Transmission Fault Diagnosis Of 3D Printers Based On Machine Learning

Owing to the advantages of personalization,complexity and rapidity,3D pringting has been gradually applied in various fields.3D printer is an important research object in the development of 3D printing technology,but the research of 3D printer in the field of intelligent fault diagnosis is not perfect enough.Aiming at the faults of the transmission system of Delta 3D printer,this paper studies the fault diagnosis and fault degree evaluation via employing attitude sensor in a data-drive way.In this study,a fault information acquisition method is proposed for the end of the transmission system,that is,the attitude sensor is installed on the moving platform.Combined with machine learning model,the fault recognition model and fault degree evaluation model of Delta 3D printer are established from shallow learning to deep learning.The main research work in this paper is as follows.(1)By analyzing the structure and working principle of Delta 3D printer,the common faults are obtained.Aiming at the high accuracy and small samples,a fault diagnosis method based on support vector machine and industrial attitude sensor is proposed.This method installs industrial attitude sensors on the moving platform of Delta 3D printer to collect fault data.Then LS-SVM fault recognition model is established by using these fault data.The average accuracy of the LS-SVM model is 94.44%.In addition,the BPNN fault recognition model is established by using the same data.The results show that the accuracy or stability of the BPNN model is lower than that of LS-SVM model.(2)In order to reduce the cost of fault pattern recognition for 3D printer,a fault pattern recognition model based on low cost attitude sensor is established.To solve the problems of low precision and high noise caused by low cost attitude sensors,a fault recognition method based on ESN and feature extraction is adopted.A 3D printer fault identification model based on ESN is established after collecting fault data with low cost attitude sensor.The model can identify more types of faults and the accuracy of fault identification is 97.17%.At the same time,the fault identification models of SVM,LPPSVM and PCASVM are established by using the fault data of low cost sensors,and the recognition accuracy is lower than ESN model.(3)After accomplishing the task of the fault recognition of 3D printer,the fault degree method of 3D printer is also studied.This method is a deep learning method based on SAE.In the evaluation of fault degree,two cases(i.e.joint bearing a and joint bearing b)with different fault degree information were collected by using low-cost attitude sensor.Normalization and extraction of statistical features in time domain and feature extraction in time-frequency domain are used to reduce data dimension,reduce calculation time of SAE model and improve model accuracy.Taking SAE hidden layer as sample matrix,the trend curves of fault degree in case 1 and case 2 are fitted by calculating the matrix similarity of normal sample matrix and sample matrix with different fault degree respectively.