Textile Spinning-Frame Roller Fault Diagnosis Based On Wavelet Analysis And Support Vector Machine

The fault diagnosis technology of spinning-frame equipment has received extensive attention in modern textile production. If there is a fault in a spinning-frame and the spinning-frame operates with the fault, not only the device will be damaged, but also life-threatening event may occur. Therefore, timely detection of the fault of the spinning-frame is essential. Traditional spinning fault detection method is to touch the device by hand to identify whether the temperature is too high or whether the vibration is too large, and to hear by ear whether the moving parts have abnormal sounds. However, estimating, the fault according to a person's feeling and experience suffers many limitations. Therefore, automatic fault diagnosis of textile spinning frame has a practical significance.According to the four common fault types, this thesis uses an improved wavelet transform and support vector machines to extract the fault information and classify the fault type, and this fault problem from three aspects, including signal acquisition, feature extraction, and fault classification.First, this thesis designs a data acquisition system based on Labview to collect the vibration signals of spinning frame. This data acquisition system changes vibration signals into voltage signals by the electric eddy sensor. The voltage signals are changed into digital signals by the acquisition card and then transfered to the computer. The vibration signals are saved by 32-bit float-point data files in computer.Then, this thesis briefly describes the experimental simulation of these common fault types. According to the data have been obtained by the data acquisition system, this thesis uses the time-domain analysis to combine the wavelet transform with the power spectral analysis to extract the feature parameters. The traditional wavelet transform can not extract the feature parameters, so this thesis uses an improved wavelet decomposition and reconstruction algorithm to extract the feature parameters. This improved wavelet algorithm combines wavelet transform and FFT analysis to reduce the frequency overlapping phenomenon.Finally, as the number of available fault samples is small, this thesis selects SVM to identify the fault. The feature parameters are extracted as the input vectors, and this thesis uses one-against-one and one-against-rest methods to train and identify the feature parameters. Trial and error method and cross validation method are used to optimize the parameters of SVM. Then, this thesis compares the SVM with BP neural network. The result shows that SVM is not only fast for diagnosis, but also of high accuracy. So SVM is more suitable for the fault diagnosis of textile spinning machines.