Study On Intelligent Spindle Autonomous Sensing Method Based On Situational Perception With Deep Learning

As the most basic production and processing unit of modern manufacturing industry,machine tools directly determine the quality and processing efficiency of a product in response to the complex and dynamic manufacturing environment.This depends on the machine's intelligent ability in perception of environmental information,optimization of processing conditions,and prediction of health conditions,etc.Among them,the spindle system is the most direct executor of machine tool processing,and its intelligent level in the aspects of autonomous sensing function,execution optimization function and reasoning decision function directly determines the processing cost,processing accuracy and surface machining quality of the components.This paper conducts a series of research work on the autonomous sensing function of the intelligent spindle system,trying to solve the problem of large scale monitoring,multiple monitoring sites,and high sampling frequency at each site in intelligent spindle sensing era,this makes the traditional fault diagnosis technology faceless when extract more sensitive features of fault types in these big data.Achieve accurate and efficient state self-diagnosis and fault self-monitoring goals for intelligent spindle systems in complex environments.Providing an effective theoretical method for autonomous sensing of intelligent machines.The main research contents include:(1)From the point of view of the manufacturing system and its complexity,the deficiencies of the traditional processing machine tools in the complex manufacturing environment,and the solutions proposed by the intelligent spindle system in response to these deficiencies,the description and modelling of the autonomous sensing problem of the intelligent spindle system are described.The definition and main characteristics of the intelligent machining machine are given.The main advantages of the intelligent processing machine for adapting to complex environments and solving complex problems are analyzed.From this,the definition and general model of the intelligent spindle system are inferred,and the autonomous sensing of the intelligent spindle system is concluded and described;(2)Because single sensor information has the defects of single characterizing ability,high contingency,and being vulnerable to external environmental factors,the use of single sensor information to solve the problem of autonomous sensing in complex environments has great limitations.This paper presents a multi-sensor multimodal information fusion method,through multi-sensor acquisition of the spindle system,feature extraction of the collected signals,and then using the traditional machine learning algorithm for processing timing information-Gaussian mixture Hidden Markov model(GMM-HMM)to identify the processing state and fault type of the spindle of the machine tool,and using the D-S evidence theory to fuse information of multiple physical domains to improve diagnostic efficiency.At the same time,combining long-term memory network(LSTM),a deep learning method,provides advantages in timing information modeling,automatic extraction of feature information,and processing of high-dimensional and complex nonlinear data,proposing a multi-physics domain information modeling method based on LSTM and used it in the process of condition monitoring of machine tool spindles to highlight the outstanding effects on feature extraction and pattern recognition of deep learning.Finally combined with situational awareness theory,Bayesian statistical method is used to establish the information fusion model between traditional machine learning and modern deep learning at the decision-making level.Multi-sensor information fusion is realized to further improve the accuracy of perception.(3)To test and verify the validity of the proposed method,in this paper,the design,development and simulation experiments of the simulation experiment system are carried out on the existing bearing data sets and on the spindle of the CTC-650 CNC 4-axis CNC drilling and milling machining center.The diagnostic accuracy of using GMM-HMM on existing datasets and test datasets was 98% and 97% respectively.The diagnostic accuracy of using LSTM was 98.3% and 97.51% respectively,and the use of Bayesian theory improved the diagnosis.Confidence,experimental results show that the proposed method has certain theoretical and practical significance.