Study On Improvement Of Shuffled Frog Leaping Algorithm And Its Applications In Rotating Machinery Fault Diagnosis

The machinery fault diagnosis is a comprehensive science to identify the running state machines or unit. Its core is effectively obtaining, transmissing, processing, regenerating and utilizing the diagnostic information. Accordingly it has the capacity of accurate condition identification and diagnosis decision-making under the set environment, which has great significance to ensure the safe operation of the equipment. At present, the working condition of the mechanical equipment is more and more complex, the structure and function is towards the direction of large-scale, integration and automation. How to extract useful information from these machine, and determine the operation of equipment by the state and timely diagnosis to the failure, this becomes a new trial to the existing fault diagnosis methods.The shuffled frog leaping algorithm (SFLA), the new swarm intelligence optimization theory, and its improvements are applied to the process of machinery fault diagnosis. From the perspective of intelligent optimization, the solution to the sensor networks placement, the parameters optimization of BP neural networks model, and the cost function and the classification number of unsupervised machine learning in the diagnosis system are accomplished. The main work in the dissertation can be summarized as follows:(1) On the basis of analysis to the related concepts and mathematical model of SFLA theory, the frog individual update model is simplified and its global convergence theoretically is proved through analyzing the dynamic behavior of the worst frogs based on the z transform; Combined with the concept of Markov model and the desired convergence time of the algorithm, the theoretical convergence rate and the complexity of SFLA itself are completed, which is to perfect the theory of SFLA; The single factor variance analysis of statistic theory is applied to analyze the effect of 5 basic parameters and the related performance of the algorithm. Then, based on experimental data and analysis results, the basic guiding principle of the parameter setting about SFLA is first given.(2) In view of 0-1 variable optimization problem, a new improved discrete SFLA (1D-SFLA) based on the crossover and mutation operators of GA is proposed. When the worst frog is updating, the average optimal frog is obtained by crossover operation. The probability of a mutation operation on the worst frog depends on the Haming distance between it and the average frog. Numerical simulation results show that the ID-SFLA has an excellent global search ability and outstanding convergence performance; Then a mathematical model is established to illustrate the sensor network optimization based on fault-sensor dependence matrix. The ID-SFLA is applied to the sensor’s optimal selection about location and number for a gearbox. In comparison with GA and discrete shuffled frog leaping algorithm (DSFLA), the new ID-SFLA not only poses an effective solving method, but also provides a new thought for the solution of related 0-1 integer NP-hard problem.(3) The new improved SFLA is developed on the basis of a chaotic operator and the convergence factor of particle swarm optimization to overcome the shortcomings of conventional SFLA. The frog population initialization can be completed and the frog individual can quickly jump out of the worst position utilizing the chaotic sequence. And through using the convergence factor, the worst frog can update only by one step, rather than the previous two steps. Testing results show that the proposed algorithm can effectively improve the solution accuracy and convergence properties, exhibits an excellent ability of global optimization in high-dimensional space. Then the presented ISFLA is employed to optimize the weights and threshold values of BP neural network for the early fault diagnosis of rolling bearings, which is usually difficult to distinguish. And the sample entropy of IMFs after EMD is used as the characteristic input of networks. Results indicate that the developed new method demonstrates better generalization capability and stronger robustness. And the recognition rate also greatly improved.(4) Individual update strategies of DE algorithm are firstly applied to the update process of the worst frogs in SFLA, and a new strategy of the multiple evolution ways existing and competing at the same time is developed into a new modified SFLA(MSFLA-Co). By comparing the principle of K-Means, the new adaptive clustering analysis method combined with the MSFLA-Co is put forward. And it uses the DB indicator as a measure of clustering effect. The method not only can quickly find the clustering center vector of massive data, but also can accurately determine the optimal clustering number of data sets; Through some testing data sets and the examples of mechanical fault diagnosis, the effectiveness of the algorithm is validated;(5) A novel modified SFLA with inertia weight is introduced. Firstly, its global convergence is proven in theory.To extend the scope of the direction and length of the worst frog (vector), three different control strategies of inertia weight are proposed and the testing results show that the modified algorithms can effectively improve the solution accuracy and convergence properties. Then inspired the idea of weighted fusion, MSFLA-VPMCD model, combining the MSFLA and VPMCD principle, is put forward and applied on pattern recognition of rolling bearings. At last, through the simulation experiment, the performance of three improved algorithm for continuous variables are tested and results show that each has advantages and disadvantages.