机载燃油泵故障诊断及实验平台研究

Fault diagnosis and test platform for airborne fuel pumps

以机载燃油泵实时状态监测为背景,针对目前燃油泵故障数据少、诊断方法效率低、成本高等问题,研制了机载燃油转输系统实验平台,提出基于小波包分析和改进粒子群支持向量机(M-PSO-SVM)的故障诊断方法。该实验平台可针对燃油泵5种典型故障模式进行实验,测取泵故障状态下的振动信号和出口压力信号。利用小波包分解提取振动信号不同频段的能量值作为特征参数,并结合出口压力均值构造故障特征向量。提出混合遗传变异思想的粒子群算法对SVM分类模型进行参数优化,用得到的故障特征向量训练并验证该分类模型。实验分析表明,该实验平台可有效采集泵的故障信号,并且测试点可进一步优化,M-PSO-SVM在诊断速度、诊断精度等方面都优于传统Grid-SVM和GA-SVM,能够满足实际故障诊断的需求。

Under the background of real-time status monitoring for airborne fuel pumps,aiming at lack of fault data and efficiency,and high-cost of now available fault diagnosis methods,a test platform of a fuel transfer system was developed and a fault diagnosis method based on wavelet packet analysis,modified particle swarm optimization and support vector machine（ M-PSO-SVM） was proposed. The test platform could run tests for five typical fault modes of fuel pumps to acquire vibration signals and outlet pressure signals under malfunction conditions. The energy of different frequency bands of vibration signals extracted with the wavelet packet decomposition was regarded as characteristic parameters to construct fault feature vectors combined with the mean outlet pressures. The particle swarm optimization algorithm with the thought of genetic variation was presented to optimize the parameters of a SVM classification model. Meanwhile,the fault feature vectors were used to train and validate this classification model. The examples demonstrated that the test platform is quite effective to get fault signals of fuel pumps and the measurement points can be further optimized; the M- PSO-SVM has higher performances than Grid-SVM and GA-SVM do and it can meet the requirements of practical fault diagnosis.