Research On The Diagnosis Of Grain Threshing Drum Blocking Load State Based On Vibration Signal

With the development of agricultural modernization and scale,the future development direction of domestic combine harvester is general,automatic and intelligent.The investigation report of agricultural machinery failure shows that the threshing separation device accounts for 30% of the blocking failure of the combine harvester,which seriously restricts the working efficiency of the machine.In the field operation,the threshing separation device has the characteristics of nonlinearity and time-varying.The load state changes depend on the driver’s experience for a long time.Therefore many problems are caused such as personal empiricism and low accuracy of judgment.If the precise load state of threshing and separating device of combine harvester is not mastered,the blockage fault could occur and it will take time and efforts to fix this problem.The researchers had used different methods to detect the load state of the combine harvester by monitoring the feed quantity,threshing drum speed,torque,oil pressure of the cylinder after concave plate,but the above methods have some limitations.In order to improve the working quality and production efficiency of the combine harvester,this paper proposed a method of diagnosis of the load blocking state of threshing based on vibration signal,in order to provide a new way for on-line monitoring and blocking prevention of the load state of the threshing separation unit of the combine harvester.The main contents of this paper are as follows:(1)A vibration signal acquisition system composed of several acceleration sensors,constant current adapter,data acquisition card and upper computer was built to obtain the multi-point vibration signal on the outer surface of threshing separation device under different working conditions of the combine harvester.The original vibration signal is a nonlinear random vibration signal and the source is complex,which accumulates a lot of real data for the subsequent analysis and processing.(2)In view of the problem that the original vibration signal has high frequency random noise and the field condition is complex,a signal processing method based on short-term feature set was proposed.Firstly,the original data samples were collected according to the working conditions of the test site,and the load state of threshing and separating device in field operation of the combine harvester was divided into five categories: no-load,low load,high load,tendency to blockage and blockage;Secondly,the original signal was divided into two seconds time segment,with the overlapping rate of 50%;Finally,the five point three smoothing method was used to eliminate the high frequency random noise and trend term,and the processed time segment was analyzed and extracted in time and frequency domain in time and frequency domain,and the feature domain was formed by information fusion.(3)Aiming at the problem that the dimension of feature set was high and there was a large amount of redundant information,which affected the operation speed and accuracy of subsequent classification and recognition.A multi-domain feature fusion and dimension reduction method based on principal component analysis(PCA)and kernel principal component analysis(KPCA)was proposed to reduce the dimension of feature set in time domain,frequency domain,time-frequency domain and mutidomain feature respectively.The analysis results showed that the dimension reduction method of KPCA feature set based on total domain had the best clustering effect for five load states.(4)Support vector machine(SVM),limit learning machine(ELM)and random forest(RF)were used to recognize,classify and diagnose the state of each domain feature set after dimensionality reduction of PCA and KPCA.The classification effects of the three algorithms were analyzed and compared from the diagnosis accuracy and diagnosis recognition time.The results showed that in comparison with the time domain,frequency domain and time-frequency domain feature sets alone,the three algorithms all had higher recognition accuracy for the proposed multi-domain fusion feature sets.In terms of diagnosis and recognition time,ELM < RF < SVM.From the perspective of diagnostic recognition accuracy,RF > SVM > ELM.It could be seen that the vibration signal analysis technology can realize the early prediction of the blockage fault of the threshing and separation device of the combine,and provide a new idea for the load state identification of the combine.