东北大学学报（自然科学版）

东北大学学报（自然科学版） ›› 2023, Vol. 44 ›› Issue (10): 1464-1474.DOI: 10.12068/j.issn.1005-3026.2023.10.013

• 机械工程 • 上一篇    下一篇

基于深度学习的挖掘机工作阶段的分类与识别

刘伟嵬1,2， 邓剑洋1,2， 张靖文1,2， 牛东东3   

  1. (1. 大连理工大学 机械工程学院， 辽宁 大连116024； 2. 大连理工大学 高性能精密制造全国重点实验室， 辽宁 大连116024； 3. 徐州徐工挖掘机械有限公司， 江苏 徐州221004)
  • 发布日期:2023-10-27
  • 通讯作者: 刘伟嵬
  • 作者简介:刘伟嵬(1981-)，男，辽宁丹东人，大连理工大学副教授，博士生导师.
  • 基金资助:
    国家重点研发计划项目(2020YFB1709903).

Classification and Identification of Excavators’ Working Stages Based on Deep Learning

LIU Wei-wei1,2， DENG Jian-yang1,2， ZHANG Jing-wen1,2， NIU Dong-dong3   

  1. 1. School of Mechanical Engineering， Dalian University of Technology， Dalian 116024， China; 2. State Key Laboratory of High-performance Precision Manufacturing， Dalian University of Technology， Dalian 116024， China; 3. Xuzhou Construction Machinery Group Co.， Ltd.， Xuzhou 221004， China.
  • Published:2023-10-27
  • Contact: LIU Wei-wei
  • About author:-
  • Supported by:
    -

RichHTML

22

PDF (PC)

200

摘要： 为实现对挖掘机作业循环各工作阶段的自动识别，采用以执行机构先导压力、主泵压力和功率为识别对象的智能识别方法.根据执行机构先导压力的变化划分工作阶段，并用主泵压力和功率验证.以各工作阶段起始特征波形作为其起始标志，以时间窗滑移方式提取起始特征并确定最佳时间窗宽度，采用深度学习的方法识别各标志.对比了深度学习中分类识别领域应用广泛的ResNet和LSTM的识别效果，发现LSTM的识别效果更好，对测试集的识别准确率最高可达到99.75%.采用LSTM对测试数据进行识别，识别正确率仅有82.54%，说明存在误识别.提出以挖掘机工作阶段的逻辑顺序和设定主泵功率阈值作为校正依据对误识别进行校正，识别正确率可提升至99.72%.结果表明，该方法识别准确率高，可有效识别作业循环各工作阶段.

关键词: 液压挖掘机；工作阶段；残差神经网络(ResNet)；长短期记忆(LSTM)神经网络；智能校正系统

Abstract: In order to realize the automatic identification of each working stage of excavators’ operation cycle， an intelligent identification method is adopted which takes the pilot pressure of actuators， and the pressure and power of the main pump as the identification objects. The working stages were divided by the pilot pressure change of each actuator， and then verified by the pressure and power change of the main pump. Defining the waveform that begins with each working stage as the staged symbol， the data features were extracted and the optimal time window width was determined in the form of time-window slips. Deep learning was used to identify each segment marker. Comparing the identification effects of ResNet and LSTM neural networks， which are widely used in the field of classification identification in deep learning， it was found that LSTM has better identification effects， and the identification accuracy of the test set can reach 99.75%. LSTM is used to identify the test data， and the identification accuracy is only 82.54%， indicating that there exists misidentification. Based on the logical sequence of excavators’ working stages and the power threshold of the main pump， the identification accuracy can be increased to 99.72%. The results show that the proposed method has high identification accuracy and good real-time performance， and can effectively identify each working stage of the operation cycle.

Key words: hydraulic excavator; working stage; residual neural network(ResNet); long short-term memory(LSTM)neural network; intelligent correction system

中图分类号: