In the identification of fatigue damage for critical load-bearing parts of automobiles and welded connection points in steel structures under random cyclic loads, a time lag exists. To address this, this paper proposes an intelligent industrial fatigue damage identification and early warning method that incorporates time-series feature analysis.First, a multi-source fatigue time-series dataset covering both automotive and steel structure scenarios is established. Information such as strain, vibration, acoustic emission, and load is uniformly converted into sliding-window samples and categorized into four damage stages according to the fatigue evolution process.Next, a joint model is developed, combining multi-scale time-series encoding, domain-consistent constraints, a damage identification head, and one risk regression head, for realizing the coordinate optimization of discrete stage recognition and continuous risk output. Based on this foundation, a smoothed early warning index and a continuous trigger mechanism are brought in, in order to change the model’s output result into a stable early warning signal which is suitable for being directly used in online monitoring. Experimental outcome indicate that the put forward method attains recognition accuracy of 97.8% and 96.9% in automobile and steel structure situation, one by one, with Macro-F1 scores of 0.972 and 0.964. It maintains good robustness under conditions of noise and missing channels, with an average early warning lead time of 14.1%.Case studies further demonstrate that the proposed method can detect the temporal precursors of microcracks from their initiation to the stable propagation stage at an early stage, showing significant potential for real-time deployment in industrial monitoring workflows.
