Research On Nonlinear Dynamics Of FDM 3D Printing

3D printing technology is used in many fields.According to the principle of molding,it can be divided into Laminated Object Manufacturing(LOM),Fused Deposition Molding(FDM),Three-dimensional Printing and gluing(3DP)and Direct Metal Laser Sintering(DMLS)and other technologies,of which the most widely used is FDM printing technology.FDM 3D printers often generate some non-linear and non-smooth problems during the printing process.Such as print delamination,nozzle scratching and drawing,and there is relatively little research related to this.At the same time,most of the FDM 3D printers do not provide alarms for printing failures and indicate specific problems.Therefore,it is necessary to build a real-time printing monitoring system to capture and classify the signals of different printing situations.The overall structure of this thesis is shown as follows:(1)By building a 3D printing platform and installing a signal acquisition device to capture signals for normal conditions,nozzle scraping,print delamination and drawing conditions that occur during the printing process of the printed workpiece.The 3D printing monitoring platform includes an FDM printer,accelerometers,a dynamic signal acquisition system(DAQ),and a personal computer(PC)in order to capture the smoothest signal and to determine the location of the sensors.The digital signals corresponding to different states during the printing process were captured using the sensors and DAQ system,which helped in the later analysis of the signals corresponding to different printing states with data.(2)During the printing process,it is divided into two categories: printing normal and printing fault.The different signals obtained are preprocessed separately and decomposed into wavelet packets.The number of decomposition layers is set to 3～5 layers,and the recurrence plot is drawn for each decomposed wavelet,and the recurrence plot is used for feature extraction,where the features include RR,DET,LMAX,ENT,LAM and TT.These several characteristic values obtained by computation are embedded in the artificial neural network algorithm(ANN)and the K-Nearest Neighbor(KNN)algorithm and support vector machine(SVM)to build the classification model,and then trained and predicted using the K-fold cross-validation method.The specificity,sensitivity and accuracy were used as classification judging indexes,which finally yielded 97.5% sensitivity,93.1% specificity and 96.1% correct rate under the ANN algorithm.100% sensitivity,83.5% specificity and95.8% correct rate under the KNN algorithm.97.33% sensitivity,95.71% specificity and 96.97% correct rate under the KNN algorithm.(3)Further,the same binary classification as previously described is used to perform multiple classification of printing situations using ANN algorithm and KNN algorithm.The printing process states are classified into four categories: printing normal,nozzle scratching,printing delamination and part drawing.The wavelet packet decomposition is performed and then the recurrence plot of wavelet subsequence is drawn.The corresponding eigenvalues are derived from the recurrence plot,and the ANN algorithm is used for classification and prediction.Finally,the sensitivity of the printing normal,printing drawing,printhead scratching and printing delamination cases are about 96.3%,96.4%,99.6% and 98.9%,respectively,the specificity is about 89.3%,89.2%,98.3%,and 97.5%,the correct rate is 92.7%.The KNN algorithm yielded an average sensitivity of higher than 95% for printing normal,printing drawing,printhead scratching and printing delamination cases,the specificity is about 86.1%,86.8%,92.2% and 98.3%,respectively,and the correct rate is 90.8%.(4)The previously extracted signals are analyzed to extract the relevant nonlinear information and thus evaluate the print quality.The signals captured by the sensors are used to calculate their corresponding recurrence plots,and then a convolutional neural network is used to perform deep learning of the recurrence plots with and without thresholds for different printing cases.Experiments are conducted by the K-fold cross-validation method,and the experimental results show that the deep learning method based on recurrence plots has a high correct rate and is able to evaluate the quality of FDM 3D printing products.The experimental analysis conducted above yielded a better classification effect in the printing process,which can help in the quality of the printing products and thus promote the market development of FDM printing products.