(339q) Supply Chain Monitoring Based on Principal Component Analysis

A supply chain is a network consisting of suppliers, manufacturing facilities, warehouses, retailers, and customers. In the network, there is material flow from upstream suppliers to downstream customers, and information flow in the opposite direction. A well-functioning supply chain is critical to the competitiveness of an enterprise. Supply chins are susceptible to various types of risks such as procurement failures, poor yield, delays, inaccurate forecasts, uncertain demands, and disruptions due to natural disasters (Chopra and Sodhi, 2004). Disturbances may lead to economic loss or even breakdown of the supply chain if there is not an effective mitigation strategy (Tang, 2006). Therefore, improving the ability of disturbance detection for a supply chain can help reduce the risks and increase its resilience (Chopra and Sodhi, 2014). The purpose of this work is to investigate the application of data analytics to supply chain monitoring, fault detection and diagnosis.

This work focuses on a well-established multivariate statistical method, principal component analysis (PCA), which is designed for extracting uncorrelated components from correlated data (Wold et al., 1987). PCA has been widely applied to industrial process modeling, monitoring and diagnosis (Kresta et al., 1991; Qin, 2012). By using a time lag shift technique, PCA is able to capture dynamic behavior of a system, giving rise to dynamic PCA (DPCA) (Ku et al., 1995). In the current literature, PCA has been applied to some aspects of supply chain analytics including its use in determining market segments for new products (Lei and Moon, 2015), reduction of redundancies of performance indicators, thus aiding multiobjective supply chain optimization (How et al., 2018), and characterization of uncertain parameters in supply chains by reducing the dimensionality of correlated uncertainty data (Ning and You, 2018). However, applications of PCA to supply chain monitoring are significantly lacking, in contrast to applications of PCA in processes plant operations.

In this work, PCA and DPCA are applied for the fault detection and diagnosis of supply chain systems. Supply chain data such as inventory levels, market demands, and material in transit are collected. PCA and DPCA are employed to model the normal operating conditions (NOC) of the supply chain. Two monitoring statistics, the Hotelling’s T² and the squared prediction error (SPE), are adopted to detect deviations from the NOC of the supply chain. Contribution plots are adopted to identify the fault-related variables when at least one index exceeds the limits. The proposed supply chain monitoring method is evaluated on two case studies, one of which is a four-echelon supply chain with single product, and the other a two-product supply chain with materials transported both from upstream to downstream and in the opposite direction. A Python-based supply chain simulator is developed, and different scenarios are simulated and analyzed. The results show that the SPE by DPCA is more reliable than the other fault detection indices considered in this work, while that by PCA is not sensitive enough. Abnormal behavior of the supply chain, such as transportation delay, low production rate and supply shortage, are successfully detected by DPCA. The proposed method is shown to apply as well to non-contiguous data and seasonal demands. Furthermore, the contribution plots help in interpreting the abnormality and identifying the fault-related variables.

References

Chopra, S., Sodhi, M.S., 2004. Managing risk to avoid supply-chain breakdown. MIT Sloan Management Review, 46, 53-62.

Chopra, S. Sodhi, M. S., 2014. Reducing the risk of supply chain disruptions. MIT Sloan Management Review, 55, 73-80.

How, B.S., Lam, H.L., 2018. Sustainability evaluation for biomass supply chain synthesis: Novel principle component analysis (PCA) aided optimisation approach. Journal of Cleaner Production, 189, 941-961.

Kresta J.V., MacGregor J.F., Marlin, T.E., 1991. Multivariate statistical monitoring of process operating performance. Canadian Journal of Chemical Engineering, 69, 35-47.

Ku, W., Storer, R.H., Georgakis, C., 1995. Disturbance detection and isolation by dynamic principal component analysis. Chemometrics and Intelligent Laboratory Systems, 30, 179-196.

Lei, N., Moon, S.K., 2015. A decision support system for market driven product positioning and design. Decision Support Systems, 69,82-91.

Ning, C., You, F., 2018. Data-driven decision making under uncertainty integrating robust optimization with principal component analysis and kernel smoothing methods. Computers and Chemical Engineering, 112, 190-210.

Qin, S. J., 2012. Survey on data-driven industrial process monitoring and diagnosis. Annual Reviews in Control, 36, 220-234.

Tang, C. S., 2006. Robust strategies for mitigating supply chain disruptions. International Journal of Logistics Research and Applications, 9, 33-45.

Wold, S., Esbensen, K., Geladi, P., 1987. Principal component analysis. Chemometrics and Intelligent Laboratory Systems, 2, 37-52.