Frequency spectral features of mechanical vibration and acoustic signals relate to difficult-to-measure product quality and quantity parameters of complex industrial process.Selective ensemble(SEN)modeling can selectively fuse these spectral features.Combination of several optimal ensemble sub-models with SEN cannot guarantee the best final model.In this paper,we use several techniques to achieve a global optimal model for SEN,i.e,dual-layer SEN(DLSEN).Genetic algorithm and kernel partial least squares are used to construct the inside layer SEN model based on each mechanical vibration and acoustic feature sub-set.Branch and bound and adaptive weighting fusion algorithms are integrated to select and combine outputs of the inside layer SEN models.Then,the outside layer SEN model is constructed.“Manipulating input features”-based and “sub-sampling training examples” –based ensembles construction methods are integrated together.The learning parameters of DLSEN model are optimized simultaneously.This novel approach is applied to a laboratory ball mill grinding process.The comparison results with the other methods show that the proposed DLSEN method is effective for modeling mechanical vibration and acoustic signals.