Spatio-Temporal Data Fusion-Based Method for Working Cycles Identification of Hydraulic Excavators

Accurately describing and identifying the operational modes of excavators is fundamental to achieving efficient and energy-saving control of excavators. Existing methods predominantly rely on hydraulic power system parameters, utilizing machine learning models to recognize the operational states of excavators. However, the limited operational data from a single or a few sensors leads to inputs that inadequately reflect the diverse action states of the equipment and overlook the spatial correlations between actions of different actuators, failing to effectively learn the distinctive features between operational conditions. To address these issues, this article introduces a spatio-temporal fusion method for condition recognition based on multidimensional sensor data, encapsulating comprehensive operational information of the entire machine. This study developed an attention mechanism linked to an operation-related attention module (OAM), integrated with a designed dual-channel feature extraction network for time and space, to enhance and capture the spatio-temporal correlations between adjacent operational states. Furthermore, a classifier based on imbalanced interclass distance metrics was designed to adaptively learn the distances between category samples, thereby mitigating the learning challenges posed by imbalanced condition samples. Experiments conducted under various conditions on single and multiple working cycles achieved recognition accuracies of 99.17% and 99.24%, respectively, demonstrating the effectiveness and robustness of our proposed method and significantly improving the accuracy of condition recognition.