Abstract: As the goal of Smart Manufacturing 2025 approaches,enterprises are joining the ranks of Smart Manufacturing Maturity Assessment to understand their own capability level.However,due to the complexity of the smart manufacturing maturity assessment criteria,enterprises lack their understanding of their level in the industry,which leads them to apply hastily,wasting their own time while taking up a lot of assessment resources.A new assessment process was designed,and the whole assessment process was reconstructed using text processing algorithms.By utilizing the intelligent manufacturing maturity assessment standards in the national standard documents as a training set,an intelligent assessment algorithm based on the combination of pre-trained language model BERT and Text Convolutional Neural Networks (BERT+TextCNN) was used instead of manual assessment.Validation on a real enterprise smart manufacturing dataset showed that the BERT+TextCNN assessment model achieved an accuracy of 85.32% when it assessed smart manufacturing maturity with the convolution kernel of ［2,3,4］,six iterations and the learning rate of 3e-5.This indicated that the proposed method could help enterprises complete the self-assessment of smart manufacturing maturity more accurately,which helped enterprises understand their own smart manufacturing capability level and formulate the correct development direction.

Key words: capability maturity smart manufacturing model, BERT pre-trained language model, text convolutional neural networks, reconfiguration of assessment process

摘要： 随着智能制造2025目标的临近,企业为了解自身能力水平纷纷加入到智能制造成熟度评估的行列中。然而,由于智能制造成熟度评估标准的复杂性,企业缺乏其对行业水平的了解,导致企业贸然申请,浪费自身时间的同时又占用大量评估资源。鉴于此,设计了一种新的评估流程,采用文本处理算法对整个评估过程进行了重构,通过利用国标文件中智能制造成熟度评估标准,将其作为训练集,采用基于预训练语言模型与文本神经网络(BERT+TextCNN)相结合的智能评估算法代替人工评估。在真实的企业智能制造数据集上的验证表明,当BERT+TextCNN评估模型在卷积核为［2,3,4］、迭代次数为6次、学习率为3e-5时,对智能制造成熟度进行评估,准确率达到85.32%。这表明所设计的评估方法能够较准确地帮助企业完成智能制造成熟度自评估,有助于企业了解自身智能制造能力水平,制定正确的发展方向。

关键词: 智能制造成熟度模型, BERT预训练语言模型, 文本卷积神经网络, 评估过程重构