Abstract:

In the water intake head and pipeline of the seawater circulation cooling system, fouling organisms will block the pipeline, accelerate corrosion, and seriously affect the normal operation of the equipment. The biocide dosing scheme is closely related to the type of fouling organisms. Due to the difficulty of monitoring, the fixed biocide dosing scheme is usually adopted. The attachment of fouling organisms on the pipe wall structures is mainly in the form of aggregation of dominant species. Therefore, cameras can be installed at the head of the seawater pipe and in the pipe to realize the monitoring of fouling organisms, so as to adjust the dosing scheme in time. In this paper, the convolution neural network algorithm was used to establish the classification and recognition model of fouling organisms, and to realize the automatic classification and recognition of common fouling organisms. The cross entropy loss function and accuracy rate were used as model evaluation indicators to train the model. The model could be used for automatic identification of fouling organisms in automatic dosing equipment. On this basis, with automatic dosing equipment, the automatic real-time adjustment of dosing scheme could be realized to improve the refined management level of seawater circulation cooling system.

Key words: seawater cooling, convolutional neural networks, computer vision, classification of fouling organisms

摘要：

海水循环冷却系统取水头部及管道中污损生物附着会堵塞管道，加速腐蚀，严重影响设备的正常运行。杀生剂的投加方案与污损生物的种类密切相关，由于监测困难，现在通常采用固定杀生剂投加方案。污损生物在管壁建构筑物上的附着主要是以优势种群聚集附着的形式，可以在海水管道头部及管道中加装摄像头，实现污损生物的监测，以便及时调整加药方案。利用卷积神经网络算法，建立污损生物分类模型，实现常见污损生物的自动分类。利用常见污损生物聚集图像作为模型训练数据，以交叉熵损失函数和准确率作为模型评价指标，进行模型训练。该模型分类准确率较高，可用于自动化加药设备中污损生物的自动识别，以此为基础，配合自动化加药设备，可实现杀生药剂投加方案的自动实时调整，提高海水循环冷却系统的精细化管理水平。

关键词: 海水循环冷却, 卷积神经网络, 计算机视觉, 污损生物分类

CLC Number: 

• P747

• TP181