Recognition methods of threshing load conditions based on machine learning algorithms

Abstract

In order to identify the load conditions of threshing and separating device of combine harvester quickly and accurately, this experiment was conducted to collect vibration acceleration signals of the outer surface of threshing and separating device under different load conditions by field test in 2020 with the new rice variety Nanjing Jinggu as the research object. Firstly, based on the statistical analysis and signal analysis method, time domain, frequency domain, and time–frequency domain characteristics were extracted and fused into total domain characteristics to characterize the overall signal attributes of load conditions of the threshing and separation device so as to reduce the difficulty of data decision. Secondly, Principal Component Analysis (PCA) and Kernel Principal Component Analysis (KPCA) were used to remove the correlation and nonlinearity among the extracted characteristics, reduce the dimensionality of the characteristic vector and improve the accuracy of the diagnostic model. Finally, Support Vector Machine (SVM), Extreme Learning Machine (ELM) and Random Forest (RF) were used to diagnose the load conditions of the dimension reduction collection of total domains, and the accuracy rate and recognition time were used as evaluation indexes for the comparative analysis of model recognition. The results showed that the KPCA clustering separation effect is significant. RF has the highest recognition accuracy, which the accuracy of training set and prediction set are 100% and 98% respectively. The accuracy of ELM-KPCA model training set and prediction set is 100% and 90% respectively, and the analysis time is 6.206 s. This model accuracy is high, and the analysis time is the shortest, then ELM-KPCA model can be the best model for load conditions recognition of combine harvester.