为了进一步提高滚动轴承故障检测的准确性、改善时间卷积网络模型(Temporal Convolutional Network,TCN)存在的过拟合问题,本研究提出了增加平均池化层的时间卷积网络(Temporal Convolution with Average Pooling Network,TCAPN)模型。...为了进一步提高滚动轴承故障检测的准确性、改善时间卷积网络模型(Temporal Convolutional Network,TCN)存在的过拟合问题,本研究提出了增加平均池化层的时间卷积网络(Temporal Convolution with Average Pooling Network,TCAPN)模型。该方法首先使用膨胀因果卷积代替传统卷积神经网络,其次在残差模块多个地方加入平均池化层改善模型过拟合问题,最后结合多个改进残差模块构建本研究提出的TCAPN模型。实验结果表明,在相同工况条件下,TCAPN模型能够更快地收敛,并且平均故障诊断准确率达到了98.73%,相较于TCN模型提高了2.87%,验证了该模型具有高准确性和鲁棒性。展开更多
Industrial wireless sensor networks adopt a hierarchical structure with large numbers of sensors and routers. Time Division Multiple Access (TDMA) is regarded as an efficient method to reduce the probability of confli...Industrial wireless sensor networks adopt a hierarchical structure with large numbers of sensors and routers. Time Division Multiple Access (TDMA) is regarded as an efficient method to reduce the probability of confliction. In the intra-cluster part, the random color selection method is effective in reducing the retry times in an application. In the inter-cluster part, a quick assign algorithm and a dynamic maximum link algorithm are proposed to meet the quick networking or minimum frame size requirements. In the simulation, the dynamic maximum link algorithm produces higher reductions in the frame length than the quick assign algorithm. When the number of routers is 140, the total number of time slots is reduced by 25%. However, the first algorithm needs more control messages, and the average difference in the number of control messages is 3 410. Consequently, the dynamic maximum link algorithm is utilized for adjusting the link schedule to the minimum delay with a relatively high throughput rate, and the quick assign algorithm is utilized for speeding up the networking process.展开更多
文摘为了进一步提高滚动轴承故障检测的准确性、改善时间卷积网络模型(Temporal Convolutional Network,TCN)存在的过拟合问题,本研究提出了增加平均池化层的时间卷积网络(Temporal Convolution with Average Pooling Network,TCAPN)模型。该方法首先使用膨胀因果卷积代替传统卷积神经网络,其次在残差模块多个地方加入平均池化层改善模型过拟合问题,最后结合多个改进残差模块构建本研究提出的TCAPN模型。实验结果表明,在相同工况条件下,TCAPN模型能够更快地收敛,并且平均故障诊断准确率达到了98.73%,相较于TCN模型提高了2.87%,验证了该模型具有高准确性和鲁棒性。
基金supported by Beijing Education and Scientific Research Programthe National High Technical Research and Development Program of China (863 Program) under Grant No. 2011AA040101+2 种基金the National Natural Science Foundation of China under Grants No. 61173150, No. 61003251Beijing Science and Technology Program under Grant No. Z111100054011078the State Scholarship Fund
文摘Industrial wireless sensor networks adopt a hierarchical structure with large numbers of sensors and routers. Time Division Multiple Access (TDMA) is regarded as an efficient method to reduce the probability of confliction. In the intra-cluster part, the random color selection method is effective in reducing the retry times in an application. In the inter-cluster part, a quick assign algorithm and a dynamic maximum link algorithm are proposed to meet the quick networking or minimum frame size requirements. In the simulation, the dynamic maximum link algorithm produces higher reductions in the frame length than the quick assign algorithm. When the number of routers is 140, the total number of time slots is reduced by 25%. However, the first algorithm needs more control messages, and the average difference in the number of control messages is 3 410. Consequently, the dynamic maximum link algorithm is utilized for adjusting the link schedule to the minimum delay with a relatively high throughput rate, and the quick assign algorithm is utilized for speeding up the networking process.
