摘要
机器人已广泛应用于汽车涂胶生产线,其突发故障会对生产节拍与成本造成很大的影响。目前,机器人本体的齿轮箱故障一般采用事后维修,因此迫切需要实施预防性维护措施。针对当前齿轮箱故障预测困难的问题,通过传感器采集齿轮箱的状态信息,建立故障预测深度学习模型,识别出可能导致故障的异常模式,从而实现故障的预测。首先建立基于生成对抗网络(GAN)的多变量时间序列信息异常检测框架,通过改进损失函数增强生成器的收敛性;然后引入基于时间扭曲编辑距离(TWED)的重构误差计算方法,精确计算时间序列信号的差异;其次采用基于局部异常概率(LoOP)的异常评价方法,对每个数据点进行异常评分,提高检测的准确率;最后以某白车身涂装单元对方法的有效性进行了应用验证。
Robots are extensively utilized in automotive gluing lines,where their sudden failures can significantly impact production pace and costs.Presently,gearbox failures in robot bodies are typically addressed post-occurrence,necessitating the urgent implementation of preventive maintenance measures.To tackle the challenge of difficult gearbox fault prediction,sensors are employed to collect the gearbox's state information,and a deep learning model for fault prediction is established.This model aims to identify abnormal patterns that could lead to faults,thereby enabling fault prediction.Initially,a multivariate time series anomaly detection framework based on a Generative Adversarial Network(GAN)is developed,which enhances the generator's convergence through an improved loss function.Subsequently,a reconstruction error calculation method based on Time-Warped Edit Distance(TWED)is introduced to accurately compute differences in time series signals.Furthermore,an anomaly evaluation method based on Localized Probability of Anomaly(LoOP)is adopted for scoring anomalies at each data point,thereby improving detection accuracy.Finally,the method's effectiveness is applied and validated in a body-in-white painting unit.
作者
史天一
时轮
何其昌
Shi Tianyi;Shi Lun;He Qichang(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)
出处
《传动技术》
2024年第1期37-43,共7页
Drive System Technique
关键词
齿轮箱
故障预测
多变量时间序列
生成对抗网络
重构误差
gearbox
fault prediction
multivariate time series
generative adversarial networks
reconstruction error
