考虑传感器位置的Bagging集成卷积神经网络柱塞泵故障诊断模型
A Sensors Location Considered Piston Pump Fault Diagnosis Model With Bagging Based Convolutional Neural Network
滕建强 1罗风 2张菁 1张玉涛 3夏唐斌2
作者信息
- 1. 中国石油化工股份有限公司西北油田分公司,新疆乌鲁木齐 830011
- 2. 上海交通大学 机械与动力工程学院 机械系统与振动国家重点实验室 上海交通大学弗劳恩霍中心,上海 200240
- 3. 嘉洋智慧安全科技(北京)股份有限公司,北京 100012
- 折叠
摘要
针对柱塞泵故障诊断中存在的振动数据不平稳和诊断精度低问题,提出一种考虑传感器位置信息的 Bagging 集成卷积神经网络(Sensors location considered-bagging-convolution neural network,SLC-B-CNN)故障诊断模型.首先将短时傅里叶变换后的二维振动时频信息与经独热编码后的传感器位置信息匹配以构建数据集,然后设计混合双输入CNN模型作为基分类器,最后将数据集输入以简单平均法聚合的SLC-B-CNN模型中,在实验柱塞泵数据集上验证了所提出的SLC-B-CNN模型,在测试集上的准确率高达92％,各类故障平均召回率为89％,表现优于CNN模型和随机森林(Random forest,RF)模型.
Abstract
Aiming at the problems of unstable vibration data and low diagnostic accuracy in the fault diagnosis of piston pumps,a fault diagnosis model of sensors location considered-Bagging-convolution neural network(SLC-B-CNN)is proposed.Firstly,the two-dimensional vibration time-frequency information after short time Fourier transform is matched with the sensors location information encoded by one-hot to construct the dataset.Then,the dual-input hybrid CNN model is designed as the base classifier.And finally the dataset is put into the SLC-B-CNN model that is aggregated by simple averaging.The proposed SLC-B-CNN model is verified on the experimental piston pump dataset,with an accuracy rate of 92％on the test set and an average recall rate of 89％for various faults.The performance is better than the CNN model and the random forest model.
关键词
柱塞泵/故障诊断/卷积神经网络/集成学习/传感器位置信息Key words
piston pump/fault diagnose/convolutional neural network/ensemble learning/sensors location引用本文复制引用
出版年
2024
NETL
NSTL
万方数据