基于噪声智能感知的高速化纤卷绕机群组预测性维护研究

高速化纤卷绕机是纺织行业的重要设备，目前大多企业对其采取点检和周期性维护模式，导致维护成本巨大。虽然预测性维护能够有效地避免此问题，但在实施过程中还存在着以下难点：一是高速化纤卷绕机关键部件的故障机理不明确，采集到的状态监测信号中含有大量强背景噪声，故障表征微弱，设备运行工况多变，故障难以检测；二是高速化纤卷绕机工况变化分割不明晰，噪点数据和有效样本的数据分布不同且数据量不平衡，退化机制难以建模；三是现有的预测性维护决策模型大多以单一指标进行维护优化，而车间级的设备群组维护决策通常包含可靠性、可用性等多种指标。针对上述问题，本项目在研究高速化纤卷绕机关键部件故障机理的基础上，基于噪声有益性理论，构建了噪声智能感知的设备微弱故障检测和退化机制建模的方法，在此基础上，综合考虑各种相关因素，建立了博弈框架下的多目标设备群组维护决策模型。本项目的研究将完善预测性维护理论体系，提高其工程实践力。

The high-speed fiber winding machines are the key equipment in the textile industry. Most companies conduct the spot inspection and periodic maintenance on winding machines, thus huge maintenance cost is generated accordingly. Even though predictive maintenance can solve this problem, there are still many issues to be considered when conducting predictive maintenance on the high-speed fiber winding machines. First, the fault mechanism of the key components of a high-speed fiber winding machine is unclear. In addition, there are lots of strong background noises which make the fault characteristics weak. Moreover, the operational condition of a high-speed fiber winding machine is time-varying, which enhances the detection difficulty of a weak fault signal. Second, the segmentation of varying operational conditions is unclear and there are lots of noisy samples which have different distributions from those of valid samples. Besides, the noisy samples and valid samples are unbalanced. Therefore, the degradation prediction models are hard to build. Third, most of the current predictive maintenance models are built with single optimization criterion, but the maintenance for machines always involves in many requirements from the perspective of workshop. To solve the above issues, new weak fault detection methods and degradation prediction methods are proposed based on the noise-intelligence prognostics in consideration of noise benefit theory after studying the fault mechanism of key components of a high-speed fiber winding machine. Thereafter, new maintenance decision models are built with multiple objectives under the framework of game theory in consideration of different requirements. This study will improve the theory and application of predictive maintenance.