Research On Dynamic Production Process Perturbation Identification Method Of Smart Shop Based On Digital Twin

With the advancement of manufacturing intelligence,the reliability and stability of the production process on the shop floor,as the carrier of manufacturing,is critical.Failure to accurately and quickly identify the perturbations occurring in the workshop production process may lead to serious consequences.At present,there have been some theoretical and practical explorations of workshop perturbation identification methods based on information physical systems,machine learning and other technologies,but there are still problems such as low identification accuracy,slow identification speed and small breadth of obtainable data.The emergence of digital twin technology provides new ideas to solve the above problems.By constructing a workshop digital twin model and using model simulation to obtain data that are difficult to be collected by perception devices,the breadth and depth of data can be improved.Based on the above concept,this paper proposes a digital twin-based dynamic perturbation identification method for intelligent workshops,with the following main research:(1)The workshop digital twin model construction method and workshop perturbation feature library construction method are studied for the problems of small breadth of data available in the workshop production process,difficulty in acquiring some data,and difficulty in matching perturbation types and perturbation features.The discrete cutting shop digital twin model is constructed using Simscape language and subdivided into production line layer digital twin model,production unit layer digital twin model and equipment layer digital twin model;the shop perturbation events are classified according to equipment layer,production unit layer and production line layer,the perturbation feature representation is designed,and a three-layer structure(perturbation location-perturbation type-perturbation(1)The workshop perturbation feature library(perturbation location-perturbation type-perturbation feature)is designed.(2)For the problem of large amount of data in the workshop production process and the difficulty of extracting workshop production process state features,the workshop state features extraction method is studied.A Hadoop distributed platform is built to store the massive operation data of the workshop with multiple heterogeneous sources;feature extraction is carried out for the workshop digital twin model;after classifying and pre-processing the workshop operation data according to data types,multi-domain feature extraction is carried out in time domain,frequency domain and time-frequency domain.(3)The current perturbation recognition methods have problems such as slow recognition speed,low accuracy rate and poor generalization.In order to solve the above problems,the workshop perturbation recognition criterion is designed,and the workshop perturbation recognition method is studied by combining the characteristics of the periodic change of the operation state at the same location of the workshop.According to the size of the operating state change cycle,high frequency perturbation identification method and low frequency perturbation identification method are designed.For high-frequency perturbations,the structure of various types of commonly used convolutional neural networks and their respective advantages and disadvantages are studied,and the design method of convolutional neural network structure is given and validated with cases;for low-frequency perturbations,the Mealy time automaton perturbation recognition model is designed,and the time automaton model is validated using the UPPAAL simulation environment.Finally,the perturbation recognition method proposed in this paper is verified.For the high-frequency perturbation recognition method,the CNC milling machine tool is selected as the validation object and the tool wear perturbation recognition experimental bench is built to extract the state features from the tool machining data,input the feature matrix into the convolutional neural network perturbation recognition model,and then realize the tool wear perturbation recognition,and update the recognition results to the workshop digital twin model to prove the effectiveness of the method;for the low-frequency perturbation recognition method,the For the low-frequency perturbation recognition method,the motor shaft is selected as the validation object,and the actual processing process and processing time information of the motor shaft are collected,and the Mealy time automaton perturbation recognition model is constructed,and the correctness of the model is verified using the UPPAAL simulation environment,which proves the validity of the method.