A clustering algorithm based on principal components analysis (PCA) is proposed for clustering autocorrelated and cyclic data sets typical of continuous chemical processes. A moving window approach is used to adjust the temporal properties of the solution; different parametrizations of the moving window can be used to isolate the high-and low-frequency content of the time series measurements. A framework is proposed to combine separate cluster analyses performed at different time scales to identify all process states and accurately detect the transition points between the states in the face of a periodic disturbance affecting a process. The method is tested on experimental data from a continuously operated pilot-scale process, and is shown to be superior to traditional k-means clustering. PCA variable contribution analysis is used to diagnose the nature of the various operating regimes and faults identified by the cluster analysis.