Search

Multivariate statistical analysis is often used in fault diagnosis. PCA is a commonly used multivariate statistical analysis method, based on the threshold method for fault diagnosis. But the PCA is the original space of the equivalent that didn't increase any amount of information, often resulting in a lot of minor failures difficult to diagnose. Deep Learning has a successful application in pattern recognition, linear representation of detail based on multi-level network. But Deep Learning is not explainable, only to give training results and there is no theoretical basis. On the basis of the original PCA, firstly expands and then reduces the dimension. The former is intend to fully express the information which cannot be expressed in original space by adding a projection, two projection,..., on the basis of original projection, and so on. And then generate a multi-level projection frame representing the space from the sample space high-dimensional feature. Experiment shows that the minor faults that cannot be detected in the original projection space can be detected under the projection frame of the new multi-level high-dimensional feature, and the detection rate of fault detection is improved.

In this paper, from the problem of the orientation of sudden pollution source which continuously release pollutants in three-dimensional space, while the orientation of the pollution source in the real environment is random and complex under the influence of the internal movement factors of the lake reservoir fluid, the problems are discussed from two aspects. Firstly, the nonlinear least squares estimation is carried out based on the pollution source diffusion model. Then, the sequential idea is combined with the sequential unscented Kalman filter, and the pollution source is precisely orientated by using the sequential unscented Kalman filter. Secondly, under the condition that the pollutant diffusion model is unknown, the diffusion model of pollutants in the unknown diffusion model is trained by ELM to solve the problem of the orientation of sudden pollution sources. The simulation results show that the ELM trains a pollutant concentration diffusion equation, which can effectively avoid the influence of internal movement, wind or other factors, and more in line with the actual environment. After training the concentration diffusion model, it is possible to reduce the complexity of numerical calculation, improve the performance of orientation and improve the robustness of orientating based on the application of the sequential unscented Kalman filter.

A new kind of roller thrust-torsion coupling was designed in this paper, with low friction and simple structure, linear movement of proportional solenoid can be changed into rotation of the spool movement, at the same time, proportional solenoid's thrust is amplified. With large flow and simple structure, 2D Electro-hydraulic Proportional Directional Valve (EPDV) which is position feedback mainly consists of roller thrust-torsion coupling and proportional solenoid. Based on the principle of roller thrust-torsion coupling, 2D EPDV had been studied through its mathematical model which was simulated with software MATLAB. Direct-acting and pilot-operated experiment results show: with good input and output characteristics, high pressure and large flow control can be achieved by this valve, no-load hysteresis is 1%, flow hysteresis is under 5% when system pressure is at 21 MPa, its linearity and repeating characteristic are all good.