基于数据驱动的波形护栏监测数据故障诊断方法研究

A Method for Diagnosing Waveform Guardrail Monitoring Data Fault Based on Data-driven

为了研究波形护栏撞击事件的监测算法,实现能够从过程数据中消除环境噪声,区分正常震动干扰数据和撞击数据,同时能准确定位故障发生的位置,采用了基于数据驱动的故障诊断方法,该方法将监测数据视为"过程数据",将撞击类数据视为一种"数据故障",将环境干扰数据视为"噪声数据",将大型货车经过时产生的正常幅度的震动数据视为"干扰数据"。采用对比试验的方式来挑选适合护栏监测的数据驱动方法,再经过对该方法中各环节的优化和组合,提出了一种基于多尺度费舍尔判别分析的数据故障诊断模型。分析了基于数据驱动方法和基于传统利用经验阈值的方法的不同,以及基于数据驱动在监测波形护栏撞击事件的优点。分析了上述噪声数据的特点,优化了前期研究中的小波阈值除噪算法。分析了上述其他3类数据的特点,选择数据驱动中PCA,PLS,FDA这3类可能适合波形护栏撞击监测的算法,并分析了各类算法的工作原理。对比试验结果表明:基于数据驱动方法的护栏碰撞监测方法比基于阈值的监测方法准确率更高,误报率更低;经过优化后的小波阈值除噪算法能明显降低环境噪声对数据质量的影响;FDA相比较PCA和PLS更适合护栏撞击的数据故障诊断;与优化的小波阈值除噪算法组合后构成的MSFDA模型相比较,FDA模型抗噪能力更强,准确度更高。

In order to study the monitoring algorithm of waveform guardrail impact event,to eliminate environmental noise from process data,to distinguish normal vibration interference data and impact data,and accurately locate the location of the fault,a data-driven based fault diagnosis method is adopted,which regards the monitoring data,the impact data and the environmental interference data as"process data","data fault"and"noise data"respectively,and regards the normal amplitude vibration data generated by the passing of large trucks as"interference data".A data-driven method suitable for guardrail monitoring is selected by means of comparative test,and a data fault diagnosis model based on multi-scale Fischer discriminant analysis is proposed after optimizing and combining each link of this method.The differences between the data-driven based method and the traditional method using empirical threshold is analyzed,and the advantages of data-driven based method in monitoring the waveform guardrail impact event is also analyzed.The characteristics of noise data mentioned above are analyzed,and the wavelet threshold denoising algorithm used in previous research is optimized.The characteristics of the other 3 types of data mentioned above are analyzed.Three types of data-driven algorithms(PCA,PLS,and FDA)which may be suitable for waveform guardrail impact monitoring are selected,and the working principles of different algorithms are analyzed.The comparative experiment result shows that(1)the guardrail impact monitoring method based on data-driven has higher accuracy and lower false alarm rate than the threshold based monitoring method;(2)the optimized wavelet threshold denoising algorithm can obviously reduce the influence of environmental noise on data quality;(3)compared with PCA and PLS,FDA is more suitable for data fault diagnosis of guardrail impact;(4)Compared with the MSFDA model formed by the combination of optimized wavelet threshold denoising algorithms,FDA model has stronger anti-noise ability and higher accuracy.