The SCM thesis Predicting Semiconductor Component Lead Time for an Oil and Gas Company: A Dynamic Safety Stock Model with Machine Learning was authored by Chong Zhang and supervised by Dr. Thomas Koch (thakoch@mit.edu). For more information on this research, please contact the thesis supervisor.
The fluctuation of semiconductor delivery time presents a persistent challenge to different industries. Inaccurate forecasts of delivery lead time, often deviating significantly from an initial supplier’s promised delivery lead time, can severely disrupt production schedules, escalate manufacturing costs, and undermine carefully orchestrated customer delivery plans. Companies whose business is characterized by volatility and unpredictable lead times need rigorous supply chain management strategies to mitigate risk and ensure operational continuity.
For industries operating within demanding environments, such as oil and gas exploration and production, where high-reliability components are paramount, the consequences of semiconductor supply chain instability are particularly acute. Unexpected delays can have cascading effects throughout the value chain, impacting critical operations ranging from exploring to drilling and production. In this context, the imperative for accurate lead time forecasting and robust inventory management systems has never been greater.
One leading multinational oilfield services company recognized the need for a proactive solution to this pervasive issue. While they had implemented a centralized inventory management system for semiconductor components, two critical vulnerabilities remained unaddressed. First, their system lacked a dynamic safety stock mechanism to mitigate the impact of fluctuating lead times. Second, the supplier-quoted lead time, forming the foundation of their planning process, proved consistently unreliable. This combination presented a significant risk, as downtime in this industry can translate into millions of dollars in lost revenue.
Decoding the endogenous and exogenous data
Determined to unravel the enigma of fluctuations in semiconductor delivery lead times, the company embarked on an ambitious data-driven initiative. They delved into a decade’s worth of historical purchasing data, meticulously analyzing a vast array of factors potentially influencing lead times. This comprehensive approach extended beyond internal metrics, encompassing external variables such as global economic indicators, industry-specific indices, and even component-level specifications.
The objective was clear: to uncover hidden correlations and predictive patterns within the data that could anticipate the potential lead time extensions. However, the complexity of the data demanded a more sophisticated approach than traditional forecasting methods could provide.
The appropriate approach emerged from the realm of machine learning. After rigorous testing of various models, the XGBoost algorithm emerged as the clear frontrunner, demonstrating an exceptional ability to navigate complex datasets and extract subtle patterns. The insights gleaned from this analysis were revelatory. One was that factors like the S&P 500 Semiconductor and Semiconductor Equipment indices, despite being related to semiconductor companies, do not demonstrate a significant impact on lead-time prediction. In contrast, the S&P indices for technology, healthcare, aerospace, and communication, which include companies with substantial semiconductor consumption like Apple, GE Health, Boeing, and Cisco, included key features that influence delivery lead times.
This discovery illuminated the interconnectedness of the global semiconductor market. A surge in smartphone sales or increased demand for medical imaging equipment could reverberate through the supply chain, impacting lead times for components utilized for oil and gas equipment.
A dynamic solution against the disruption
Armed with its powerful predictive model, the company was not satisfied with simply forecasting lead times more accurately. It took it a step further, developing a dynamic safety stock model that adapts in real time to changing market environments.
This innovative approach considers both the predicted lead time and its potential variability. Instead of relying on static safety stock levels, the system continuously adjusts inventory targets. When the model predicts longer lead times or increased volatility, safety stock levels automatically increase. Conversely, during periods of stability, the system can optimize inventory levels to reduce carrying costs.
The impact was significant. Simulations showed a dramatic reduction in inventory holding cost caused by component shortages. The supply chain procurement team and production planners gained a newfound confidence in their ability to predict lead times, manage semiconductor supply volatility, and optimize inventory levels. The ripple effects extended throughout the supply chain, with improved on-time delivery rates and reduced component shortage costs.
By leveraging the data-driven approach and machine learning algorithms, the company transformed its semiconductor components inventory management from a reactive to a proactive position. The new capability ensures consistent production schedules and timely project deliveries. The success of this initiative has demonstrated that machine learning algorithms can be an effective solution for lead-time forecasting in the oil and gas industry even with limited data availability.
For more about this capstone project, and to see the full results of this research, visit the Supply Chain Management Review online at SCMR.com.
Sponsor a Capstone Project
Collaborate with seasoned MIT SCM students and researchers to tackle real-world supply chain challenges head-on. Through a 10-month partnership, your team gains actionable insights and fresh solutions as student teams apply analytical skills and effective strategies to your current projects. Joint teams that include MIT SCM students and MIT CTL faculty work on real-world problems. If you are interested in learning more about sponsoring a capstone project for the upcoming academic year, please email scm-capstone@mit.edu to connect with a member of our team.
Contact Us