| 摘要: |
| 为防止渔网破损造成养殖鱼类逃逸，有必要对网衣进行破损检测。为了克服人工检测劳动强度大且效率低下的缺点，实现渔网的精准实时监测，本研究提出了一种基于数字孪生的网衣破损检测方法，可利用传感器代替人工监测。该方法首先从渔网的数值仿真模型获取大量的仿真传感数据，然后将数据用于人工神经网络的训练与测试，最后生成可进行网衣破损识别的数字孪生体。数字孪生体可根据传感器监测到的数据来判断网衣是否发生破损。在数值模拟中，考虑各种波浪条件以及网衣的破损情况。在训练人工神经网络中，将有效波高Hs、谱峰周期Tp以及横纲竖纲的拉力值作为输入变量，将网衣完整状态以及破损状态作为输出。经过测试分析，该识别模型根据传感器数据识别网衣是否破损的平均准确率为94.32%，由此可见，数字孪生技术能准确检测到渔网的损坏，可以作为网衣破损检测的一种新方法。 |
| 关键词: 数字孪生 人工神经网络 网衣 破损检测 |
| DOI：10.19663/j.issn2095-9869.20210825001 |
| 分类号: |
| 基金项目: |
|
| Research on the damage detection method of the plane fishing net based on the digital twin technology |
|
LIAN Likai1, ZHAO Yunpeng1,2, BI Chunwei1, XU Zhijing2, DU Hai1
|
|
1.State Key Laboratory of Coastal and Offshore Engineering, Dalian University of Technology, Dalian, Liaoning 116024, China;2.Ningbo Research Institute, Dalian University of Technology, Ningbo, Zhejiang 315016, China
|
| Abstract: |
| If damages to aquaculture nets are not found in time, they will result in the escape of fish, thereby, causing considerable losses to farmers. Therefore, it is necessary to detect whether damage to fishing net occurs. At present, the primary method for detecting damage to fishing nets is the manual inspection of staff diving into the water, but this method is labor-intensive and inefficient. This paper proposes a damage detection method based on digital twin, which uses sensor monitoring instead of manual monitoring to overcome these limitations and realize real-time monitoring of fishing nets. The research first shows that the numerical simulation data in good agreement with the physical model experimental data can be obtained through the numerical simulation of the lumped mass mechanical model. In the numerical simulation, considering a kind of damage to the fishing net, a total of 11 simulations were carried out: the tensile values of the horizontal and vertical ropes of the fishing net, nine sea conditions as training samples, and two sea conditions as test samples. The artificial neural network adopts the error backpropagation training method that takes the significant wave height Hs, the spectral peak period Tp, and the tensile value of the vertical and horizontal rope as inputs, and the complete state and damaged state of the fishing net as the outputs. After training, the recognition model recognition accuracy rates for the training and test samples were 99.21% and 95.11%, respectively. The measured actual physical sensor data were also used as test data. The recognition accuracy of the recognition model is 94.32%, which indicates the feasibility of the practical application of digital twin technology in the damage detection of the net. It can, therefore, be used as a new method of fishing net damage detection. As the wave-current environment is more complex in the actual sea area, our following research will focus on dealing with the sensing data and detection of the damage of fishing nets in a more realistic sea condition. |
| Key words: Digital twin Artificial neural network (ANN) Fishing net Damage detection |
