Research On Tool Wear Condition Recognition Based On Chaos Theory

In metal cutting process, surface quality and dimensional accuracy of the workpiece is affec ted by cutting-tool wear condition. So it is important to study the cutting-tool wear, especially in automation production. Cutting-tool wear is a complex process, it affected by various factors like cutting parameters, material characteristics and cutting environment, etc. Metal cutting is a nonli near system, there are a lot of non-stationary signals used in condition monitoring and fault diagn osis. Vibration, force and acoustic emission(AE) are the typical signal type widely used in cuttin g-tool wear research. In this paper, AE signal is chosen to be the carrier in analyzing cutting-tool wear. AE is the class of phenomena whereby transient elastic waves generated by the rapid releas e of energy when the materials are distorted or under the outside load. The AE signal produced b y cutting-tool wear is high-frequency and the bandwidth is nearly 50kHz~1MHz, so it can weak en the influence of low-frequency noise like mechanical noise and ambient noise. Experiment for data acquisition is design based on orthogonal experiment rule.The measured signal sometimes contains some high-frequency noise. In this paper, chaos theory is adopted in analyzing the nonli near characteristics of the AE signal. Chaos theory is sensitive to noise, so performing noise redu ction is done with the method based on EMD-Wavelet before computing. EMD is short for empir ical mode decomposition, the signal can be decomposed into several intrinsic mode functions wh ich is from high-frequency to low-frequency, then it is to determine the noise dominated intrinsic mode functions based on consecutive mean square error(CMSE) proposed by Boudraa and restr ain them. A new signal can be reconstructed by adding the rest intrinsic mode functions together and a further and last de-noising is using wavelet to processing the new one in order to get more pure signal. Before extracting the chaotic character, an important step is to reconstruct a phase sp ace from the de-noised signal. To get the phase space vector, two key parameters delay time and embedding dimension have to be determined. Method based on mutual-information is utilized in computing delay time and Cao method for embedding dimension. After reconstructing the phase space, the chaos attractors of lorenz system and real AE signal are presented which can obviousl y reflect the cutting-tool wear condition. The structure of the attractor differs with tool wear.Through chaotic time series analysis, the chaos features : correlation dimension, Lyapunov exponent and Kolmogorov entropy are extracted. These features prove that the tool-wear AE sig nals are chaotic. A result show that the three characters have a certain relevance and trend with t ool wear loss. Then a eigenvector matrix is set up mainly contains delay time, embedding dimens ion, correlation dimension, Lyapunov exponent and Kolmogorov entropy. Input the matrix to SV M for learning will get a model by which the test samples have a over 90% accuracy recognition.The result shows that it is effectively to combine chaos theory and SVM in tool wear condition r ecognition and wear loss prediction.