Data Driven Optimization Control And Its Research And Application In The Monitoring Of Flexible Production Process Of Auto Parts

In order to realize the flexible production of automobile parts,a large number of digital equipment are used in the production process,a large number of equipment data and process data are generated,which contain a large number of production information.Using this information,enterprises can avoid abnormal events caused by various uncertain factors in the production process,improve the utilization rate of equipment and optimize the production process.Therefore,the production process monitoring method based on data-driven optimal control has attracted wide attention and has important industrial value.Traditional production process monitoring methods often assume that process variables obey a single linear relationship or a simple non-linear relationship,and consider that process modeling data are large and regular.With the increasing flexibility of production process,more and more kinds of data are produced by digital equipment,and the requirements of the types,complexity,accuracy and efficiency of the processed automobile parts are becoming higher and higher,which leads to the more complex process modeling characteristics,mainly manifested in the difficulties of multi-source heterogeneous data fusion,strong non-linear relationship of variables,high data dimension,small number of samples and other multiple characteristics.Therefore,more stringent requirements are put forward for the modeling,optimal control and fault prediction of digital production process and equipment monitoring.Data-driven optimization control method based on deep neural network can abstract features in depth by using discrete sample data and powerful non-linear factors in the network,realize the modeling of digital production process and equipment monitoring,and accurately classify and predict features,so as to better solve the problems of modeling,optimal control and prediction.However,in the process of non-linear modeling based on deep neural network,on the one hand,because of the high dimension of network layers,weight values and the number of neurons in hidden layer,it is easy to cause "dimension disaster".On the other hand,in order to realize the self-adaptive optimization design of network structure,it is necessary to study the multidimensional and multi-objective optimization for the hyper parametric optimization of network model.The algorithm is solved.According to the idea of data fusion-virtual-real modeling-simulation prediction-intelligentcontrol,this paper focuses on the research of digital twin data fusion and modeling method,the modeling of non-linear system based on evolutionary depth confidence neural network,multi-objective optimization and decision-making,and the prediction of reverse clearance error value of machining center.Questions mainly include the following contents:1.Aiming at the problem that it is difficult to fuse multi-source heterogeneous data in the operation of digital equipment,the method of digital twin data fusion and modeling is studied.By constructing a five-dimensional digital twin model,the adaptive matching protocol analysis method is studied to solve the problem of heterogeneous network transmission,the sliding window and Euclidean distance are studied to solve the problem of uncertain redundant data,and the multi-granularity heterogeneous data fusion method of probability transmission is studied to solve the problem of temporal and spatial correlation of sensor data..2.Aiming at the problem of "dimension disaster" in the process of nonlinear system modeling by deep neural network,the design of high-dimensional multi-objective optimization algorithm is studied.Firstly,the framework of complex biogeography optimization algorithm based on decomposition is designed.The high-dimensional optimization objectives are decomposed into several subsystems by using uniform distribution weight vector and mean aggregation method,and then introduced into metropolis twice Criterions migrate within and across subsystems,and use PBI distance to calculate the neighborhood Island distance to balance the convergence and diversity of solutions,so as to obtain the optimal Pareto solution set and improve the efficiency of the algorithm;.Then,the parameter sensitivity analysis of MHDB and PBI distance parameter theta in the algorithm is carried out.The results show that the algorithm has good reliability.Finally,compared with the mainstream multi-objective optimization algorithms in recent years,such as NSGAIII,MOEAD/PBI and BBO/Complex,it shows that the algorithm can effectively avoid the "dimension disaster".3.In order to solve the problem that it is difficult to automatically design the optimal architecture of neural network in the process of nonlinear system modeling by deep neural network,the method of combining deep belief neural network with evolutionary algorithm is studied.Firstly,the number of hidden layers,connection weight and activation function areinitialized by unsupervised method.Then,the hyper-parameters are coded,and the fitness function is designed by minimizing the reconstruction error of the network as the optimization object.The better parameters are selected according to the fitness value to generate a new population,and then the crossover is carried out.And mutation operation.Finally,the model structure of deep confidence neural network is automatically optimized to find the optimal number of hidden layers and hidden layer nodes,reduce the demand for computing resources,and solve the problem of automatic design of network structure under a large number of marked data,so as to carry out adaptive optimization control for the modeling process.4.Aiming at the problem of reverse clearance error prediction during machining center operation,a hierarchical predictive maintenance model of reverse clearance error is established,and the reverse clearance problem in the operation of machine tool is modeled.The prediction of reverse clearance error is verified by experiments.5.Based on the above research results,a flexible production line monitoring system model of an automobile parts factory is designed and developed,and the model is optimized.The core component of the model and algorithm program is integrated into the.NET framework by using the MATLAB program,and it becomes the core component of the intelligent production line dynamic monitoring management system.Practice shows that the system can enable factory managers to fully grasp the production site information in time,improve the efficiency of equipment operation,enhance the overall production capacity and improve the level of production management,and bring good economic benefits to enterprises.