インダストリアルエンジニアリングとマネジメント

Fault Diagnosis of Spinning Industrial Systems by Using Wavelet Analysis and Soft Computer Classifier

Abstract

Ahmed E. Amin Mohamed

Any industry needs an efficient predictive plan in order to optimize the production quality and improve the economy of the plant by reducing the defective product and produce final product fits its end-use.
In the textile industry, mass variation is an important property of yarn and is generally described by the diagram and spectrogram. The main parameter used for characterizing yarn evenness is the Coefficient of Variation (C.V %).
This parameter is based on the “cut and weight” method. Practice in the industry test length of 8 mm was used. This means that the C.V% corresponds to an electronic cut length of 8 millimeter. A diagram of the mass variation can also be produced to provide an overall profile of yarn irregularity.
For spun yarns, the evenness tester produces a spectrogram that covers a range of wavelengths from 2 cm to 1280 m. It assesses periodic mass variations, which occur at least 25 times as being statistically significant. Periodic variations are typically caused by mechanical defects (e.g. drafting faults).
This paper presented a new method based on multi-resolution analysis, for the classification of yarn mechanical defects diagnosis. It is constituted of two stages architecture: in the first stage a set of features are extracted from yarn signal by a wavelet analysis. The second stage is devoted to the classification of defect from the features by using Probabilistic Neural Network (PNN). Naïve Bayes algorithm and Bayes net algorithm is taken for classification and compared. The novelty of the proposed method resides in the ability not only with higher precision, but also with dimensionality reduction and higher speed than method of Fourier transform and mathematical statistics.