深度神经网络是一种具有复杂结构和多个非线性处理单元的模型,目前也已逐步被应用在工业生产过程中。但由于神经网络不可解释,不可控制的"黑箱"问题,以及海量的数据需求问题,使得深度学习在工业领域的应用仍有巨大的障碍。提...深度神经网络是一种具有复杂结构和多个非线性处理单元的模型,目前也已逐步被应用在工业生产过程中。但由于神经网络不可解释,不可控制的"黑箱"问题,以及海量的数据需求问题,使得深度学习在工业领域的应用仍有巨大的障碍。提出一种新的深度神经网络模型:知识深度置信网络(Knowledge-based deep belief network,KBDBN)。这种逻辑符号语言与深度神经网络的结合,不仅使得模型具有良好的模式识别性能,还可自适应地确定网络模型并具有可解释和可视化特性。进一步提出基于KBDBN的工件表面粗糙度加工过程的预测模型,实现了精确预测且有效地提取了制造过程的关键知识。试验结果证明:相较于传统机器学习器,KBDBN的网络性能更加优越,具有可解释性,可应用性更强。创新性的将符号规则与深度学习相结合并建立加工粗糙度预测模型,可以在精准预测的前提下提取工艺知识,指导加工工艺优化。展开更多
The top relief surfaces of an hourglass worm gear hob are ground manually in the traditional manufacturing process,which cannot ensure the width of the land surfaces of the hob.Moreover,each geometric feature of the h...The top relief surfaces of an hourglass worm gear hob are ground manually in the traditional manufacturing process,which cannot ensure the width of the land surfaces of the hob.Moreover,each geometric feature of the hob has been produced through different manufacturing techniques and machine tools,which results in low efficiency.To solve this problem,we propose a semi-automatic computer aided design(CAD)method for hobs.The point clouds of each feature surface of a hob are calculated by combing mathematical equations of the top relief surfaces built by the proposed method with other existing equations of hob surfaces.According to the point clouds,the method can achieve the automatic modeling for the hob in three-dimensional(3D)software by classifying and extracting the parameter information of the feature-hierarchical knowledge of the hob.Based on the generated 3D model,the entire surfaces of the hob can be manufactured on a four-axis computer numerical control(CNC)milling machine through only twice clamping.Verification of the width of the land surface of the hob manufactured by semi-automatic CAD method on a measuring projector proved the precision of the designed width can be ensured.The edge of the contact area on the worm wheel in a meshing experiment is clear and distinct,which means the worm gear drive is meshed well and the hob manufactured by the proposed method has improved machinability.The method simplifies the processing technique,and improves the design efficiency and production accuracy.展开更多
文摘深度神经网络是一种具有复杂结构和多个非线性处理单元的模型,目前也已逐步被应用在工业生产过程中。但由于神经网络不可解释,不可控制的"黑箱"问题,以及海量的数据需求问题,使得深度学习在工业领域的应用仍有巨大的障碍。提出一种新的深度神经网络模型:知识深度置信网络(Knowledge-based deep belief network,KBDBN)。这种逻辑符号语言与深度神经网络的结合,不仅使得模型具有良好的模式识别性能,还可自适应地确定网络模型并具有可解释和可视化特性。进一步提出基于KBDBN的工件表面粗糙度加工过程的预测模型,实现了精确预测且有效地提取了制造过程的关键知识。试验结果证明:相较于传统机器学习器,KBDBN的网络性能更加优越,具有可解释性,可应用性更强。创新性的将符号规则与深度学习相结合并建立加工粗糙度预测模型,可以在精准预测的前提下提取工艺知识,指导加工工艺优化。
基金supported by The Development of Design and Manufacturing Technology of Double-Enveloping Worm Pair in High-precision with Little Centre Distance(No.69190135)We would like to thank the Professor Workstation of China Agricultural University in Hebei Baoding Laibo Transmission Machinery Manufacturing Co.LTD(No.20200901).
文摘The top relief surfaces of an hourglass worm gear hob are ground manually in the traditional manufacturing process,which cannot ensure the width of the land surfaces of the hob.Moreover,each geometric feature of the hob has been produced through different manufacturing techniques and machine tools,which results in low efficiency.To solve this problem,we propose a semi-automatic computer aided design(CAD)method for hobs.The point clouds of each feature surface of a hob are calculated by combing mathematical equations of the top relief surfaces built by the proposed method with other existing equations of hob surfaces.According to the point clouds,the method can achieve the automatic modeling for the hob in three-dimensional(3D)software by classifying and extracting the parameter information of the feature-hierarchical knowledge of the hob.Based on the generated 3D model,the entire surfaces of the hob can be manufactured on a four-axis computer numerical control(CNC)milling machine through only twice clamping.Verification of the width of the land surface of the hob manufactured by semi-automatic CAD method on a measuring projector proved the precision of the designed width can be ensured.The edge of the contact area on the worm wheel in a meshing experiment is clear and distinct,which means the worm gear drive is meshed well and the hob manufactured by the proposed method has improved machinability.The method simplifies the processing technique,and improves the design efficiency and production accuracy.
