Accurate anomaly detection is the premise of production process control and normal execution of production plan. The implementation of Internet of Things (IoT) provides data foundation and guarantee for real-time perception and detection of production state. Taking abundant IoT data as support, a density peak (DP)-weighted fuzzy C-means (WFCM) based clustering method is proposed to detect abnormal situations in production process. Firstly, a features correlation and redundancy measure method based on mutual information (MI) and conditional MI is proposed, unsupervised feature reduction is completed based on the principle of maximum correlation-minimum redundancy. Secondly, a DP-WFCM based clustering model is established to identify clusters with fewer samples to detect production anomalies. DP is used to obtain the initial clustering centers to solve the problem that FCM is sensitive to the initial centers and the clusters number needs to be determined manually in advance. MI-based similarities are introduced as weight coefficients to guide the clustering process, which improves convergence speed and clustering quality. Finally, a real case from an IoT enabled machining workshop is carried out to verify the accuracy and effectiveness of the proposed method in anomaly detection of manufacturing process.

准确的异常检测是生产过程控制和生产计划正常执行的前提。物联网（IoT）的实施为实时感知和检测生产状态提供了数据基础和保证。以物联网数据为支撑，提出一种基于密度峰值（DP）加权模糊C均值（WFCM）的聚类方法，以检测生产过程中的异常情况。首先，提出了一种基于互信息（MI）和条件MI的特征相关和冗余度量方法，基于最大相关最小冗余的原理完成了无监督的特征约简。其次，建立了基于DP-WFCM的聚类模型，以识别样本较少的聚类来检测生产异常。DP用于获取初始聚类中心，以解决FCM对初始聚类敏感的问题，并且需要事先手动确定聚类数。引入基于MI的相似性作为权重系数，以指导聚类过程，从而提高了收敛速度和聚类质量。最后，从支持IoT的机加工车间进行了实际案例验证了所提出方法在制造过程异常检测中的准确性和有效性。