ICA (Independent Component Analysis) is a blind source separation technique that decomposes a multivariate signal into statistically independent components — unlike PCA (which finds uncorrelated components), ICA finds maximally independent sources, revealing the underlying independent physical causes.

How Does ICA Work?

• Model: $X = AS$ where $S$ are independent source signals and $A$ is the mixing matrix.
• Objective: Find the unmixing matrix $W = A^{-1}$ that maximizes the statistical independence of the estimated sources.
• Independence Criteria: Maximizing non-Gaussianity (kurtosis or negentropy) or minimizing mutual information.
• Algorithms: FastICA, Infomax, JADE.

Why It Matters

• Source Separation: Separates mixed signals into independent physical sources (e.g., separating fault signatures from normal variation).
• Beyond PCA: PCA gives uncorrelated components; ICA gives truly independent ones — better for identifying root causes.
• Fault Isolation: Each independent component may correspond to a separate physical mechanism.

ICA is finding independent causes in mixed data — separating overlapping signals to reveal the truly independent sources of variation.