The invention discloses an industrial multi-loop oscillation detection method based on multidimensional essential time scale decomposition. The method comprises steps that in to-be-detected control loops, process output signals of all the to-be-detected control loops are acquired; the output signals are projected along a series of projection directions to acquire a set of signals; multidimensional signal extreme points are solved, and multidimensional baseline node extraction of the extreme points is carried out; a multidimensional linear transformation formula is utilized to acquire linearity extraction results; the linearity extraction results are integrated to acquire a mean value estimate and decomposition subsignals of the output signals; stop conditions are determined, if the stop conditions are satisfied, decomposition stops; zero crossing point regularity indexes of the decomposition subsignals are calculated, oscillation degrees of the subsignals and whether the industrial process has plant-level oscillation are determined according to the indexes. The method is advantaged in that quantitative detection on multi-loop oscillation behaviors in the industrial process can be carried out, the rule degree and the period of each oscillation component can be further acquired, and abundant data supports are provided for oscillation behavior evaluation and fault source diagnosis.