| In actual engineering applications,considering the convenience of use and beautiful appearance of the product,industrial products will try to reduce non-essential parts.Relatively speaking,the information monitoring and feedback link will be relatively weak,and only some necessary sensors will be left.Therefore,for such industrial products,it is difficult to obtain accurate and sufficient sampling data only by relying on the sensors carried by itself,and often only sparse or even non-uniform sampling data can be obtained.If we want to further develop the equipment,the first problem is that a more accurate mathematical model is needed to provide a basis for research.Considering this relatively extensive engineering phenomenon,it is of great practical significance to achieve more accurate modeling for such a real system that can only obtain non-uniform sparse sampling data.In response to this practical engineering problem,this article makes the following innovations.Firstly,a data reconstruction method for non-uniform sparsely sampled data is proposed.A detailed analysis of the characteristics and difficulties of the current research objects,highlighting the problems that cannot be solved by the existing methodology.Taking the summary basic content as the inspiration point,considering the physical characteristics of the research object and experimental design from multiple perspectives,the preprocessing process and two data reconstruction methods are proposed.Then proceed to the experimental design,introduce the experimental device and the choice of excitation signal,and conduct a large number of experiments.At the same time,based on the original sampling data,we further analyze the sampling mechanism of a kind of non-uniform sparse sampling of rotating machinery.Finally,the data reconstruction method is used to process the non-uniform sparse sampling data to obtain multiple reconstructed data sets.Secondly,data-driven modeling based on reconstructed data sets is realized.Due to the problem of data acquisition in non-uniform sparse sampling systems,it is difficult to directly carry out data-driven modeling.Therefore,under the precondition of the data reconstruction link,based on the physical characteristics of the object to be modeled,we realize a dynamic modeling method based on the gradient descent method.During the experiment,taking into account the operational characteristics and data sampling characteristics of the research object,different reconstructed data sets are used as the control group.Finally,a number of model evaluation indicators are proposed to objectively evaluate the results of multiple sets of experiments,so as to analyze and compare the obtained training models from multiple perspectives.Thirdly,the evaluation and analysis of multiple training models have been completed.Aiming at the problem of inaccuracy and imperfection in evaluating multiple training models using only raw measurement data,we carry out an open-loop cross-validation test experiment design based on multiple training models.Subsequently,a large amount of open-loop testing work is performed on multiple sets of training models.Based on reliable model evaluation indicators,the test results of all training models are analyzed,and individual models are used as examples to describe them with graphics.In addition,this article analyzes the characteristics of the model from the perspective of control,including analysis of controllability,observability and stability.On this basis,the application is carried out,and several controllers are designed for the system,and they are analyzed and compared one by one. |
