Research On On-line Perception And Intelligent Early Warning Algorithm Of Gear Wear State In Whole Life Cycle

With the intensification of the demand for agricultural machinery equipment and the promotion of the concept of intelligent operation and maintenance,how to make the safe,efficient and sustainable operation of agricultural machinery equipment during service has become the focus of attention.Gear transmission system is the core component of high-end agricultural machinery equipment.Operation condition restricts the reliable operation of agricultural machinery equipment seriously.Wear is the main cause of gear failure.Online perception and intelligent warning of gear wear state are very important for reliable operation and health management of equipment.In this thesis,the on-line oil analysis and vibration monitoring method are used to study the gear box of agricultural machinery equipment.The wear state information of the whole life cycle of the gear was obtained online.Research on online perception and intelligent warning algorithm of gear wear state is carried out.The main work of the thesis includes the following.In order to reveal the wear law of gear in different wear stages,the pitting damage mechanism analysis and the whole life cycle wear state modeling of gear were carried out.The rheological theory was applied to the contact force analysis of gaussian and non-Gaussian rough surfaces,and the empirical crack initiation criterion was modified.The crack initiation Angle and propagation length were calculated to explore the influence of cracks of different lengths on meshing stiffness.From the point of view of surface roughness peak,the wear rate,chip concentration and vibration of gear in its whole life cycle were explained.In order to solve the problem of low accuracy in the extraction of abrasive features from visible ferrographic video images caused by high bubble interference,the intelligent extraction method of abrasive features from online ferrographic images with high bubble interference was studied.Firstly,according to the bubble motion characteristics in the deposition video,a PSO-GMMS bubble elimination algorithm based on moving target recognition is proposed to effectively remove the bubble interference.Second,aiming at the problems of low efficiency and limited extraction accuracy in constructing different abrasive label data sets by manual marking,an intelligent marking algorithm based on prior subblock wavelet coefficient was proposed.Realize automatic construction of abrasive label data set.The average marking time is about 3 seconds per sheet.Thirdly,based on different types of abrasive labels as training sets,the Faster-RCNN detection algorithm m AP was improved by 6.15%.Finally,the improved FRcnn-max-otsu algorithm was proposed to solve the problem of anchor frame redundancy and realize the10-dimensional parameter extraction of the abrasive feature.Aiming at the shortcoming of insufficient quantitative characterization parameters of wear state extracted from single oil information source,the intelligent early-warning algorithm based on oil-vibration heterogeneous multi-source information was studied.Firstly,a vibration-oil multi-source heterogeneous high dimensional characteristic parameter data set is constructed.Then,an intelligent early warning algorithm MIV-t SNE-RNN for multi-source heterogeneous high dimensional data is proposed.This algorithm can filter fault correlation of monitored data sets.Higher prediction accuracy is achieved with less monitoring data and shorter running time,which has been preliminarily verified on fan failure data.Finally,more than 180 hours of gear life cycle wear acceleration tests were carried out.5460 sets of oil vibration data were obtained.Based on the method proposed in this thesis,34-dimensional quantitative indexes were extracted and their sensitivity was compared and analyzed.Then,MIV-t SNE-RNN algorithm is used for early warning.The results show that the prediction accuracy of gear abnormal wear is 99.87%,and it takes 5.8846 seconds.Compared with the data without t SNE dimension reduction,the accuracy is improved by2.9% and the time is reduced by 46.6%.The above results show that the proposed gear wear state model can represent the evolution process of wear state.The method can realize gear abnormal wear warning.This study helps to improve the intelligent level of agricultural machinery equipment and ensure the safe,efficient and long-life operation of gear transmission equipment.This avoids the misjudgment of important state nodes to cause malignant accidents,which has important theoretical significance and practical engineering value.