Soft-Sensing Of Aviation Hydraulic Oil Contamination Level Based On Neural Networks

In modern industry,70%-80%faults of the hydraulic system arecaused by the contamination. For the aircraft hydraulic system, damagescaused by hydraulic oil contamination are most severe, even threat to thesafety of aviation. The contaminations of airplanes’ hydraulic system aremainly result from particles produced by hydraulic system’s owncirculation, so airplanes no matter being assembled or on service shouldbe measured of their hydraulic oil contamination frequently. Existingmeasure methods including weight measuring, microscopic particlecounting method etc, mainly need to be carried out in laboratory whichreduces the efficiency, while the rest of them, timely online methods,need complex technology that is not suitable for the daily production.This paper is to solve the issue that aviation hydraulic oil cannot betimely and accurate measured, and it raises a method of soft-sensing ofaviation hydraulic oil contamination level based on neural networks. First,it analyzes the importance of aviation hydraulic oil contamination levelmeasurement in the practical production and the value of soft-sensingtechnique in this issue. Secondly, it discusses the hydraulic oil contamination of airplane and aviation hydraulic oil contamination leveldivision. According to the principle of soft-sensing technique andpurposing on hydraulic oil contamination on-line measurement, it deeplyanalyses of effects of various physical quantities combined with hydraulicoil contamination level, and chooses4independent and completelyobservable variables to be the soft-sensing’s secondary variables. Afterthe analysis of the soft-sensing technology, the author choosessoft-sensing method based on neural networks, and through theappropriate experiments gets related data under variable contaminationlevels. Thirdly, compared BP networks and RBF networks, this paperestablishes soft-sensing models using the experimental data by Matlab.Trains and tests the two different networks, and then results show that thesoft-sensing model based on RBF neural networks has good fitting abilityand prediction capability, achieves precise results. Finally, creates thecontamination level of aviation hydraulic oil measure platform, whichcontains soft-sensing function and can establish new models according toerror goal and secondary variables, using Matlab GUI. This platform iseasy to manipulate, suitable for the actual production, and can simplifythe work of measuring the aviation hydraulic oil contamination level.