GOALI: Real-time Performance Prediction of Multi-Stage Manufacturing Systems using Nonlinear Stochastic Differential Equation Models

该资助为基于非线性随机动态系统原理的制造系统性能输出（例如，吞吐量和成品率）的实时预测提供了基于物理和数据驱动的方法。现代制造企业已经投资了各种传感器和IT基础设施，以提高工厂车间系统的可见性。这为从动态角度跟踪制造系统的性能提供了前所未有的机会，而不是静态角度。传统的静态模型不足以从这些大型数据源实时预测性能变量。动态模型是必要的，例如由提议的研究得出的模型。与以前的方法不同，该方法将明确考虑影响生产线停机时间分布的退化和修复动力学。将使用s型函数理论来消除模型中的不连续。通常使用的平稳性假设将被放宽，模型阶数将减少，以促进在非平稳状态下快速和准确的实时性能估计。该方法将通过从通用汽车（GOALI合作伙伴）汽车装配线的实际操作中获得的多个数据集进行验证。如果成功，该研究结果将导致现实世界制造系统的仿真建模和实时性能预测的改进，包括汽车和半导体行业。为了使现代工业使其业务适应需求波动以及供应和能力变化，快速和准确的模拟模型以及业绩预测器正变得越来越必要。随着国内工业制造转向“按订单组装”。在范例中，建议的方法可以是极有价值的手段来跟踪和预测绩效，以支持短期和中期的决策。

This grant provides funding to pursue physics-based and data driven approaches for real-time prediction of performance outputs of manufacturing systems (e.g., throughputs and yield rates) based on nonlinear stochastic dynamic systems principles. Modern manufacturing enterprises have invested in a variety of sensors and IT infrastructure to increase plant floor systems visibility. This offers an unprecedented opportunity to track performance of a manufacturing system from a dynamic, as opposed to a static sense. Conventional static models are inadequate for predicting performance variables in real-time from these large data sources. Dynamic models, such as those resulting from the proposed research, are necessary. Unlike previous approaches, degradation and repair dynamics that influence the down time distributions in a manufacturing line will be explicitly considered in the proposed approach. Sigmoidal function theory will be used to remove discontinuities in the models. The commonly used stationarity assumptions will be relaxed and model order reduced, to promote fast and accurate real-time performance estimation under nonstationary regimes. The proposed approach will be validated using multiple data sets acquired from actual operations at General Motor's (GOALI partner) automotive assembly lines.If successful, the results of the proposed research will lead to improvements in simulation modeling and real-time performance prediction of real-world manufacturing systems, including in automotive and semiconductor industry. Fast and accurate simulation models as well as performance predictors are becoming increasingly necessary to make the modern industries adapt their operations to demand fluctuations as well as supply and capacity variations. As the domestic industrial manufacturing is moving to ?assemble-to-order? paradigm, the proposed approach can be extremely valuable means to track and predict performance to support decision-making over short- and medium-term horizons.