Abstract: In shop floor management,equipment load is a key performance indicator that directly affects productivity and production costs,but little research has focused on how to achieve equipment load stability.To this end,a smart shop floor scheduling method for equipment load stabilization was proposed.The method uses A scheduling agent which contained a deep neural network scheduling model was used to analyze the correlation between shop floor production status and equipment load,and a scheduling strategy that met the desired goal was outputted timely.For the deep neural network scheduling model,a deep neural network scheduling model trainer was designed based on Double Deep Q-learning Network(DDQN),which used reward and punishment learning to form scheduling samples unsupervised,so as to update the network parameters of the deep neural network scheduling model and realize the model self-learning.The proposed method was validated in the MiniFab semiconductor production shop floor model.The result proved that the proposed smart shop floor scheduling method could realize the control of equipment load in the smart shop floor,thus ensuring the stability of overall equipment load.

Key words: smart shop floor, equipment load, scheduling, deep Q-learning network

摘要： 在车间管理中,设备负荷是一个关键性能指标,负荷稳定直接影响了生产效率与生产成本,但目前鲜有研究关注如何实现设备负荷稳定的问题。为此,提出一种面向设备负荷稳定的智能车间调度方法。该方法通过一个含有深度神经网络调度模型的调度智能体,分析车间生产状态与设备负荷间的相关性,及时输出满足期望目标的调度方案。针对深度神经网络调度模型,设计了一个基于双深度Q学习网络(DDQN)的深度神经网络调度模型训练器,其利用奖惩学习免监督地形成调度样本,借此对深度神经网络调度模型进行网络参数更新,实现模型自学习。所提方法在MiniFab半导体生产车间模型中进行了验证,证明了所提调度方法能实现对智能车间设备负荷的控制,从而保证车间整体设备负荷的稳定性。

关键词: 智能车间, 设备负荷, 调度, 深度Q学习网络