交流接触器剩余电寿命的条件密度预测建模研究

AC contactors, which switch frequently, are widely used in electrical control systems. Operating AC contactors must be replaced before failure to guarantee safety and reliability of electrical systems. Therefore prediction of residual endurance of AC contactors is an important issue in electrics. AC contactors’ residual endurance (RE) primarily depends on their electrical residual endurance (REE). The conventional and most commonly used prediction method defines a maximum closing-breaking value for each type of AC contactor and counts operation times. This method is of low accuracy and resource wasting. At present related studies are at a theoretical stage, whose practicality is not very high. Hence the study on REE prediction of AC contactors is an issue to be solved for advanced manufacturing..Based on big dataset analysis and machine learning, this project proposes a novel prediction method building a prediction model for each type of AC contactor respectively: ① current and voltage waveforms samples between the contacts of each AC contactor are acquired and corresponding REE values are calibrated by novel methods during closing-breaking tests; ② the signal waveforms are pre-processed and characteristics are computed; ③ the relation between REE and signal characteristics are studied; ④a prediction model is built by using Conditional Density Estimation; ⑤ the model is applied to online signal samples to predict corresponding REE values; ⑥ finally online real-time prediction is implemented on intellectual AC contactors..The proposed project improves the accuracy and robustness of AC contactors REE prediction, increases valid operations, and can be implemented on products. Therefore the research is of high significance to theoretical research and practical applications.

交流接触器有着接通与断开（分合）操作频繁的特点，被广泛地运用于电气控制系统中。为确保电气系统工作的安全、可靠，必须在接触器失效之前更换掉，因此剩余寿命预测至关重要。交流接触器的寿命主要取决于其电寿命，传统的分合计数预测方法准确度不高且造成浪费，现有新型预测方法的研究尚处于理论实验阶段，距离实现应用尚有一定距离。因此交流接触器剩余电寿命的研究是先进制造业发展需要解决的一项课题。本项目提出基于大数据分析和机器学习，通过接通-断开试验对交流接触器触头电流电压进行波形采样，使用新型寿命标定方法标定每次操作的电寿命，计算波形特征并分析电寿命和波形特征变化的关系，然后使用条件密度估计进行建模预测，并且最终在智能交流接触器上实现在线实时预测。本项目的研究可以提高交流接触器剩余电寿命预测的准确性和鲁棒性，增加交流接触器的有效操作次数，同时算法可在实际产品上实现，具有重要的理论意义和应用价值。

本文基于项目团队之前对交流接触器分断电弧特性与电寿命状态的研究成果，进一步研究了交流接触器的电寿命预测以及优化的理论和技术，包括三个方面的内容：1）基于深度学习的实时电寿命预测方法。该方法采集和重构分断操作的电流电压波形作为特征量矩阵，测量电触头质量损耗作为电寿命标签，采用卷积神经网络（CNN）建立电寿命预测的分类和回归模型，获得了达到实际应用价值的预测结果；2）电寿命优化的分断策略及应用实现。研究了分断起弧相位角（APA）与电触头质量损耗的关系，提出了兼顾降低烧蚀和三相平衡的优化分断策略，开发了选相分断控制装置。实验表明该方法提高电寿命可达77%；3）分断操作动态过程的研究。本文通过实验研究了分断过程中分断相位角（BPA）和起弧相位角（APA）之间的延时，发现该延时并非常数，并通过选相分断实验分析动态过程中影响延时的原因，得出选相控制的关键因素。本文的研究成果对交流接触器电寿命预测和优化的理论基础和实际应用，均有着重要的价值。