Abstract: Accurately predicting the remaining useful life of rolling bearings is of great significance for ensuring the safe operation of mechanical systems and formulating maintenance strategies.However,in practical industrial applications,due to changes in working conditions and interference from environmental noise,it is difficult to extract useful features from the collected signals.In addition,there are issues such as low accuracy of the First Prediction Time (FPT) model and overly simplistic trend analysis models.These factors make the high-precision prediction of RUL for mechanical equipment extremely challenging.Therefore,a new method for predicting the remaining useful life of bearings based on multi-model fusion was proposed.A health indicator model combining an improved Genetic Algorithm-Deep Forest (GADF) and a FPT model using a Self-Attention-Auto Encoder (SAAE) was constructed.Based on the FPT results,a particle filter model was built to analyze the trend of the health indicator and finally calculate the remaining useful life of the mechanical equipment.Experimental results showed that the proposed method had higher prediction accuracy compared to other methods.

Key words: remaining life prediction, rolling bearings, long short term memory neural networks, convolutional neural networks, improved deep forests, health indicators, particle filtering, self-encoders, self-attention mechanism

摘要： 准确预测滚动轴承的剩余使用寿命对于保证机械系统的安全运行和制定维修策略具有重要意义。然而,在实际工业应用中,由于工况的变化和环境噪声的干扰,从采集到的信号中提取有用特征十分困难。此外,还存在首次预测时间(FPT)测定模型准确度较低以及趋势分析模型过于简单等问题。上述问题使得机械设备剩余使用寿命(RUL)的高精度预测变得极具挑战。为此,提出了多模型融合的轴承剩余使用寿命预测新方法:首先,构建了结合改进深度森林(GADF)的健康指标模型和结合自注意力机制的自编码器( SAAE)的FPT测定模型;随后,基于FPT测定结果构建粒子滤波模型进行健康指标的趋势分析,最终得到机械设备的剩余使用寿命。实验验证表明,所提方法相较于其他方法具有较高的预测精度。

关键词: 剩余寿命预测, 滚动轴承, 长短时记忆神经网路, 卷积神经网络, 改进的深度森林, 健康指标, 粒子滤波, 自编码器, 自注意力机制