Dynamic Time Warping

A distance matrix compares each sample in one sequence with samples in the other sequence. Dynamic programming then evaluates possible routes through that matrix and selects a path with the lowest accumulated cost. The resulting path pairs features that may occur at different positions, allowing similarity to reflect aligned signal behavior rather than requiring synchronized timing.

Continuity and boundary constraints limit which movements through the distance matrix are allowed. They ensure that the selected alignment follows a structured path from one sequence boundary to the other instead of choosing unrelated sample pairings. These restrictions make the minimum-cost result represent a coherent correspondence between the two complete sequences.

Local expansion and compression let one portion of a sequence correspond to a shorter or longer portion of another. This accommodates features that occur at different rates while preserving their broader order within the alignment. The capability is important when a recurring signal pattern remains recognizable even though its timing changes from one observation to the next.

First, samples from the two sequences are compared to construct a distance matrix. A dynamic-programming procedure then searches the matrix for the minimum-cost path while obeying continuity and boundary constraints. The resulting path provides the feature alignment and a cost-based measure of similarity, which can support subsequent recognition, monitoring, or signal-comparison tasks.

In machinery-condition monitoring, the method can help identify recurring signal patterns even when operating rates vary. By aligning observations whose timing differs, engineers can compare behavior without requiring every event to occur at the same speed. This supports analysis of sensor signals and helps monitoring systems recognize patterns associated with changing machine operation.

Engineering applications include speech and audio recognition, sensor-data alignment, and gesture or motion analysis. In each case, meaningful features may appear at different times or over different durations. Dynamic Time Warping helps align those features so systems can evaluate recurring patterns despite timing variation, extending sequence comparison beyond strictly synchronized measurements.