What are AI applications in Supply Chain Management?
Figure 1. Artificial Intelligence (AI) in Supply Chain Management (made by the author)
Artificial Intelligence (AI) is defined as the use of computers for logic, recognizing patterns, learning, and developing various forms of inference to solve problems and help make decisions. A multitude of AI tools, namely genetic algorithms, and artificial neural networks, was applied to resolve several supply chain management (SCM) issues. Let’s have a look at how these applications were utilized in SCM challenges.
#1: Genetic algorithm (GA)
This algorithm is based on the theory of evolution, which simply means that it scans through the possibilities, evaluates them, and then creates a solution to a given problem. Possibilities can be understood in the following context: when you make a decision, the outcome is unknown and it can be anything (we can’t know for sure what happens in the future). If you plan to cook dinner, you can give the ANN information such as ‘there are carrots, chicken, vegan burger, rice in the fridge’, ‘you don’t want to eat rice today’, ‘the supermarket is close by’, etc. The ANN will then create different possibilities (scenarios) such as ‘you can go to the supermarket to get potatoes’, ‘you can then make chicken with potatoes’, ‘you can eat a vegan burger with carrot’, or ‘you can eat carrot and potatoes’. This means ‘scanning through possibilities’ and of course, there are more possibilities depending on the criteria and information you feed the ANN. In the end, the ANN will evaluate these possibilities and select the one that meets all the requirements.
Specifically, GA translates multiple supply chain agents (e.g. supplier, retailer, manufacturer) into numerical strings called chromosomes. Through multiple iterations and mutations of chromosomes using fitness value (criteria that a potential solution needs to meet), GA creates a solution that is not necessarily the best, yet meets all the requirements. The performance of a GA depends on population size, number of generations, recombination rate, mutation rate, and termination criteria. This application has been utilized in supply chain network design, routing, and inventory management. Research from D. Kallina et al. (2021) shows a straightforward example of how GA was used to analyse the Bill of Materials to provide a more optimal delivery route from different suppliers to different customers. This research also highlighted that besides the quantitative factors, the qualitative factors should be considered in coding. This stems from the fact that different companies have different strategies (e.g., cost-driven, differentiation), so the solution should also calculate this factor by using a corresponding scale for example.
Thakur (2013) researched GA applications in inventory management. The GA, in this case, is designed to predict an optimal stock level by analysing historical data, to reduce the excess amount of stocks and prevent a shortage. Data was labelled as follows: Zero means that the supply chain (SC) contributor needs no inventory control while non-zero data requires inventory control. Logically, both the excess amount of stocks and the shortage amount should be labelled as non-zero data. The labelled data is fed to a clustering algorithm that separates stock levels that are either in excess or shortage from the ones that are neither in excess nor shortage. Then, the process starts with GA where multiple agents are randomly generated with the stock levels within the lower limit and the upper limit for all SC contributors. Through this process, fitness values, crossover function, and mutation are applied to each individual in the population (stock levels), resulting in the final optimal stock level.
#2: Artificial neural networks (ANN)
Artificial neural networks are algorithms theoretically designed to function like the brain cells. The ANN uses an interconnected network of computers to learn from past data, recognize distinct patterns and trends, and eventually make new and more accurate decisions so it simulates the process of learning. A remarkable feature of ANN is the ability to deal with incomplete or abstract data. This algorithm has been applied to solve several problems in SCM, such as demand planning, forecasting, cost prediction, performance measurement, SCM visibility, etc.
One of the most recent researches on the application of ANN in improving SCM visibility is N. Silva (2017), who developed a multi-echelon SC (which is referred to as a complex SC network) in a simulator to generate data that could feed an ANN developed using MATLAB programming*. The purpose of the research is to compare the actual output data from the current system of retailers with the output data from the designed ANN. The input data was pre-defined in a way that infers the uncertainty about the order size that the retailer must deliver to end customers. In principle, order size can be certain (known) and uncertain but in practice, as we know, the order size is usually uncertain due to multiple internal and external SC factors (competitors, demand, etc.). The output data was the time registered for each order in the current system of retailers, and the inventory levels of each SC node whenever a new order entered the system. ANN was developed as a multi-layer perceptron: one input layer, one output layer, and one or more hidden layers. After training the network, the test phase was performed with untrained data. Untrained data can be referred to as data coming from an untrained model which is simply an algorithm with random parameters. It simply produces outputs by guessing, without any criteria or requirements.
Two experiments were executed by the author of the research “Improving Supply Chain Visibility With Artificial Neural Networks” (Silva, 2017), and the results were evaluated based on a recognition rate calculated by comparing the outputs with the values that were returned from the computer simulation of an ANN. In experiment 1, ANN was designed to predict if coming orders will be instantaneously fulfilled and if not, how long it will take to finish them. In the other experiment, ANN was built to foresee which retailers would reach their re-order point and place new purchase orders to buy goods from suppliers. The result of the recognition rate was remarkable in both experiments; however, the author concluded that a longer time horizon of historical data may be unnecessary in cases when the outcomes from short periods (e.g., 2 months) and longer periods (e.g., 6 months) are practically the same. Thanks to this model, SC visibility will be improved. This means that companies can act more proactively to avoid SC disruptions or handle demand uncertainty.
In short, Genetic algorithms and Artificial neural networks have been utilized in resolving several SCM challenges that humans can’t effectively and efficiently handle. These applications are quite complex for organizations to learn and develop, but once they manage to get it worked properly, there will be certainly sustainable advantages.
* According to Mathworks (n.d.), MATLAB is a high-level programming language designed for engineers and scientists.
Recommended reading:
Optimization of Supply Chain Network using Genetic Algorithms based on Bill of materials (Kallina et al., 2021)
The role of artificial intelligence in supply chain management: mapping the territory (R. Sharma et al., 2022)
Inventory Analysis Using Genetic Algorithm In Supply Chain Management (L. Thakur, 2013)
Silva, N., (2017) Improving Supply Chain Visibility With Artificial Neural Networks
Demand forecasting using neural networks for supply chain management (A. Kochak and S. Sharma, 2015)
Application of Artificial Intelligence in Automation of Supply Chain Management (R. Dash et al., 2019)
Artificial Intelligence for Supply Chain Resilience: Learning from Covid-19 (S. Modgil et al., 2021)
Mathworks (n.d.) https://www.mathworks.com/products/matlab/programming-with-matlab.html
Thank you for reading. Please share if you find this article useful!
0 Comments