Research On Method Of Machine Learning Based Prognostics And Health Management

In some important occasions,the failure of equipment will cause huge economic losses.However,manual maintenance costs much more with the increasing integration.Nowadays,Prognosties and Health Management(PHM)plays a important role in reduce the cost of maintenace.There are many examples of using PHM technology,but for digital circuit,it is still in the exploratory stage.As digital circuit is everywhere in our life,the research on PHM of digital circuit is of great value.By summarizing the existing application PHM methods in China and abroad,this paper makes a pioneering study on fault Prognosties and Health Management scheme for digital circuits.In this paper,74LS280 is used as an example to analyze the characteristics of the chip and its related performance indicators.The data of the degraded performance are extracted from the aging test data,and the method of predicting the fault is put forward.This method has the advantages of easy realization and strong usability.In order to complete the experimental content of this paper,this paper designs a platform for testing the characteristic parameters of 74LS280.The platform design includes a hardware platform,which includes sensor selections and environment simulation,and a software platform for reading test data and transferring it to a PC.Based on the support vector machine(SVM)method,74LS280 test data was used to establish a fault model.Other chips can use this model for failure diagnose.As an extension,two data preprocessing methods are presented to optimize the training model in training speed or accuracy.After the model is established,this paper proposes to predict the remaining useful life(RUL)of 74LS280 by using the least squares method,and propose the original data rotation alternative method to improve the RUL prediction value.Although the RUL prediction value of this paper can not guarantee the accuracy of the numerical value,it is of certain practicality to alert the maintenance staff to the incoming fault.