FAULT PREDICTION OF FASTENER PRODUCTION EQUIPMENT VIA LIGHTWEIGHT REGRESSION MODELING WITH SCIKIT-LEARN: EVIDENCE FROM THE HAIYAN FASTENER INDUSTRY

In fastener manufacturing, an unexpected breakdown of a cold heading machine stops the entire production line, yet most small and medium-sized enterprises (SMEs) still maintain their equipment by rule of thumb. Targeting the SMEs in the Haiyan fastener industry cluster, this paper takes cold heading machines as the object and builds a low-cost equipment fault prediction method based on regression modeling and Scikit-learn. The remaining useful life (RUL) of key components is taken as the prediction target; four models, namely linear regression, random forest, gradient boosting regression tree (GBRT), and support vector regression (SVR), are compared, and grid search is used for parameter tuning. Since such enterprises generally lack usable historical failure data, the method is validated on real degradation data from the public XJTU-SY accelerated life test dataset of rolling bearings, which share the same data form as the key rotating components of cold heading machines. The results show that GBRT performs best (RMSE = 6.25 min, R² = 0.511), that degradation information is concentrated in a few amplitude-type statistical features such as mean amplitude, standard deviation and root mean square, and that in the early-warning replay on two test bearings, alarms are all raised ahead of failure with an average lead time of 8.0 min and no false alarm. The study indicates that lightweight AI tools are sufficient to support equipment fault prediction in county-level manufacturing SMEs, and the whole procedure can be replicated directly.