Research On Multi-population Collaborative Genetic Algorithm For Aeroengine Diagnosis And Parameter Identification

With the rapid development of science and engineering technology,the complexity of systems and equipment is increasing,and the analysis of equipment failure mechanisms and failure modes is becoming more and more complex.It is therefore quite necessary to strengthen the development of intelligent fault diagnosis technology,which could automatically find out significant features for further diagnoses through analysing the collected faults,acquire the features,and thus achieve intelligent diagnosis.The independent research and development of aero-engines in China has also put forward intelligent requirements for modeling,parameter identification,and health management technologies.In this paper,artificial intelligence methods such as computational intelligence and machine learning are applied to the field of aero-engine.In combination with the practical background and requirements of this field,the research is focused on the multi-population genetic algorithm and its application in the field of aero-engine,including the improvement of genetic algorithm,the migration strategy of multi-population genetic algorithm,the mining of feature patterns and the adjustment of adaptive parameter domain,etc.The main contents are as follows:(1)A multiple differential genetic algorithm with multi-criteria evaluation for feature selection is proposed.Firstly,the differential genetic algorithm is suggested.Based on the simple genetic algorithm,the differential genetic algorithm is improved by designing the genetic operators in combination with Relief F algorithm,inter-class distance and intra-class distance,frequent combination and differential strategy,which overcomes the limitation of the premature of simple genetic algorithm and facilitates the selection of individuals with high fitness and important features.Secondly,the multi-population differential genetic algorithm is applied to the feature selection,in which each population shares the current best individual after the evolution with short-iterations,uses it in searching for the effective feature pattern and thus achieves the spread of good and reliable models obtained from multi-population.Finally,the algorithm is verified by simulation experiments and UCI data sets.In addition,the effectiveness and advantages of the algorithm proposed is also compared with those of other algorithms in terms of convergence speed,population mean value,reliability and stability.(2)An adaptive adjustment parameter domain with multiple genetic algorithms is proposed.Firstly,the method of combining differential variation and uniform variation is adopted to improve the mutation operator of simple genetic algorithm,which realizes the exploration outside the interval and the development within the interval and,which could achieve both generalization and refinement.Secondly,multi-population collaboration is introduced and the interval containing high-quality solutions is excavated through regular communication.At the same time,the adaptive adjustment parameter domain method is designed to realize the dynamic adjustment.As the parameter value range is constantly changing,the efficiency of the algorithm will be greatly improved.The improvement of the mutation operator and the combination of multi-population can help solve the problem that when the correct value of the parameter is not included in the given initial interval,the correct value cannot be found,and the accuracy requirement for the initial value range of the parameter is reduced.Finally,the performance of the algorithm is verified by eight standard functions(high-dimensional unimodal functions and multi-peak functions).By observing the adaptive adjustment process of the parameter interval during the algorithm operation,the correctness of the parameter domain adaptive adjustment method is verified.Compared with other algorithms in the aspects of optimal solution,mean value,standard deviation and convergence speed,the superiority of the algorithm is verified.(3)The multi-dimensional feature selection of aero-engine rolling bearing fault signal is studied.A trouble with the eroengine fault diagnosis is that it has to deal with a large number of fault features,whose redundancy and mutual exclusion have a great impact on the accuracy of fault diagnosis.Therefore,how to extract and select the important "pure" feature subset is of great significance for improving the accuracy and efficiency of fault diagnosis.In this paper,multiple differential genetic algorithm with multi-criteria evaluation is applied to the fault feature selection of aero-engine rolling bearings.In addition to mining effective feature patterns from the phase results of each population,the algorithm also deals with the strong correlations in the feature patterns,so that more ―pure‖ feature patterns are passed to the offspring and guide the descendants to evolve for a higher efficiency and accuracy until a subset of features with high classification accuracy is found,and the difference between features in the feature subset is significant.Finally,the eroengine rolling bearing fault simulation experiments are conducted to obtain the fault sample set,which prove the effectiveness of the algorithm.(4)Aeroengine dynamics parameter identification is the basis for establishing accurate dynamic models.The accuracy of the parameter identification has a great effect on the accuracy of model calculation.However,in practical engineering applications,the accurate parameter range is not easy to define.In many cases,it could only be constructed out of experience.In order to reduce the impact of initial parameter interval accuracy on the identification results,this paper applies the adaptive adjustment parameter domain with multiple genetic algorithms to the aeroengine dynamic model parameter identification practice.Taking into consideration the initial interval uncertainty of the optimization parameters,the vibration modal parameter identification and the connection stiffness identification of the rotor system are studied.The identification results of the proposed method are verified and analyzed with experimental and simulation data,and its application in engineering is illustrated.