Application Of Fault Diagnosis In Electric Spindle System Based On ICA And Improved ICA Methods

As one of the core components of CNC machine tools,the performance and operating status of Electric Spindle system have a direct impact on the precision machining equipment and product quality.To ensure the normal operation of the Electric Spindle and avoid significant economic losses and personnel accidents,the fault detection and diagnosis of Electric Spindle are paid closely attention.In this thesis,taking the Electric Spindle system as the research object and the bearing crack as an example,the data-driven multivariate statistical analysis are used for fault detection and fault diagnosis.Firstly,some attempts were tried to find out the location,type and form of the fault occurs based on the analysis of the basic structure,the working principle and the fault mechanism of the Electric Spindle.The vibration signal of Electrical Spindle was collected through the AIC9916FS equipment fault simulation diagnosis and analysis system and used as the sample data to achieve fault detection and diagnosis.Secondly,two kinds of multivariate statistical analysis methods,principal component analysis method and independent principal component analysis method,were analyzed in detail.And emphasize that ICA was a signal decomposition method for high-order statistics,which was based on the assumption that independent variables were independent of each other.The probability statistics of variables were used to deal with non-Gaussian variable data.ICA was one of effectively methods to solve the problem of blind source signal whose prior knowledge was unknown.However,the characteristics of the data preprocessed by the traditional ICA method were approximately the same and difficult to extract.The relative transformation theory based on Euclidean distance was introduced and ICA fault detection method based on relative transformation(RTICA,Relative Transformation ICA)was proposed.RTICA transforms the original spatial data into relative space as new data by calculating the Euclidean distance between the sampled data.Then,ICA processing was carried out in relative space to reduce the dimension of relative space.The independent principal components with the main information were selected,and the kernel density estimation method was used to calculate the decomposition matrix and the control limit of the three statistics(I2,Ie2,SPE)in order to establish the fault monitoring model.By calculating 'the three statistic values of the newly acquired data and comparing with the control limit,the data greater than the control limit were considered as fault data and the online monitoring was realized finally.The results show that RTICA was the best method to detect the crack in the bearing of the Electric Spindle by comparing with ICA and PC A.It is verified that the RTICA method has better fault detection effect.Finally,MBRTICA,which is a combination of multi-block processing and RTICA method,was proposed to achieve the purpose of fault traceability.In the off-line modeling phase,the pre-processed training data were divided into some subblock units according to the position of the device,and these subblock units were processed by the relative transformation and ICA method respectively to calculate the decomposition matrix and the control limit of the three statistics.During the online diagnosis phase,the new data were still divided into multiple subblock units,and the statistics of each subblock unit were calculated.The purpose of fault traceability was realized by comparing the statistics and control limits to detect which subblock unit the fault occurred,the experimental results of the crack in the bearing of the Electric Spindle showed that the method was effective and feasible.