A novel data fusion based intelligent identification approach for working cycle stages of hydraulic excavators

Abstract

Accurately identifying the stage of the excavator working cycle is the prerequisite to achieve the staged energy-saving control. However, current identification methods often overlook the influence of hydraulic system latency on identification results and depend on a single model, resulting in poor generalization performance of the identification approaches. Moreover, expert calibration system remains a necessary factor for improving identification accuracy. Aiming at these issues, a hybrid multi-scale feature extractor and a decision-level data fusion classifier approach (HMSFE-DFC) is proposed to identify the working cycle stages of excavator. The input signal employs mixed signals from the main pump pressure and the control current of the proportional solenoid valve to reduce the response delay caused by the single main pump pressure signal. A hybrid multi-scale feature extractor is constructed using a convolutional neural network temporal self-attention feature extraction mechanism and one-dimensional ResNet-50 architecture to extract multiscale features. To prevent overfitting, a decision-level data fusion classifier is used to fuse the decisions information of numerous classifiers. The accuracy of stage identification for 10 consecutive working cycles reaches 95.21%, which verifies its effectiveness.