关键词: 高质量过程, 融合特征, IFCM聚类算法, 自适应神经模糊推理系统, 故障诊断

Abstract: The deep application of intelligent manufacturing technology has promoted industrial production into a high-quality production stage with high stability, high precision, and high efficiency as its core characteristics. In this context, any minor anomalous fluctuations in the production process, if not detected in a timely manner, can lead to severe ramifications, thereby imposing more stringent standards on the accuracy and timeliness of fault diagnosis. However, the high-quality process itself causes the equipment to operate in a normal state for a long time, and fault data is extremely scarce. At the same time, the cost of manual labeling is expensive, resulting in a large amount of unlabeled data that cannot be effectively utilized, which seriously limits the application of traditional supervised learning fault diagnosis methods that rely on abundant labeled samples. In addition, existing intelligent fault diagnosis models have a "black box" problem, which can only output fault recognition results and lack the interpretability of the decision-making process. To address the issue of fault diagnosis with unlabeled data in high-quality production processes, this paper proposes a fault diagnosis method based on an Improved Fuzzy C-Means (IFCM) clustering algorithm and an Adaptive Neuro-Fuzzy Inference System (ANFIS) to achieve efficient and accurate fault classification. First, by extracting statistical features, frequency-domain features, and trend ordinal pattern (TOP) features from the raw data, a multi-dimensional feature space is constructed to comprehensively reflect the global information and local fluctuations of the data. Second, the IFCM algorithm is used to cluster the unlabeled data, and the number of clusters is adaptively determined based on the silhouette coefficient. Finally, the results derived from the IFCM algorithm are utilized as inputs for the ANFIS model, thereby achieving interpretable fault-type diagnosis via its established rule base. Experiments are conducted based on simulated control chart data and a hydraulic system condition monitoring benchmark dataset developed by the Industry 4.0 Laboratory of Paderborn University in Germany. The experimental results show that the IFCM-ANFIS model proposed in this paper exhibits significantly better classification capabilities than the comparison methods on various evaluation indicators. In the hydraulic system experimental dataset, compared with the average F1-score of other unsupervised learning methods, the average F1-score of the method in this paper is improved by 28.45% compared with the AE method, 21.77% compared with VAE, 11.73% compared with CL, 1.26% compared with the iForest method, and 2.06% compared with LOF, and the average running time is 103.9965 seconds. The IFCM-ANFIS model not only has a shorter overall time consumption but also maintains a high fault identification accuracy. The efficient performance of the IFCM-ANFIS algorithm in this paper also verifies its feasibility in industrial deployment. It provides an efficient and feasible solution for equipment fault diagnosis in intelligent manufacturing scenarios, which can monitor and diagnose faults in real time in actual industrial systems, thereby improving system reliability and maintenance efficiency.

Key words: high-quality process, integrated features, improved fuzzy C mean clustering algorithm, adaptive neuro-fuzzy inference system, fault diagnosis