The batch duration in most batch units is quite long and the number of batch runs is very limited, so it is difficult to build accurate monitoring models. A powerful supervised functional monitoring method, called wavelet functional partial least squares, is proposed. First, an active learning strategy is used to extract features of process variables using orthogonal wavelet approximations and achieve a more concise model. Then the partial least squares method can be constructed using the extracted features and quality data, so the regression model is robust. Using the compact support property of wavelet functions, the process has multiple phases. The final quality can be expressed as a summation of multiple sub-qualities so multiple local models can be established for within-batch detection of both process data and quality data. The advantages and merits of the proposed method are demonstrated using a numerical case and an industrial sintering process for polytetrafluoroethylene.

大多数批处理单元的批处理持续时间很长，批处理运行的次数非常有限，因此很难建立准确的监控模型。提出了一种强大的监督函数监测方法，称为小波函数偏最小二乘法。首先，主动学习策略用于使用正交小波近似提取过程变量的特征并实现更简洁的模型。然后可以使用提取的特征和质量数据构建偏最小二乘法，因此回归模型是稳健的。利用小波函数的紧支持特性，该过程具有多个阶段。最终质量可以表示为多个子质量的总和，因此可以建立多个局部模型用于过程数据和质量数据的批次内检测。