童鞋论文网专业提供代写本科论文、代写硕士论文、代写职称论文、代写毕业论文、论文发表等服务,上万成功案例,欢迎咨询。 网站地图  |  加入收藏  |  设为首页
童鞋论文网
热门搜索
行政 教育 营销 法律 经济 护理 心理 会计 文学 工商 计算机 旅游
站内搜索:
童鞋论文网
 当前位置:童鞋论文网 > 毕业论文 > 文章正文
 
基于卷积神经网络的城市燃气管道故障诊断技术研究
论文作者:童鞋论文网  论文来源:www.txlunwenw.com  发布时间:2019/9/4 8:08:06  

摘要:随着我国经济总量的迅猛增长以及现阶段我国政府和能源消费者对环境保护逐渐重视,我国的能源结构发生了翻天覆地的变化,燃气产业得到迅猛的发展。同时,因燃气管网泄漏所造成事故数量也呈上升趋势。因此,提高燃气管道故障的识别能力以及准确度,减少燃气管道安全事故的发生,就显得尤为重要。鉴于此,本文将卷积神经网络模型运用到燃气管道故障图像判别中,进行管道故障诊断识别。

作为深度学习中主要及常用模型之一,卷积神经网络模型在图像分辨识别方面有明显的优势。运用包含信息较为全面的时频图作为管道故障诊断模型的输入样本,可以提高模型的诊断正确率。实验室条件下,通过采集管道不同运行状态下的模拟故障声发射信号,并应用MATLAB工具箱对声发射信号进行分解、消噪处理,并将管道故障信号运用小波变换转化为时频图,然后将时频样本输入到已建好的卷积神经网络模型中,用于进行管道故障分类识别。通过数次的管道故障声发射检测试验并进行分析,可以获得如下研究结果:

(1)将小波分析法用于声发射管道故障信号波形的分解与消噪,并将得到的波形图转换为时频图,可以较为准确的反应管道运行的真实状态,有助于提高卷积神经网络模型的诊断正确率。

(2)卷积神经网络的模型诊断正确率会随着迭代次数、批量尺寸、卷积核大小、卷积核个数这些模型参数的变化而变化,为了得到最高的模型正确率,通过实验研究分析,可得出最佳的参数组合,构建出相应参数组合的卷积神经网络模型,并与SOFTMAX分类器相结合,可以构建出基于卷积神经网络性能较优的燃气管道故障诊断模型。

(3)为了验证卷积神经网络在管道故障图片识别领域的性能,运用卷积神经网络进行管道故障诊断,并与同为深度学习模型的深度置信网络模型以及堆栈式自编码网络模型相比,结果显示,在相同条件下,卷积神经网络表现出更为优越的诊断识别性能。具体表现在两方面:在故障诊断平均正确率方面,卷积神经网络诊断平均正确率均高于其他两者平均正确率,由此可以看出,在图像识别方面,三者对图像的识别均表现较好,但卷积神经网络模型表现稍强;在诊断运算用时方面,卷积神经网络模型所用时间均不足其他两者的1/2。综合考虑,卷积神经网络在管道故障的图像判别领域表现更优。

With the rapid growth of China's economic aggregate and the increasing emphasis on environmental protection by the Chinese government and energy consumers at this stage, China's energy structure has undergone earth-shaking changes, and the gas industry has developed rapidly. At the same time, the number of accidents caused by the leakage of gas pipelines is also on the rise. Therefore, it is particularly important to improve the identification capability and accuracy of gas pipeline faults and reduce the occurrence of gas pipeline safety accidents. In view of this, this paper applies the convolutional neural network model to the gas pipeline fault image discrimination to conduct pipeline fault diagnosis and identification.

As one of the main and commonly used models in deep learning, the convolutional neural network model has obvious advantages in image resolution and recognition. Using the time-frequency map with more comprehensive information as the input sample of the pipeline fault diagnosis model can improve the diagnostic accuracy of the model. Under laboratory conditions, the simulated fault acoustic emission signals under different operating conditions of the pipeline are collected, and the acoustic emission signal is decomposed and denoised using the wavelet transform toolbox in MATLAB, and the pipeline fault signal is converted into time using wavelet transform. The frequency map is then input into the established convolutional neural network model for pipeline fault classification and identification. Through several pipeline fault acoustic emission detection tests and analysis, the following research results can be obtained:

(1) Using wavelet analysis method to decompose and denoise the waveform of the fault signal of the acoustic emission pipeline, and convert the obtained waveform diagram into a time-frequency diagram, which can accurately reflect the true state of the pipeline operation and help to improve the convolution. The diagnostic accuracy rate of the neural network model.

(2) The correct rate of model diagnosis of convolutional neural network will change with the change of model parameters such as the number of iterations, batch size, convolution kernel size and number of convolution kernels. In order to obtain the highest model correct rate, experimental research Analysis, the best combination of parameters can be obtained, and the convolutional neural network model of the corresponding parameter combination is constructed. Combined with the SOFTMAX classifier, a gas pipeline fault diagnosis model based on the convolutional neural network can be constructed.

(3) In order to verify the performance of convolutional neural networks in the field of pipeline fault picture recognition, convolutional neural networks are used for pipeline fault diagnosis, and compared with the deep confidence network model and the stacked self-coding network model, which are also deep learning models. The results show that under the same conditions, the convolutional neural network shows superior diagnostic recognition performance. The specific performance is in two aspects: in terms of the average correct rate of fault diagnosis, the average correct rate of convolutional neural network diagnosis is higher than the average correct rate of the other two. It can be seen that in image recognition, the three images are recognized. The performance is better, but the convolutional neural network model performs slightly better; in terms of diagnostic calculation time, the convolutional neural network model takes less than 1/2 of the time of the other two. Considering comprehensively, convolutional neural networks perform better in the field of image discrimination in pipeline failure.

关键词:燃气管道;声发射技术;卷积神经网络;管道故障诊断

gas pipeline; acoustic emission technology; Convolutional neural network; pipeline fault diagnosis

上一篇:基于声发射技术的城市管道泄漏精确定位研究     下一篇:基于视频图像处理的轨道交通客流检测方法的研究
 
论文客服 95702844
论文热线18796337551
本站网址 www.txlunwenw.com
郑重承诺 原创,包修改,包通过!
童鞋论文网真诚欢迎新老客户的光临与惠顾!
 
 热门文章
· 住宅建设用地使用权续期研究
· 用户锁定效应视角下的网易云音
· 众筹产品创新性对融资绩效的影
· 南京市城镇居民冷鲜肉认知度问
· 摩拜共享单车可持续盈利研究
· 催生地方政府隐性债务风险的原
· 社交媒体中不实信息转发主体心
· 基于网络平台的SPA模式研究——
· 基于游客感知的云台山景区微信
· 新媒体时代背景下乡村旅游企业
 
 推荐文章
· 论商业性小屏直播的主持互动策
· 新兴偶像经济发展现状及发展建
· 辅助线在小学生空间观念培育中
· 罗浮山国家级风景名胜区探险旅
· 基于《预算法》的地方政府债务
· 商品房买卖合同“霸王条款”存
· 中泰水果贸易竞争力比较分析
· 我国企业内部转移价格应用现状
· 虚拟现实技术下的创意短片《萌
· 核定征税与查账征税实施现状及
 
 相关文章
· 论当代中国画应该站在传统的基
· 新木刻艺术“经纪人”鲁迅研究
· 从“人史”浅谈鲁迅与浙东学派
· 张锐锋散文集《文学王》解构
· 参考文献在毕业论文中所反映出
· 工科本科生毕业论文质量提高研
· 高职高专毕业论文质量提高探讨
· 商务英语毕业论文文献综述写作
· 体育教育专业本科毕业论文现状
· 毕业论文写作规范介绍

论文原创,准时交稿,包修改,包通过,八年信誉,品质保证
童鞋论文网提供代写本科论文,代写专科论文,代写毕业论文,代写职称论文,代写硕士论文,论文发表及查重服务
备案号:苏ICP备11011187号-1 苏公网安备32030502000117号
版权归童鞋论文网所有 禁止一切商用盗用