Hydraulic System Based On Multi-sensor Information Fusion Fault Diagnosis Method Is Studied

As the core component of engineering vehicles, the operationalstatus of the hydraulic system affects the work of the entire vehicledirectly, it has important practical value to monitor、predict and diagnosethe faults timely and accurately. The working environment of hydraulicsystem is harsh, the failure mechanism is complex and diverse, and thefault characteristic parameters obtained by a single sensor are often vagueand uncertain. In order to achieve the comprehensive and accurateequipment diagnosis, it needs to make full use of the fault informationfrom multi-source synthetically.The paper focuses on researching the fault diagnosis of hydraulicpump which is the power component of hydraulic system. Throughconstructing an effective sensor network and take advantage of thecharacteristic information of the three vibration signal of pump housingand the temperature signal of leakage outlet. According to JDL model andcombined with the failure characteristics of the truck hydraulic system,the paper designs a three-stage multi-source information fusion diagnosissystem；The data level is mainly to carry out the feature extraction andnormalization for each fault data, the feature level conducts the subsetlocal diagnosis by building four parallel MPSO-BP neural network, thedecision-making level fuses the local diagnosis from each MPSO-BPneural network based on the improved D-S evidence theory；Finally, itrealizes the fault diagnosis.In the feature level of the system, according to the problems of thelow convergence precision、easy to fall into local minima and the slowconvergence speed in later period of the PSO algorithm, a PSO improvedalgorithm based on the sine changing inertia weight and adaptivemutation strategy is proposed in this paper, the convergence precision andconvergence speed of the algorithm have been improved and thepremature convergence is avoided effectively; Moreover, as the gradient down learning methods of BP neural network is easy to fall into localminimum value in the network learning and training process,the paperutilezes the strong global optimization ability of PSO algorithm tooptimize its weight and threshold matrix, an improved PSO-BP neuralnetwork diagnostic algorithm is put forward, the numerical exampleverifies the effectiveness of the method.In the decision-making level, in order to solve the problem that D-Sevidence theory can’t combine the high conflicting evidence, in this paper,taking the evidence sources themselves into consideration and accordingto the decision idea that the minority submit to the majority, a newimproved D-S algorithm based on the weight of evidence is proposed.The numerical example shows that the method can deal with the highconflicting evidences efficiently, the convergence speed is faster and thediagnosis results are more desirable than other improved algorithms.