This work considers the problem of change detection of working regimes from industrial processes, e.g. electric machines with rotation elements, and which generates mechanical vibrations. Two approaches are considered: (i) based on signal processing and pattern recognition methods; (ii) based on sparse methods. The objective of the paper is to evaluate the preliminary results obtained by the above approaches and to promote methods based on sparse representations and computations for change detection problems, as alternative to classical methods based on transform or pattern recognition. The results are encouraging and suggest that more studies on the method of sparse computation as an optimal candidate for change detection from time detection point of view is needed.