Research On Quality Intelligent Diagnosis For Automatic Machining Process Based On Pattern Recognition

Real-time intelligent process monitoring and diagnosis is an important part of the quality and safety supervision with big data of food and drug as well as a key part of the implementation development strategy of the intelligent manufacturing by government. Traditional statistical process control method has been difficult to meet the demands that real-time and intelligent quality control and diagnosis for the machining process. The intelligent monitoring and diagnosis method for machining process based on pattern recognition has become a new direction in the fields of quality management. At present, these researches in this area are mainly focused on the control chart pattern recognition and the abnormal pattern parameter estimation.Most of the existing recognition models based on single artificial neural networks or support vector machines have some problems that are computational complexity and low recognition accuracy. The existing parameter estimation models of abnormal patterns have this problem that is difficult to accurately estimate the parameter of the abnormal patterns with the subtle changing parameter. Therefore, there are urgent to solve these problems that how to construct a more efficient recognition model and the estimation model of the abnormal pattern parameter.In this paper, the intelligent monitoring and diagnosis method for automatic machining process is researched based on the National Natural Science Foundation project which is titled "Dynamic process quality monitoring and diagnostic based on pattern recognition". Firstly, according to the change characteristics of real-time measurement data flow for quality characteristics, the automatic machining process operation states are defined as quality patterns such as normal, trend, shift and so on.Secondly, the hybrid feature of process quality pattern is proposed and recognition model is established using artificial neural network and support vector machine. Then,an estimation model of the abnormal pattern parameter is established based on wavelet reconstruction and support vector regression with multiple kernel function.Finally, an online intelligent diagnosis framework is formed and the validity of the proposed framework is verified by an example application of precision axis automatic machining process. The specific research contents are as follows:(1) Research on quality pattern recognition model for automatic machining processThe hybrid feature of the process quality pattern is proposed that combined with low frequency approximation coefficients of wavelet decomposition and the shape features. Then, the quality pattern recognition model using neural network and support vector machine is proposed. The neural network based on mean feature is used to divide automated production process quality pattern which overall change characteristics has the similar into three categories including normal and cycle,upward, downward. Furtuer the support vector machine based on hybrid feature is used to divide three categories into specific process quality pattern such as normal,cycle and so on. Simulation results shows that the proposed hybrid feature not only the data dimension is low and the detail information is strong, which can effectively improve the discrimination of quality pattern. The proposed recognition model combined with the neural network and support vector machine through the mean feature is to greatly reduce the computational complexity.(2) Research on abnormal quality pattern parameter estimation model for automatic machining processThe abnormal quality pattern parameter estimation model based on wavelet reconstruction data feature and the support vector machine regression with multiple kernel function is proposed. Firstly, wavelet reconstruction data feature of abnormal process pattern is extracted to highlight the variation characteristics of abnormal process pattern which the changing of parameter is subtle. Then, multiple kernel support vector regression based on the linear kernel function and gauss kernel function is used to estimate the parameter of abnormal process pattern. Simulation results shows that the wavelet reconstruction data feature can enhance the discrimination of the abnormal process quality pattern that the parameters are changes slightly, and then improve the estimation performance of the parameter estimation model.(3)Research on online quality intelligent diagnosis framework for automatic machining processAn online intelligent diagnosis framework for automatic machining process by combining the recognition model, the abnormal pattern parameter estimation model and the expert diagnosis knowledge base is proposed. Firstly, the measured value of quality characteristics is online readed using the "monitoring window", and the recognition model is used to identify the quality pattern kind of the data flow in the current "monitoring window". When the quality pattern of the data flow in the "monitoring window" is abnormal, the abnormal pattern parameter estimation model is used to estimate the abnormal pattern parameter. Then, the process abnormal causes are found from the expert diagnosis knowledge base.(4)Case study on online intelligent monitoring and diagnosis for precision axis machining processThe validity of the proposed quality intelligent diagnostic framework is verified by the case study on online intelligent monitoring and diagnosis of precision axis machining process. Firstly, the quality model data of precision axis automatic machining process is simulated using Monte Carlo method and the recognition model and the parameters estimation model are trained and tested off-line. Then, the expert diagnostic knowledge base is constructed using expert experience knowledge and fault tree analysis method. Finally, the online intelligent monitoring and diagnosis of the actual precision axis machining process is carried out using quality intelligent diagnosis framework. Case study shows that the proposed quality intelligent diagnosis framework can be used in the online quality intelligent monitoring and diagnosis for automatic machining process.The characteristics and innovation of this study can be summarized as follows:(1)Because of existing fusion features are difficult to meet the demand that data dimension is low and pattern discrimination is strong, the hybrid feature combined with the approximation coefficients and shape feature is proposed. The proposed hybrid feature not only has low data dimension but also has strong ability of pattern discrimination.(2)Because of recognition models based on the single artificial neural network or support vector machine has complex structure and high computational complexity,quality pattern recognition model based on artificial neural network and support vector machine is proposed to improve the recognition efficiency of process quality model.(3) Because of the traditional parameter estimation models are difficult to estimate small changing parameter, support vector regression estimation model based on wavelet reconstruction and multiple kernel function is proposed. And then the quality intelligent diagnosis framework is proposed to provide an online intelligent diagnosis method for automatic machining process.The results of this study not only provides a set of operational quality intelligent monitoring and diagnosis methods for the automatic machining process, but aslo canbe extended to the government supervision fieldes such as monitoring and regulating of financial market, the quality and safety supervision of food and drug and so on.