中文摘要：

为提高自相关过程的统计过程控制方法的灵敏度与可靠性，提出利用Hopfield网络来检测自相关过程的均值渐进型漂移。首先将质量特性观测值分解为原形与背景噪声，通过动态编码将原形存储于网络；而后采用“相对增加”和“大于均值”原则对观测值编码，再利用网络的联想学习功能滤去背景噪声，提取原形，并判断均值是否发生渐进型漂移。研究表明，所提方法适用于具有不同参数的自相关过程，既无需过程统计模型，也无需大量的历史样本进行权值训练，具有较高的灵敏度与可靠性。

英文摘要：

To improve the sensitivity and reliability of Statistical Process Control （SPC） methods for autocorrelation processes, a Hopfield Neural Networks （HNN） was used to detect continuous mean shift of autocorrelation processes. Firstly, observation values of quality characteristics were decomposed into original patterns and background noises. Then, the original patterns were stored into HNN through dynamic encoding. After that, the observation values were encoded by using the principles of ＇ relative increase＇ and ＇ beyond mean＇, so that the background noises were filtered out by associative learning and thereby the original patterns were retrieved. As a result, the continuous shifts were detected. Research showed that the proposed method was applicable to processes with various autocorrelation parameters. It needed neither statistical model of processes nor large historical sample for network weights training. All of these manifested high sensitivity and reliability of the proposed method.