基于张量Tucker分解的发动机故障诊断

传统的发动机故障诊断方法通常基于向量模式进行数据特征的提取，可能丢失数据之间的结构信息及破坏数据间相关性。针对此问题，提出了一种张量模式下提取发动机数据特征的方法，构建了“信号类别×曲轴转角×转速”的三阶张量形式的发动机状态样本，基于交替投影的思想，使用HOSVD-HOOI张量Tucker分解的联立求解算法，对数据特征进行提取。分别以不进行数据特征提取和基于张量Tucker分解进行数据特征提取两种情况，对发动机正常工作、单缸失火和轴系不对中三种状态下的实验数据进行处理，并分别采用网格参数优化法、遗传算法、粒子群算法对分类模型中的参数进行优化。以预测准确率和模型学习时间为评价指标进行对比分析，实验结果表明，基于张量Tucker分解的发动机数据特征提取及诊断方法预测准确率更高，分类模型学习时间更短。

Engine Fault Diagnosis Based on Tensor Tucker Decomposition

The traditional engine fault diagnosis method based on vector mode to extract data features was a method which might lose the structure informations of the data and damage the correlation between the data. To solve this problem, the paper presented a method to extract engine data features in tensor mode, and the engine state samples of the three order tensor form of the “signal class×crank angle×rotation rate” were constructed.Based on alternating projection theory, the HOSVD-HOOI tensor Tucker decomposition of the simultaneous solution algorithm was used to extract data features.The data were processed without feature extractions and with the method based on the tensor Tucker decomposition respectively,and the experimental data were processed under three kinds of state of normal operation, single cylinder misfire and shaft misalignment .The parameters of the classification model was optimized by using the grid parameter optimization method, genetic algorithm and particle swarm optimization respectively.Then the prediction accuracy and model learning time were compared and analyzed as evaluating indicator.The experimental results show that the prediction accuracy of engine data feature extraction and diagnosis method based on tensor Tucker decomposition is higher, and the learning time of classification model is shorter.