Process Monitoring Based On Extreme Learning Machine

With the rapid development of science and technology,especially the progress of automation and computer,modern manufacturing process become more and more complex,large-scale,automatic,which makes it difficult to guarantee the production process' s safety and reliability.In addition,the manufacturing process is accompanied by very strict production conditions and environment,such as high temperature,high pressure,inflammable and explosive,so safe and reliable operation is crucial,and is key factors to ensure the quality of products.Therefore,in order to improve product quality and ensure that the process operation conditions satisfy the given performance index,it is necessary to detect,diagnosis and elimination abnormal conditions or fault,so online monitoring,fault detection and quality control become an urgent and necessary research subject to the scientific researcher.Flexible Printed Circuit(FPC),as one of the most important electronic interconnection technology,has been widely used in various electronic products because of its superior performance.However,as a high-precision product,the manufacturing process of FPC is pretty complex and precise,it becomes more and more important to ensure safe and reliable manufacturing process of FPC.The manufacturing process of FPC includes hundreds of processes,among which the etching process is one of the key processes.In this paper,the process monitoring methods and application of extreme learning machine are researched based on the etching process.The research works are shown as following:(1)A new online extreme learning machine algorithm with varying weights and decision level fusion has been proposed,which increases the weights of the new samples predicted wrongly by current data monitoring model in learning,and introduced decision level fusion to improve integrated decision-making ability of the model.Performance comparisons of the method are presented using UCI datasets and Tennessee Eastman process.The results show that the proposed algorithm produces comparable or better performance with higher accuracies and lower training time.(2)An online regularized sequential extreme learning machine with ensemble strategies including kernel strategy,weight mechanism and differential evolution optimized parameters(WOS-DE-RKELM)is proposed.Regularization,weight mechanism and kernel strategy have been introduced into an online sequential extreme learning machine(OS-ELM)separately to improve algorithm performance.In this paper,to take advantage of different strategies,regularization,weight strategy and reduced kernel projection strategy are combined in one OS-ELM to make the algorithm more outperformance.However,the ensemble strategies have side effects as several parameters which directly affect algorithm performance are introduced at the same time;therefore,to wake the effects,an effective optimization tool,differential evolution is adopted to tune these parameters.Experiments are carried out on benchmark regression datasets from the UCI repository and time-varying datasets.The results show that the proposed algorithm is more efficient than the popular OS-ELMs in term of prediction accuracy and robustness.(3)The thesis analyses the characteristics of the technology in the manufacturing process of high density FPC and the etching process of manufacturing process of FPC,while summarizes the possible types,source and faults,then simulation data of etching process generated accordingly.The proposed OS-ELMs are employed to manufacture etching process and recognize fault.