When it comes to optimizing transportation, logistics, and shipping, artificial intelligence (AI) and machine learning (ML) algorithms have a vital new role to play. While getting the right product in the desired quantity and at the lowest price sounds easy in theory, many variable factors are in constant play, including data flows too massive to be managed by human operators, continuous disruptions in the distribution chain, fuel price volatility, the presence of multiple suppliers for the same products, and ever-changing, unpredictable levels of consumer demand.
To forecast future inventory needs, all sectors of logistics are therefore leaning into machine learning (ML), the branch of AI that makes machines smarter by feeding them data, from which they can “learn” what to do with it. But nowhere is the need for ML more sharply felt than in the shipping and maritime transportation industry. Here is just one practical application that looks at best practices in AI as they apply to shipping: predictive maintenance and spare parts management.
Optimizing parts management
Focused on the need for predictive maintenance on ships, this case study relates to our parts optimization work with a company that does drilling and exploration for new oil deposits. This company uses ships called FPSOs, which stands for Floating, Production, Storage, and Offloading. FPSOs are vessels used in the oil industry in locations far from the coast that cannot be reached by oil or gas pipelines. The management of spare parts in this type of vessel must take into account that the ship is an itinerant warehouse with very limited space.
This company's main objectives, therefore, were to avert stock breakdowns, increase the availability of spare parts, and avoid so-called “dead stock,” that is, the storage of materials that take up space unnecessarily on board.
ToolsGroup started by conducting a preliminary audit, for which we collected and validated master data of spare parts and ships, stock levels, history of consumption, and other statistics relating to the consumption of and demand for parts.
Next, we developed an artificial intelligence algorithm to address a “what-if” maintenance need that went beyond traditional preventive maintenance—in other words, the AI we engaged served to enable predictions and scenario planning. In so doing, we effectively built the shipping company a new business model that enabled them to better manage the process of predicting what spare parts each ship would need, taking into account all the logistics constraints.
While this process began with analyzing the current performance of these FSPO vessels, we were able to propose an entirely new business analysis and optimization model that allowed a view into “what-if” scenarios and evaluated different options for resolving them.
Typically, traditional preventive maintenance is an evaluation of all factors related to cyclicality or past events. But by plugging in multiple eventualities, the system was able to predict the need for given replacements outside the normal range of maintenance and expected breakdowns or timed obsolescence. Using AI thus allowed us to forecast or predict which spare parts would be needed and which should be on hand preventively, optimizing inventory levels and the transport of spare parts. In this case, we developed a form of machine learning comprising a self-adapting and self-learning algorithm specific to maintenance, repair, and operations on these ships. The system is also capable of calculating advanced consumption forecasts of parts. Hence the optimization of stock levels of mechanical spare parts and consumables, with stock levels based precisely on forecast algorithms, answered to the need for safety as well as convenience on these vessels—along with not getting stranded at sea.
The supply chain planning software the shipping company adopted used a phased approach—that is, we introduced the implementation in a conscious sequence, replacing old systems, processes, and methodologies gradually. We used probability forecasting and machine learning technologies that were designed to work together seamlessly and automatically. Starting from a basis of data on historical demand, the ML engine went on to improve the baseline probability forecasts by applying machine learning technology to the existing historical data. This helped to produce a more robust, reliable baseline forecast that accurately models the phenomena shaping the demand. The tool then layers on more sophisticated machine learning by leveraging additional external data sources.
That said, our experience at ToolsGroup suggests that forecasting can’t be completely based on machine learning techniques. Instead, it requires a solid statistical backbone to deal with the changing and often random nature of demand. In this case, we recommended that the company use a hybrid approach that employs probability forecasting and machine learning technologies which work together seamlessly and automatically.
To do this, we introduced a self-adaptive model for probabilistic forecasting using granular historical demand. We’ve found that for this shipping company and others, this approach is critical to success when using advanced machine learning—and yields significant benefits on its own. Applying machine learning technology to the existing historical data further improves the probability forecast, resulting in a more robust, reliable baseline that accurately models the phenomena shaping the demand. From there, the system can engage in more sophisticated machine learning, using external data sources such as weather forecasts, nautical indicators, availability through distributors and stores, social media and online search, Internet of Things, and more.
Machine learning engines thus improve the calculation of factors that affect demand. For this shipping company, ML produced a more accurate future forecast—resulting in lower costs, optimized inventory of parts needed, and reduced risk of downtime.
The quantitative, qualitative, and green benefits
Beyond helping to resolve some common industry problems, optimizing shipping supply chains has wider implications, as well. In the project discussed here, the benefits were first and foremost quantitative, since stock optimization coincides with the reduction of waste. The approach also enabled the avoidance of two common risks in logistics—stock-outs or the presence of excess stock. There are also qualitative benefits. For example, as planning improves, downstream interventions (and consequently costs resulting from re-negotiation with suppliers) decrease. Finally, greater efficiency is a source of greater sustainability, which is determined both in the reduction of waste and in the containment of potential toxic events. Enhanced forecasting forestalls corrective actions that can correspond to additional and therefore more costly and polluting transportation.
In general, one of the strengths of AI-powered technologies is their ability to crunch multiple demand variables to automatically generate a reliable demand forecast. This “self-tuning” approach allows the system to predict demand behavior much more accurately than considering demand history alone. Supply chain professionals understand the importance of accurate demand forecasting, yet this is a difficult task due to the extreme complexity of modern demand planning. Increasing forecasting complexity and rapidly shifting consumer demand are often exacerbated by seasonality, new product introductions, promotions, and myriad causal factors such as weather and social media. A high level of automated machine learning is an ideal application to improve forecast accuracy in supply chain planning. ML also supports the development of more resilient supply chain planning practices because it enables the whole system to react to changes and disruptions in a timely manner. Businesses that use ML-augmented supply chain platforms can harness real-time data for immediate action and become more resilient and future proof.
Authors’ Note: This case study was presented at a recent conference held in Genoa, Italy, “Digital Infrastructure and Predictive Logistics: Strategies, Risks and Opportunities in Transportation Supply Chain Data Exchange." The event was sponsored by Logistic Digital Community, a virtual community created through the initiative of Confcommercio-Conftrasporto in collaboration with Federlogistica and Consorzio Global.
Leo Cataldino has extensive international planning, project management, forecasting, reengineering, and supply chain management experience. He is a partner-manager and principal in the Logistics practice of ToolsGroup, a global firm focused on AI-driven supply chain planning.
Inna Kuznetsova is CEO of ToolsGroup. They can be reached respectively at email@example.com and firstname.lastname@example.org.
Copyright ©2023. All Rights ReservedDesign, CMS, Hosting & Web Development :: ePublishing