基于深度神经网络水下清淤机器人绞龙的可靠性分析  

Reliability Analysis of Packing Auger of Desilting Robot Based on Deep Neural Networks

作　　者：邵可鑫 桑建兵[1] 田魏昌 石政加 袁国秩 SHAO Kexin;SANG Jianbing;TIAN Weichang;SHI Zhengjia;YUAN Guozhi(School of Mechanical Engineering,Hebei University of Technology,Tianjin 300401,China)

机构地区：[1]河北工业大学机械工程学院,天津300401

出　　处：《机械科学与技术》2024年第11期1894-1900,共7页Mechanical Science and Technology for Aerospace Engineering

基　　金：国家自然科学基金项目(12102123);河北省自然科学基金项目(A2020202015,A2021202014)。

摘　　要：绞龙作为水下清淤机器人的淤泥疏松装置,其作业时常与淤泥中的硬物发生碰撞,静应力强度校核并不能很好地反映碰撞过程中的瞬态动力响应问题,同时传统的可靠性分析方法在解决多源不确定性瞬态动力响应问题时往往存在效率低下且精度不高等问题。为了克服上述缺点,基于数值模拟、深度神经网络(DNN),与Monte Carlo(MC)方法提出一种用于绞龙与硬物碰撞问题的可靠性分析方法。首先,基于理论推导确定影响碰撞力大小的不确定性变量,并对其进行多参数敏感性分析得出最终深度神经网络的输入量;其次,利用拉丁超立方采样(LHS)根据深度神经网络的输入量分布情况进行采样,采用有限元软件ANSYS/LS-DYNA建立与采样数据相对应的绞龙与硬物碰撞有限元模型,并提取深度神经网络的输出量;最后,通过单轴拉伸试验确定绞龙的损伤准则,并结合深度神经网络与Monte Carlo方法 (DNN-MC)对绞龙的可靠度与失效概率进行了预测。结果表明本文方法精度远高于工程要求精度,与传统可靠性分析方法相比本文方法在保证高精度的前提下具有更高的效率和更好的鲁棒性。The packing auger of an underwater desilting robot often collides with hard objects in the sludge during operation.The static stress check cannot reflect the transient dynamic response in the collision process.At the same time,the traditional reliability analysis method is often inefficient and inaccurate in solving the problem of multi-source and uncertain transient dynamic response.A reliability analysis method for the collision between the winch and hard objects is proposed based on numerical simulation,deep neural networks(DNN)and the Monte Carlo(MC)method to overcome the shortcomings of the static stress check.Firstly,the uncertain variables affecting the impact force are determined based on theoretical derivation,and the final input of DNN is obtained with sensitivity analysis.Secondly,the Latin hypercube sampling(LHS)technique is used to sample according to the input distribution of each DNN,and the finite element model of the collision between the winch and hard objects corresponding to the sampling data is established by using the finite element software ANSYS/LS-DYNA to extract the output of DNN.Finally,the damage criteria of the packing auger are determined through experiments,and its reliability and failure probability are predicted by combining deep neural networks with the Monte Carlo method(DNN-MC).The results show that the accuracy of this method is much higher than the required engineering accuracy.Compared with the traditional reliability analysis method,this method has higher efficiency and better robustness to ensure high accuracy.

关 键 词：绞龙 可靠性分析 深度神经网络 拉丁超立方采样技术 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP242[自动化与计算机技术—控制科学与工程]