An exploratory study to identify rogue seasonality in a steel company's supply network using spectral principal component analysis

Variability in the information flows within a supply network requires production companies to either track the variations, hence leading to increased production on-costs, or to buffer themselves via the use of inventory which leads to stock holding costs.Customer demands generate variability, often in the form of seasonal patterns, but must be satisfied. In contrast, ?rogue seasonality?, i.e. unintended variability, may be generated by a company?s own internal processes such as inventory and production control systems. Importantly, rogue seasonality may propagate through a supply network. Thus there is a motivation for automated detection of network-wide rogue seasonality and for the diagnosis of its root cause.In this article, a data-driven technique known as spectral principal component analysis is used to detect and characterise cyclical disturbances in a supply network that indicate seasonality. All the information and material flows participating in each disturbance are detected, and the distribution of each disturbance enables a hypothesis to be reached about its root cause. The technique is applied to a supply network consisting of four autonomous business units in the steel industry. Two main cyclical disturbances were detected and diagnosed. One was found to be rogue seasonality and the other was externally induced by the pattern of customer orders. Variability in the information flows within a supply network requires production companies to either track the variations, hence leading to increased production on-costs, or to buffer themselves via the use of inventory which leads to stock holding costs.Customer demands generate variability, often in the form of seasonal patterns, but must be satisfied. In contrast, ?rogue seasonality?, i.e. unintended variability, may be generated by a company?s own internal processes such as inventory and production control systems. Importantly, rogue seasonality may propagate through a supply network. Thus there is a motivation for automated detection of network-wide rogue seasonality and for the diagnosis of its root cause.In this article, a data-driven technique known as spectral principal component analysis is used to detect and characterise cyclical disturbances in a supply network that indicate seasonality. All the information and material flows participating in each disturbance are detected, and the distribution of each disturbance enables a hypothesis to be reached about its root cause. The technique is applied to a supply network consisting of four autonomous business units in the steel industry. Two main cyclical disturbances were detected and diagnosed. One was found to be rogue seasonality and the other was externally induced by the pattern of customer orders.