基于Mask R-CNN的污泥热解工艺状况智能诊断  

Intelligent Diagnosis of Sludge Pyrolysis Process Conditions Based on Mask R-CNN

作　　者：张珂[1] 白禹启 王泽华 郑宾国[1] ZHANG Ke;BAI Yu-qi;WANG Ze-hua;ZHENG Bin-guo(School of Civil Engineering and Architecture,Zhengzhou University of Aeronautics,Zhengzhou 450046,China;Zhengzhou Sewage Purification Co.,Ltd.,Zhengzhou 450000,China)

机构地区：[1]郑州航空工业管理学院土木建筑学院,河南郑州450046 [2]郑州市污水净化有限公司,河南郑州450000

出　　处：《山东工业技术》2022年第5期3-7,共5页Journal of Shandong Industrial Technology

基　　金：国家留学基金(青年骨干教师出国研修项目,201809905008);河南省科技攻关项目(222102320273)。

摘　　要：针对目前污泥热解工艺状况判断主观性强、人力成本高、反馈效率低等问题,结合深度学习理论在诊断预测领域的优势,提出了基于掩膜区域卷积神经网络(Mask R-CNN)的污泥热解状况智能诊断模型。以某污泥处理厂市政污泥处置项目为例,采集污泥热解出渣图像数据并构建规范化样本数据库,根据出渣图像所反映的工艺优劣状况,将样本数据分为5类状态,对模型进行训练与测试。实验结果表明,所构建模型实现了对污泥热解工艺状况的有效学习与诊断,为推动低碳高效智能的污泥处置模式提供了技术支持。Sludge pyrolysis process condition diagnosis is highly subjective, labor-costing and with low feedback efficiency. Aming at these problems, an intelligent diagnosis model of sludge pyrolysis condition using Mask Region-based Convolutional Neural Network(Mask R-CNN) is proposed, using the advantages of deep learning theory in such fields. Taking a sludge treatment plant as an example, the sludge pyrolysis residue image data were collected and a normalized sample database was built. According to the process conditions reflected by the residue images, the sample data were divided into 5 categories for the training and testing of the Mask R-CNN model. The experimental results show that the model achieves effective learning and diagnosis performance of sludge pyrolysis process conditions, which provides technical support to promote a low-carbon,efficient and intelligent sludge disposal pattern.

关 键 词：污泥处置 热解工艺 工艺诊断 Mask R-CNN 图像识别 

分 类 号：X703[环境科学与工程—环境工程]