| The production assembly of the aircraft is a highly complex and extremely important task.In order to ensure the progress of the assembly work and ensure the quality of the product,it must be assembled according to the text of the process operation instructions,but the preparation process is a complicated and cumbersome work.The process operation description has certain similarity and reusability,and it can form a typical and highly reusable template.The use of the template can improve the efficiency of process programming.Aiming at the problem of obtaining the working operation description template,this paper proposes a method based on the combination of entity recognition and cluster analysis to generate the template form of the process operation description.The method firstly identifies the process parameters in the text and generalizes them into a word slot form.Secondly,cluster the generalized texts;finally,extract standardized and normalized templates from them.First of all,according to the construction requirements of the process operation specification template,this paper defines 11 types of entities such as engineering drawings,reference standards and part models,with a total of 15990 sentences.Since the operating instructions in the aircraft assembly process are highly domain-oriented,they exhibit strong context dependence compared to entity recognition in the general field.Based on this,this paper uses a two-way long short-term memory(LSTM)neural network and a conditional random field(CRF)model combined with dictionary and rules to solve such problems.The experimental results show that the overall F1 value is 1.52% and 4.14% higher than the baseline method on the two experimental corpora.Secondly,because the process operation can be divided into 11 categories such as part positioning,inspection and installation,each category can be further divided into different sub-categories.Based on this,this paper proposes a two-stage clustering algorithm.In the first stage,this paper uses K-means clustering algorithm to perform rough division according to the above 11 categories.In the second stage,this paper uses hierarchical clustering algorithm to refine the process data objects in each category.Finally,under the guidance of professional craftsmen,97 and 284 process operation description templates were extracted from the two experimental corpora,and the requirements of process design were met. |
