A Federal Recursive RUL Prediction Approach for Inter-Client Information Complementarity for Missing Data

RUL (Remaining useful life, RUL) prediction is one of the most important methods to ensure the safe operation of mechanical equipment. When continuous time segments are missing from the monitoring data, the relevant information from the sequence itself is not enough for a single RUL client to learn an accurate bearing degradation trend model. Therefore, a federated recursive RUL prediction method for inter-client information complementarity (FedRC) for missing data is proposed. High RUL prediction accuracy is ensured by information complementation within inter-client segments and gradual recursion between intra-client segments. A joint central loss function-driven aggregation weight and local loss joint optimization strategy is designed to reasonably allocate the weights of the uploaded model parameters of each client, which further improves the RUL prediction accuracy. In the case of missing local continuous time data segments, the proposed method ensures the RUL prediction accuracy through a segment-by-segment complementary recursive joint strategy among clients.