noise, the Student's denotation of non-Gaussian latent space serves as a natural and elegant probabilistic implementation for describing non-Gaussian behavior. Moreover, an integration of PPCA for monitoring Gaussian residual information after non-Gaussian extraction can further capture the residual variances. For practical implementation, it should be necessary to: (i) collect sufficient amount of data from industrial systems; (ii) preprocess industrial data well and eliminate irregular entries; and (iii) select proper IC/PC numbers based on training samples.

Calibration diagnostics for point process models via the probability integral transform

Score-based calibration testing for multivariate forecast distributions

A Probabilistic Model with Spike-and-slab Regularization for Inferential Fault Detection and Isolation of Industrial Processes

On Calibrating Diffusion Probabilistic Models

Comparison of Point Process Learning and its special case Takacs-Fiksel estimation

Calibrate: Interactive Analysis of Probabilistic Model Output

Independent component models for replicated point processes

Non-Gaussian Industrial Process Monitoring with Probabilistic Independent Component Analysis

On goodness-of-fit testing for self-exciting point processes

LOO-PIT: A sensitive posterior test

New validation metrics for models with multiple correlated responses

PCA for Point Processes

A Generalized Bayesian Approach to Model Calibration

Calibration through the Lens of Interpretability

Assessing the calibration of multivariate probabilistic forecasts

Deterministic and statistical calibration of constitutive models from full-field data with parametric physics-informed neural networks

Non‐parametric inference on calibration of predicted risks

Probabilistic recalibration of forecasts

Fixed Inducing Points Online Bayesian Calibration for Computer Models with an Application to a Scale-Resolving CFD Simulation

Modelling for Poisson process intensities over irregular spatial domains

A cross-validation-based statistical theory for point processes