Abstract: The spinning process parameters are numerous and interrelated,which blurs the important information between the process parameters and the yarn quality indicators.The quality fluctuation of the spinning process caused by strong coupling of the spinning process parameters is difficult to detect and control.Aiming at this problem,an abnormal detection method for quality fluctuation of multi-association parameters in spinning process based on feature space transformation was proposed.The relationship between the parameters of the spinning process was analyzed,and the partial least squares method was used to eliminate the correlation between the parameters,and a feature space with orthogonality was obtained.Then the new feature space data was used to calculate the Local Outlier Factor (LOF) statistics and determine whether the spinning process was stable,locate the abnormal fluctuation area.On this basis,the abnormal fluctuation data after correlation analysis was used as the input of the deep belief network model to identify the type of abnormal fluctuations in the spinning process.The algorithm analysis was verified,and the result showed that the proposed model could improve the abnormal detection accuracy of multi-related parameters in the spinning process to 98.2%.

Key words: spinning process, quality fluctuation, abnormal detection, local outlier factor, deep belief network, feature space transformation

摘要： 纺纱过程参数众多且相互关联,模糊了过程参数与纱线质量指标间的重要信息,针对纺纱过程参数强耦合带来的纺纱过程质量波动难以检测与控制的问题,提出一种基于特征空间变换的纺纱过程多关联参数质量波动异常检测方法。首先,分析纺纱过程参数间的关联关系,采用偏最小二乘法(PLS)消除参数之间的相关性,获得具有正交性的特征空间。然后,利用新的特征空间数据计算局部离群因子(LOF)统计量,判定纺纱过程是否稳定,定位异常波动区域。最后将相关性分析后的异常波动数据作为深度置信网络(DBN)模型的输入,识别纺纱过程异常波动类型。通过算例进行验证,结果显示所提模型将纺纱过程多关联参数异常检测精度提高到98.2%。

关键词: 纺纱过程, 质量波动, 异常检测, 局部离群因子, 深度置信网络, 特征空间变换