Cybermanufacturing Analytics

The third industrial revolution, introduced in Chapter 1, involved embedded systems such as sensors and programmable logic controllers to achieve automation in manufacturing. With the extensive use of embedded systems, the third industrial revolution significantly improved throughput, efficiency, and product quality in the manufacturing industry, while reducing reliance on manual operations. This opened the era of ``smart manufacturing”, which is utilizing sensor data to enable data-driven decision-making and equipment managed by numerical controllers (Kenett et al., 2018b). Cybermanufacturing is the next phase in industry, leveraging advances in manufacturing technologies, sensors and analytics (Kenett and Redman, 2019; Kang et al., 2021b), Suppose that $X$ is a set of process variables (scalar, vector, or matrix) and $Y$ are performance variables (scalar, vector, or matrix). Modeling consists of identifying the relationship $f$, such that $Y=f(X)$, or an approximation $f’$, such that $Y=f’(X)+ \varepsilon$ where $\varepsilon$ is an error term. Modeling and analysis provide the foundation for process monitoring, root-cause diagnosis, and control.

Modeling can be based on physical principles such as thermodynamics, fluid mechanics and dynamical systems and involve deriving exact solutions for ordinary or partial differential equations (ODE/PDEs) or solving an approximation of ODE/PDEs via numerical methods such as finite element analysis. It aims at deriving the exact relationship $f$, or its approximation (or discretization) $\tilde{f}$ , between process variables $X$ and performance variables $Y$ (Chinesta, 2019; Dattner, 2021). When there is a significant gap between the assumption of physical principles and the actual manufacturing conditions, empirical models derived from statistically designed of experiments (DOE).

Smart manufacturing exploits advances in sensing technologies, with in situ process variables, having an impact on modeling and analysis. This enables online updates and provides improvements in real time product quality and process efficiency. Another advance has been the development of soft sensors which provide online surrogates to laboratory tests.

There are two important challenges in manufacturing analytics: 1) data integration, also called data fusion and 2) development and deployment of analytic methods. Data fusion refers to the methods of integrating different models and data sources or different types of datasets. Machine learning refers to the building a mathematical model based on data, such that the model can make predictions without being explicitly programmed to do so. In particular, deep neural networks have shown superior performance in modeling complex manufacturing processes. Data fusion and data analytics play crucial roles in cybermanufacturing as they provide accurate approximations of the relationship between process variable $X$ and performance variables $Y$, denoted as $f’$, by utilizing experimental and/or simulation data. In predictive analytics, models are validated by splitting the data into training and validation sets. The structure of data needs to be accounted in the cross validation. An approach called befitting cross validation (BCV) is proposed in Kenett et al. 2022.