Search

Building a model of flight accident scenario and quantitatively assessing flight risk are of great significance in improving system design and guiding aircraft's using. In view of the deficiency of traditional Probabilistic Risk Assessment (PRA), an evolution model of flight accident scenario is established based on Bow-Tie, and flight risk is evaluated based on Bayesian Networks. Taking engine shutdown in flight for an aircraft as an example, simulation results show that the method proposed can sort weak parts of flight safety system, and provide basis for system improvement. Comparing with traditional method, Bayesian Networks is more suitable for modeling of complex system and easy in calculation. So rationality of the method proposed above is validated.

Based on the highest density domain analysis, a new algorithm is proposed to perform prediction of the aviation safety in an uncertain framework. In order to perform the prediction of the aviation safety, highest density domain (HDR) is combined with uncertainty description technique to obtain the aviation safety interval and the corresponding interval probability level in the proposed method. Firstly, the kernel density estimator is further utilized with to produce the probability distribution. Secondly, highest density regions based on the estimated probability density function have been considered, to further investigate the underlying information. Then the aviation safety interval and the corresponding interval probability level are given to complete the aviation safety prediction. Compared with the methods of traditional aviation safety causal prediction and aviation safety time series prediction, this paper considers the uncertainty properties of aviation safety more thoroughly. Discussions with the flight accident hourly rate of the US Air Force from 1991 to 1999 have evidenced the applicability and adaptability of the proposed method.

Random walk plays an important role in heuristic algorithm. The traditional Cuckoo Search used only Levy random stemmed from Levy distribution. This paper extends two symmetric random walk methods, named Gauss random and Cauchy random, which are consistent with α-stable distribution and can be applied to the Cuckoo Search. The characteristics of three random methods are compared and analyzed, and the Cauchy random is revised. Based on the infrastructure framework of the Cuckoo Search algorithm, two new Cuckoo Search modes using new random walk methods are constructed. By comparing with the original Cuckoo Search in solving benchmark functions optimization and pressure vessel design problem, the conclusions can be drawn that Cauchy random has a great advantage for low dimensional problems and Gauss random is quite competitive for multidimensional unimodal problems, while Levy random is suitable for multidimensional and multimodal problems. And it has certainly guiding significance for the optimization design of algorithm and some optimization problems.

Use flue gas wet desulphurization technology to reduce the emissions of sulfur dioxide. The improved BP neural network model based on LM algorithm was established according to the factors that influence sulfur dioxide emissions, and the network weights and threshold value were adjusted repeatedly. The real data obtained from thermal power plants are simulated and verified. The results show that the improved BP network prediction model of LM algorithm has higher prediction accuracy of desulfurization efficiency, and has certain guiding significance for field operation.

The reliability analysis and reliability sensitivity analysis are two important works in modern engineering design. Because the reliability membership function presented previously is difficult to compute the reliability sensitivity, a new reliability index is proposed to cope with the reliability model with mixture of random and fuzzy variables. Firstly, the fuzzy variables at λ membership level are transformed into uniformly distributed variables within the λ-cuts, and the reliability model with mixed variables is transformed into one with random variables. Secondly, the reliability or the failure probability are computed by traditional reliability analysis methods, then the reliability index with mixed variables is obtained by use of the decomposition theorem of the fuzzy theory. Based on that, the reliability sensitivity analysis with respect to distribution parameter of random variable can be derived to evaluate the effect of distribution parameter on the new reliability index. The new reliability index not only provides an intuitive and compact index instead of the reliability membership function to describe the fault of reliability model, but also offers an approach to compute the reliability sensitivity with mixed variables. What's more, the point estimated method and the saddlepoint approximate method are suggested to solve the above indices because they have high precision without too much computational cost in computing the reliability and reliability sensitivity. At last, Two examples including a speed reducer and X-type compressor mechanism are used to testify the feasibility and reasonable of the proposed index. Their computational results verify that the presented methods in their corresponding application fields have high efficiency and acceptable accuracy.