Search

Based on a brief introduction to the basic theory of Volterra series, PSO algorithm is applied to obtain the three order time domain of volterra. Experiments are constructed based on the Jeffet cracked rotor simulation model. The results prove this model can identify the different crack depth rotor crack fault.

For gear fault classification problem, most of fault diagnosis methods only use the sensor output signal to diagnose. For this reason, they can not reflect the system as a whole transmission characteristic and so inevitably there are limitations. a fault diagnosis method based on Volterra kernel, which uses input and output signal in gear system, is presented. The method uses the first order, second order and third order time domain kernel, which is solved by recursive least squares algorithm, to determine the current fault status of the gear. The experimental results show that the method can judge accurately the gear fault from the characteristics of time domain kernel.

For estimating fuzzy system which is imprecise and represented by fuzzy sets, a regularized least squares support vector fuzzy regression model is proposed. The proposed fuzzy regression model is applying the fuzzy sets principle in weight vector and bias term. Determining the weight vector and the bias term of the proposed fuzzy regression model only requires a single matrix inversion, as against the solution of a complicated quadratic programming problem in previous support vector fuzzy regression model. Numerical examples are given to demonstrate the effectiveness of the proposed fuzzy regression model.

For estimating faults of the system with unknown nonlinear term, a novel fault diagnosis method based on nonlinear compensation term and proportional multiple-integral observer is proposed. In this method, the nonlinear compensation term is constructed by support vector machines(SVM), which can reduce the influence of unknown nonlinear part. Proportional multiple-integral observer based nonlinear compensation term can estimate actuator faults when there are sensor faults or output disturbance in the system. The proposed method is applied in civil aircraft system. Simulation experiments are given to demonstrate the efficiency.

For discrete time-delay systems with unknown nonlinear fault, a fault observer is proposed based on least square support vector machines (LS-SVM) and deconvolution method. The proposed fault observer uses a LS_SVM regression model to estimate the nonlinear fault. The training samples of the LS_SVM regression model are the control vector sequence which is acquired from the discrete time-delay systems by deconvolution method. Stability of the observer is proved by Lyapunov function. Simulation experiments demonstrate the effectiveness of the proposed fault observer.

For estimating the fault signals of the discrete flight control system with unknown senor faults and actuator faults simultaneously, a discrete proportional integral observer (PIO) is presented. The proposed PIO uses the augment states to estimate the sensor faults of the flight control system. Moreover, this observer uses an additionally introduced integral term of the output error to obtain the estimation of actuator faults. The convergence of the PIO is proved. The proposed fault observer is applied in flight control system. Simulation results are given to demonstrate the effectiveness of the proposed fault observer.

For nonlinear time series fault prediction online, an incremental learning least squares support machine (LSSVM) is presented to replace LS-SVM which is as a kind of regression method with good generalization ability and trained offline in batch way. The incremented learning LS-SVM fully utilizes historical training results and reduces memory and computation time, which guarantee to predict time series online. Two simulations results show that the incremental learning LSSVM has good performance for predicting nonlinear series fault prediction online.

Land degradation is a major problem world-wide. Soil degradation is the core of land degradation. Soil degradation at regional scale is related to susceptibility to erosion, soil suitability, and soil characteristics. There is a pressing need for an objective measure on land degradation at regional scale. The study area is in Hengshan county in ShanXi province, where is in the agriculture-pasture mixed area in Northern China with complex physical geographical situation. On the basis of discussing the spectrum feature of soils and soil line parameters and analyzing the spectrum feature parameters responding to land degradation, the study is to develop and evaluate a spectral approach for soil features for degraded soil using Hyperion by exploiting the potential of its red bands and near infrared bands. This index called Degraded Soil Line Index (DSLI) is based on SAM classification approach and the soil line concept. The potential of Hyperion image data for mapping soil degradation through the application of this spectral index and the spectral angle mapping (SAM) approaches in semi-arid area of North China will also be compared and investigated. It was indicated that results showed that both the SAM and DSLI approaches have the ability to map soil degradation and clearly show the degraded soil class. It also showed the difference between the DSLI and SAM methods is significant, and showed the interest and the contribution of the DSLI index in the study of land degradation. The validation of the results obtained with the DSLI index and soil samples served as an additional tool showed that that the map of distribution of land degradation types using DSLI method depends on the physico-chemical characteristics of the different classes. It indicated that the highly degraded soils are associated with low soil organic matter and available phosphorus, high slope gradient and sand particle percentage and total phosphorus.

Tikhonov regularized SVM is a kind of new SVM which can convert multi-class problems to be single optimized problems. Since SVM has some limitations in disposition of big data collection, this paper puts forward a new reduction Tikhonov regularized SVM by utilizing pruning algorithm to gain reduction data collection. Meanwhile, the paper applies genetic algorithm to make automatic selection from the balance parameter and kernel function parameter of Tikhonov regularized SVM. The experiment proves this newly improved Tikhonov Regularized SVM is more advantageous for classifying precision and train rate.

To select parameters is important in the research area of support vector machine. Based on particle swarm optimization, this paper proposes automatic parameters selection for least squares support vector machine (LSSVM). The effect of this proposed method is demonstrated by function regression problem. Besides, an equipment fault classification further illustrates that LSSVM based on particle swarm optimization has better classification ability than LSSVM based genetic algorithm under the same condition.