Research On Evaluation And Prediction Of Tubing Coupling Roundness Error Based On Neurual Network

Tubing coupling is important oil steel, which is used to joint the tubing. Oil is usually transferred by high-pressure in the tubing, in order to ensure that the oil is not leakage, the inner wall of the tubing coupling has high roundness requirement. Most of high-precision, high-strength tubing couplings are imported, rare is domestic production. Some domestic manufacturers which can produce such tubing couplings are trouble with serious problem of coupling roundness tolerance which seriously affects the efficiency and results in a huge waste of human and material resources. This paper carries out technology research to improve the quality of inner tubing coupling. A neural network prediction model of roundness error is applied to provide selection and optimization of processing parameters in order to improve the rate of qualified products. This paper has made researches around the key issue as follows:1. Through the analysis of the principle of BP neural network algorithm, found that the algorithm has some shortcomings, such as slowly convergence speed , easy enter into local minimum point and the number of hidden nodes is mostly empirically determined. In order improve these shortcomings, learning rate adaptation is introduced to increase convergence speed, genetic algorithm is introduced into determine initial parameters value of network and grey correlation analysis is introduced to determine of the hidden units number.2. Based on the similarity of least mean square algorithm in the linear neural network and least square method in the roundness error evaluation, the linear neural network is introduced in tubing coupling inner wall roundness error evaluation. Through the comparison with algorithms and measurement methods indicates that this method can correctly and effectively evaluate roundness error.3. In order to get data of processed couplings inner wall roundness error and processing parameters, orthogonal processing and tubing coupling inner wall roundness error detection experiment are designed. Combined with the roundness instrument, linear neural network is introduced to process roundness error, and achieved very good results.4. Establish a prediction model of tubing coupling inner wall roundness error with improved BP neural network. Through the comparison with tradition BP neural network predict model and measurements, the result so that the improved BP neural network has fast convergence rate and strong generalization capability, which can predict the maximum the roundness error of tubing coupling accurately.