| 摘 要: 针对口罩机在多工序生产中故障特征难以诊断的问题,提出了一种基于自组织映射(SOM)和误差反向传播网络(BP)的故障检测模型。首先针对4种减速机故障类型搭建SOM-BP复合型神经网络模型并完成检测分类,其次通过提取原振动信号的20组时域和频域参数作为SOM网络的输入样本进行初步聚类,并根据仿真结果确定最佳竞争层结构,最后将聚类后结果输入BP网络进行预测并完成分类,实现故障检测。研究结果表明,7×7竞争层结构下的SOM-BP复合型神经网络对于减速机的8种时域和频域参数的检测效果最优,分类准确率可达93.5%,173次迭代即可收敛,数据拟合度最高达0.998 76,达到实际检测要求,验证了该方案的有效性和可行性。 |
| 关键词: 口罩机;自组织映射;BP神经网络;故障检测 |
|
中图分类号: TP206.3
文献标识码: A
|
| 基金项目: 浙江省科技计划项目2022C01065 |
|
| Transmission System Fault Detection of Fully Automatic Mask Machine Based on SOM-BP |
|
PENG Laihu1,2, LIU Xudong1, WAN Changjiang1,2
|
(1.Zhejiang Sci-Tech University, Hangzhou 310000, China;
2.Longgang Research Institute, Zhejiang Sci-Tech University, Wenzhou 325000, China)
laihup@ zstu.edu.cn; 970960889@qq.com; wanchj@zstu.edu.cn
|
| Abstract: To address the difficulty in diagnosing fault characteristics in the multi-process production of mask machines, this paper proposes a fault detection model based on SOM-BP (SOM: Self-Organizing Map; BP: Error Back Propagation) network. Firstly, a SOM-BP composite neural network model is built for four types of gearbox faults to perform detection and classification. Secondly, 20 sets of time-domain and frequency-domain parameters extracted from the original vibration signals are used as input samples for the SOM network to conduct initial clustering. The optimal competitive layer structure is determined based on simulation results. Finally, the clustered results are input into the BP network for prediction and classification to achieve fault detection. Research results indicate that the SOM-BP composite neural network with the competitive layer structure shows the best detection performance for eight timedomain and frequency-domain parameters of the gearbox, achieving an accuracy rate of 93.5% . It converges after 173 iterations, with the highest data fitting degree of 0. 998 76, meeting the requirements of practical detection. This validates the effectiveness and feasibility of the proposed solution. |
| Keywords: mask machine; Self-Organizing Map; BP neural network; fault detection |
