(59x) Data-Driven Supply Chain Monitoring Based on Canonical Variate Analysis

A supply chain is a network of entities including suppliers, manufacturers, distributors, and retailers. It deals with the procurement of raw materials, the manufacture of intermediates and final products, and the distribution of products to meet customers’ demand (Patel and Swartz, 2019). These entities are typically involved in a flow of materials from upstream suppliers to downstream customers, and a flow of information in the opposite direction. Good supply chain management improves the long-term performance of the supply chain system, and therefore is important to the general economic performance of an enterprise.

The environment under which a supply chain is operated may be unstable. Modern supply chain networks are prone to uncertainty due to their increasing complexity and interrelation. The presence of uncertainty in a supply chain system, such as uncertain demand, process yield, and transportation, gives rise to many types of risks in the system and makes supply chain management complicated (Tang, 2006). Unexpected events may disrupt supply chain operations and cause substantial negative effects that propagate across the supply chain. The ability to detect abnormal supply chain operations in a timely manner is crucial to the functioning and economics of a supply chain system.

The goal of SCMo is to support supply chain decision-making through characterizing the normal operating conditions, raising alarms of abnormal events, identifying potential reasons, and providing suggestions for mitigation (Wang et al., 2020). Research on the SCMo problem involves contributions from a variety of domains. Most studies adopt certain performance indicators to monitor the supply chain. Although many approaches have been proposed for SCMo, research on applying data-driven statistical process monitoring methods to SCMo is still scant. Wang et al. (2020) investigate the fault detection and diagnosis of supply chain systems using dynamic principal component analysis (PCA) and agent-based modelling. Their study shows that PCA-based monitoring statistics and variable contributions are effective in detecting and diagnosing abnormal supply chain operation caused by transportation delay, degradation in manufacturing yield, and supply shortage. This indicates the potential of applying data-driven methods to SCMo, which is referred to as data-driven SCMo in this work.

The present work aims to explore the use of another data-driven method, canonical variate analysis (CVA), in SCMo. CVA is multivariate statistical analysis method that discovers the relationship between the past and future system data. Since its introduction to system and process identification by Larimore (1990), it has received widespread attention in statistical process monitoring (Russell et al., 2000; Severson et al., 2016). In this work, a data-driven SCMo method based on CVA is presented. In this method, a supply chain is considered as a dynamic system and the state-space model of the supply chain system is developed. The normal operating conditions of the supply chain are characterized using CVA. Then, CVA-based monitoring statistics and variable contribution plots are used to detect supply chain faults and identify the fault-related variables. To address singular covariance matrices, a sparse CVA algorithm is adopted (Lu et al., 2018). An approach to hyperparameter tuning for the sparse CVA method is presented. Furthermore, a novel fault impact prediction method that utilizes the time-dependent relationships inherent in the CVA model is proposed.

The proposed SCMo scheme is validated on two case studies – the classical beer distribution game, and a more complicated supply chain system that has material flow in forward and reverse directions. The performance of CVA in fault detection is examined and compared against dynamic PCA in terms of dimensionality reduction, false alarm rate, missed detection rate, and detection delay. A Python-based supply chain simulator is developed for the simulation study. Results show that CVA can detect abnormal supply chain operations and achieve comparable performance to dynamic PCA in a lower-dimensional latent space. Moreover, the proposed fault impact prediction method is effective in identifying the variables that will potentially be impacted by a fault.

References

Larimore, W.E., 1990. Canonical variate analysis in identification, filtering, and adaptive control. In: 29th IEEE Conference on Decision and Control, vol. 2. pp. 596-604.

Lu, Q., Jiang, B., Gopaluni, R.B., Loewen, P.D., Braatz, R.D., 2018. Sparse canonical variate analysis approach for process monitoring. J. Process Control 71, 90-102.

Patel, S., Swartz, C.L.E., 2019. Supply chain design with time-limited transportation contracts. Comput. Chem. Eng. 131, 106579.

Russell, E.L., Chiang, L.H., Braatz, R.D., 2000. Fault detection in industrial processes using canonical variate analysis and dynamic principal component analysis. Chemometr. Intell. Lab. Syst. 51 (1), 81-93.

Severson, K., Chaiwatanodom, P., Braatz, R.D., 2016. Perspectives on process monitoring of industrial systems. Annu. Rev. Control 42, 190-200.

Tang, C.S., 2006. Perspectives in supply chain risk management. Int. J. Prod. Econ. 103 (2), 451-488.

Wang, J., Swartz, C.L.E., Corbett, B., Huang, K., 2020. Supply chain monitoring using principal component analysis. Ind. Eng. Chem. Res. 59 (27), 12487-12503.