动态数据流驱动的再制造拆解工艺知识图谱构建方法

Dynamic data flow-driven knowledge graph construction method for remanufacturing disassembly process

高效稳定的再制造拆解是实施规模化再制造的重要前提。然而,废旧零部件的内部结构、损伤特征等存在较大差异,需要对拆解工艺进行频繁调整,从而造成再制造拆解效率低、质量稳定性差,严重制约了再制造规模化效益。为此,提出一种从动态数据流中提取个性化拆解工艺知识的知识图谱构建方法,利用知识的高效更新与重用提升再制造拆解效率与质量。首先,分析再制造拆解工艺数据的动态时序性特点,以拆解工位为数据分析节点,建立动态数据流模型。其次,基于工位数据对拆解工艺知识进行分类,构建包含拆解过程知识、资源知识和特征知识的本体模型,并利用命名实体识别模型(ALBERT-BiLSTM-CRF)和自然语言处理方法自动抽取拆解工艺知识。进而,提出一种动态数据流驱动的图谱更新机制,实现知识图谱对拆解工艺调整的快速响应。最后,以某型号废旧动力电池包拆解为例,验证所提方法的有效性。

Efficient and stable remanufacturing disassembly is an important prerequisite for the implementation of large-scale remanufacturing.However,the large differences in the internal structure and damage characteristics of used parts require frequent adjustments to the disassembly process,which results in low efficiency and poor quality stability of remanufacturing disassembly,and severely restricts the benefits of remanufacturing scale-up.In this respect,a knowledge graph construction method was proposed to extract personalized disassembly process knowledge from dynamic data streams and to improve the efficiency and quality of remanufacturing disassembly using efficient updating and reuse of knowledge.The dynamic temporal characteristics of remanufacturing disassembly process data were analyzed,and the dynamic data flow model was established with the disassembly stations as the data analysis nodes.The disassembly process knowledge was classified based on the data of each workstation,the ontology model containing disassembly process knowledge,resource knowledge and feature knowledge were constructed,and the disassembly process knowledge was automatically extracted by using the ALBERT-BiLSTM-CRF model and natural language processing method.Further,a dynamic data flow-driven graph update mechanism was proposed to realize the fast response of knowledge graph to the adjustment of disassembly process.The effectiveness of the proposed method was verified by taking the dismantling of a certain type of used power battery pack as an example.