【目的】为评估自动引导车(AGV)驱动电机的健康状态、预测故障概率，改进AGV的保养维护和工作策略，本文结合AGV驱动电机的历史状态数据和预定义的工作负荷，提出了一种AGV驱动电机的故障预测与健康管理(PHM)模型。【方法】首先，采集AGV驱动电机的负载电流、振动信号以及温度数据，并对采集的数据进行去噪和归一化处理，以提高模型的收敛速度和泛化能力。然后，采用自回归模型和卷积神经网络模型对AGV驱动电机的负载电流、振动信号及温度数据的变化趋势进行预测，并将采集的数据和预测的数据转化为对称点图案(SDP)。最后，基于YOLOv11网络对生成的SDP进行分类，从而确定AGV驱动电机的健康等级。根据电机温升将健康等级分为不健康、亚健康和健康三类，基于电机的负载电流、振动信号采用本文所提模型检测驱动电机健康状态并估计驱动电机属于某类健康等级的概率，基于健康状态的测定可以得到AGV驱动电机的故障概率。【结果】对本文模型在验证集和测试集上进行验证测试以评估模型性能。结果表明AGV驱动电机的3类健康状态的平均诊断准确率为99.7%，其中健康和不健康两类的诊断准确率达到了100%。为进一步验证本文模型的优越性，与其他两种模型进行对比，结果表明本文模型的诊断准确度高于其他模型，具有较高的可信度。【结论】本文提出的PHM模型对故障概率预测和健康状态评估的准确度较高，将AGV工作负荷信息整合到PHM模型中，可为AGV的工作任务制定和保养维护提供数据参考。

[Objective] A method has been proposed to preprocess, predict, and classify the load data of automated guided vehicle (AGV) drive motors, and then predict the health status and failure of AGVs. The aim is to evaluate the health status and failure probability of AGV motors, and improve the maintenance and work strategies of AGVs. [Methods] Based on the experimental collection of AGV motor channel current, vibration signal data, and temperature data, the data is sampled. The proposed method uses autoregressive model and convolutional neural network model to predict the health status and calculate the failure probability of AGV drive motor's current, vibration signal, and temperature rise data trends. The collected and predicted data are converted into symmetrized dot pattern (SDP) images using SDP algorithm for classification detection, thereby determining the health level of the working motor. After classifying the health level of the dataset into three levels based on the temperature rise data of the motor, the autoregressive model and Convolutional network model is used to detect the health status of the driving motor and estimate the probability of the driving motor health level based on the load current and vibration signals of the AGV motor. Based on the determination of the health status, the statistical model can calculate the failure probability of the AGV driving motor. [Results] The data verification through the acceleration test of the driving motor shows that the accuracy of this method in evaluating the health status diagnosis of AGV driving motors reaches an average of about 99.7%, with an accuracy of 100% in classifying the test samples as healthy and unhealthy. When predicting the probability of AGV drive motor failure under planned workload, the root mean square error of AGV motor state data prediction reaches around 0.053. [Conclusion] By applying deep learning methods to the current, vibration, temperature rise and other data of AGV motors, the health level classification of AGV motors (healthy, sub healthy, unhealthy) is achieved. Based on the evaluation results of health status and the calculation of failure probability, reference is provided for the assessment of AGV workload intensity and maintenance plan.