The world is ever-changing and the drivers behind the rapid change are companies. As businesses grow at unprecedented paces and spread their services/products globally, a smarter supply chain has become a key competitive factor. There has been a gradual shift from day-to-day operations management to techniques like MRP to integrated ERP systems. This is now giving way to intelligent systems. Machines which enhance efficiency by merging the computational powers of computers with the decision-making skills of human beings are the next step in better supply chains.
This paper looks into the ways AI is being implemented currently, requirements for a more efficient supply chain and some research and industrial applications of AI in supply chain management.
Artificial Intelligence(AI) aims to enable machines to mimic human thinking and essentially make machines which are capable of decision making. AI is primarily divided into 3 classes:
a. Weak AI or Narrow task AI: Capable of performing a single task
b. Strong AI or Artificial General Intelligence(AGI): Hypothetical AI at least as smart as a normal human being and able to learn and take better decisions with experience
c. Superintelligence: Hypothetical AI systems which far surpass human intelligence
Since the early 1990s, research into artificial intelligence has gained momentum. While AI is still in it’s nascent stages, applications based around Machine Learning(ML) and Natural Language Processing(NLP) are already commercially in use today, for instance smart chatbots which are able to reply like a normal human being would are integrated into popular social media websites. While AI has been applied in some fields like game playing and robotics, AI’s application to supply chain management has not been thoroughly explored (Russell and Norvig 1995, Luger 2002).
The definition of AI is all-encompassing which has led to the development of multiple roadmaps to the achievement of intelligence in machines. From the development of Artificial Neural Networks (ANN), which enables a machine to think like a human by emulating human thinking, to Machine Learning (ML), which tries to enable machines to become self-learning; they all seek to make machines smarter. (Min, Hokey 2010)
Artificial Intelligence in Supply Chain Management:
The tricky aspect of efficiently managing a supply chain resides in the sheer size and number of decisions which need to be taken to keep the entire process synchronized and working at the optimum level. To efficiently deal with this vast selection of activities, the decisions which need to be taken can be classified into 3 major aspects on the basis of frequency of required decisions and scale:
a. Strategic decisions like facility planning
b. Tactical decisions which encompass mid-management level decisions like supplier selection and inventory planning
c. Day-to-day operational decisions like routing and picking (Min, Hokey 2010)
Individual tasks are being assigned to AI systems to add a USP to a product. The advent of self-driving cars, like the ones implemented by Tesla are an application of AI. The cars operate as a swarm with shared intelligence. The shared data is then used to make the cars’ self-driving feature better. Having a fleet of self-driving vehicles which are programmed to take the best route available would massively reduce the lead time of any supply chain. Similarly, research into applications of AI to SCM is underway and below we discuss a few of them.
Forecasting and Inventory Control:
For a supply chain, forecasting is one of the key and generally the first tactical decision taken. Forecast accuracy determines the profits for a company. As globalization puts a strain on companies to increase their product offerings, AI becomes crucial to handle the large number of SKUs. AI is also necessary to account for the huge deviations seen in the VUCA world. In 2017, Amirkollai et al wrote a paper on demand forecasting of spare parts for the Dassault company’s airlines. Neural network decision making modelled upon mean squared error in forecasting is used to get forecasts. On comparison with conventional forecasting methods, AI based-forecasting is observed to have higher accuracy.
Inventory management is a key financial driver for most firms. From forecasting to replenishment techniques to safety stock norms, all of them need to be in synchronization for profit maximization. Conventional techniques of inventory forecasting have been compared with an optimal situation of merging multiple forecasting techniques at different nodes to get optimal output. Ponte, et al compared forecasting techniques of moving average, exponential smoothing and naïve forecasting with an agent-based AI approach. It was observed that the standard deviation reduces. This can be attributed to the learning of AI which optimizes the entire organization inventory rather than local optimization of nodes.
Visibility Enhancement and Chaos Control:
Many of the decision making nodes of a supply chain tend to add uncertainty to a supply chain. Factors like customer behavior and bullwhip effect tend to cause blockages in supply chain by inventory mismanagement and inefficient logistics. Non-AI based chaos control has been discussed by a few researchers with solutions ranging from the popular linear feedback control to methods like adaptive control and impulse control. Kocamaz, et al 2016 discuss the application of ANN and Adaptive Neuro-Fuzzy Inference System (ANFIS) to reduce the uncertainty in modern supply chains. A trained neural network with linear feedback control performs better than a simple linear feedback control. Taking it a step further to incorporate ANFIS reduces the time for synchronization of chaotic supply chains.
Visibility of data across the entire supply chain is instrumental in reducing deviations and synchronizing activities. Silva, et al 2017 explore the usage of ANN for decision-making of order transfer to different nodes of the supply chain based on requirement of the next node. The paper uses a simulation of a complex multi-level supply chain to compare AI techniques with conventional techniques. The ANN is able to provide high recognition rate for the next 3 time periods of on-time order fulfilment and slippages. The system also focuses on inventory level recognition by analyzing the usage of entities at each node. This would allow for timely ordering of raw material so as to avoid slippages.
AI research into maintenance modelling has been going on for the past 3 decades. AI’s application to maintenance is applied in multiple tasks. In their paper in 2011, Mozami et al used various factors like road width and traffic volume index to prioritize maintenance activities. Fuzzy Logic (FL) has been used in this paper. Genetic algorithms have been used for project management in Lapa et al’s paper in 2006. The paper looked into the application of genetic algorithms to maximize the availability of a nuclear power system by scheduling maintenance.