基于多传感器数据融合的拖拉机驱动轮滑转率检测方法与试验

Drive Wheel Slip Rate Detection Method and Experiment for Tractor Based on Multi-sensor Data Fusion

摘要:

犁耕作业拖拉机驱动轮滑转程度的精准检测是实现滑转率控制的基础，有助于提高拖拉机的牵引效率和犁耕作业质量。针对重负荷犁耕作业环境下滑转率检测准确性、实时性难以保证的问题，提出了一种基于多传感器数据融合的滑转率检测方法。搭建了拖拉机犁耕作业数据实时采集平台，通过田间试验获取不同工况下的犁耕速度、前进加速度、车身俯仰角、侧倾角、犁耕阻力、耕深以及对应的滑转率，构建数据集并进行统计学分析，基于随机森林（Random forest，RF）、支持向量回归（Support vector regression，SVR）、径向基神经网络（Radial basis function neural network，RBFNN）、人工神经网络（Artificial neural network，ANN）4种机器学习算法搭建滑转率检测模型，采用网格搜索（Grid search）、粒子群算法（Particle swarm optimization，PSO）对模型超参数进行寻优，并使用决定系数（R2）、均方根误差（RMSE）以及平均绝对百分比误差（MAPE）3种评价指标评估模型性能。结果表明，除RF模型存在过拟合现象外，其他3种模型均可以较好地实现滑转率的预测（R2>0.9）。其中，使用PSO进行初始阈值权重寻优的ANN模型在测试集的性能（R2为0.937，RMSE为1.0%，MAPE为7.6%）优于SVR模型（R2为0.918，RMSE为2.6%，MAPE为8.9%）与RBFNN模型（R2为0.903，RMSE为3.0%，MAPE为8.8%），能够更加可靠、精确地实现拖拉机驱动轮滑转率的检测，为实现滑转率精准在线检测与控制提供了新思路。

Abstract:

Accurate detection of the drive wheel slip in tractor ploughing operations is fundamental to achieving slip control, which helps improve the traction efficiency and ploughing quality of a tractor. To address the issue of ensuring the accuracy and real-time performance of slip rate detection under heavy load ploughing conditions, a slip rate detection method based on multisensor data fusion was proposed. A real-time data collection platform for tractor ploughing operations was established. Field experiments were conducted to obtain tillage speed, forward acceleration, tractor pitch angle, roll angle, traction resistance, tillage depth and corresponding slip rate under different operating conditions. A datasets was constructed and subjected to statistical analysis. Four machine learning algorithms—random forest (RF), support vector regression (SVR), radial basis function neural network (RBFNN), and artificial neural network (ANN)—were employed to construct the slip rate detection models. The hyperparameters of these models were optimized by using grid search and particle swarm optimization (PSO). The performance of the models was evaluated via three metrics: the coefficient of determination (R2), the root mean square error (RMSE), and the mean absolute percentage error (MAPE). The results indicated that, except for the RF model, which exhibited overfitting, the other three models achieved satisfactory slip rate predictions (R2>0.9). Among them, the ANN model optimized with particle swarm optimization (PSO) for initial threshold weight optimization outperformed both the SVR model with an R2 of 0.918, an RMSE of 2.6% and a MAPE of 8.9%, and the RBFNN model with an R2 of 0.903, an RMSE of 3.0%, and a MAPE of 8.8% on the test dataset, achieving an R2 of 0.937, an RMSE of 1.0%, and a MAPE of 7.6%. This method enabled reliable and accurate detection of tractor drive wheel slip rate, offering an approach for precise online detection and control of slip rate.