Online Optimization Of Burn-in Strategy And Residual Life Prediction Based On Degradation Data

With the intensification of market competition,reliability has become an important product performance indicator,which thus attracts much attention of users and the market.Reliability analysis aims at improving the reliability of products and ensuring their ability to accomplish specified tasks by eliminating defective units and preventing unit failures,thus having important engineering application value.As two hot issues in the field of reliability,burn-in test and failure prediction are important tools to ensure the reliability of products and systems.With the development of manufacturing,the reliability of products has been greatly improved.In this case,the early fault-based burn-in test is particularly inefficient.Thanks to the development of modern measurement technology and sensor technology,burn-in test and fault prediction have made great progress in the research,but still face many challenges.To solve the burn-in test for highly reliable products and failure prediction under multi-sensor data,based on degradation data,the optimal burnin strategy and its online optimization and residual life prediction are studied.The main research work of this thesis is as follows.(1)Firstly,investigate the research status of burn-in test and residual life prediction,describe the principles and technical characteristics of existing methods,summarize the advantages and disadvantages of these methods,and thus lay the foundation for the subsequent work.(2)Research on the generation algorithm of burn-in strategy for reliable products and complete the online optimization of burn-in time.The difficulty lies in how to obtain the reasonable burn-in time based on the given degradation sequences,and adjust the burn-in time according to the actual degradation situation during the burn-in test and complete the screening task.In the design of the algorithm,a new burn-in framework is proposed,which combines a sliding window strategy with one-dimensional convolutional neural network,completes the off-line training for classification model,and then obtains the optimal burn-in time under a group-accuracy strategy.And an online optimization algorithm is constructed based on the definition of effective information.(3)Research on the residual life prediction under multi-sensor data.An available solution for the prognosis with multi-sensor data should not only provide a continuous visualization progression of system degradation,but also ensure that the generated fusion signal effectively performs in RUL prediction.Moreover,human exploration of degradation failure mechanism and the priori assumption of model parameter distributions should not be required.Therefore,in this thesis,a residual life prediction framework based on deep learning is proposed,and the relevant application design is carried out.The framework compromise two deep learning models,which are used for data fusion and residual life prediction respectively.The biggest feature of the framework is that in the off-line phase,the two sub models are trained by the proposed supervised joint training scheme.(4)Build simulation database and determine available public data sets,and carry out necessary data visualization analysis and preprocessing.Through experiments,the effectiveness of all the algorithms designed in this thesis has been verified.