Since times immemorable route optimisation has been taught as one of the starting tricks in operations research and associated subjects in supply chain management. All the theory and mathematics of the problem is summarised in this post. I will not repeat the material because the original does such a good job of describing it.
I want to focus on the practical aspects of route optimisation which supply chain managers frequently encounter.
I was introduced to the problem way back in 1988 when I was the navigating officer of a merchant vessel shuttling between the paper and pulp mills of west coast of north America (Mainly Pacific North West - Canada and Alaska), and their customers in China, Taiwan, Indonesia, Malaysia, Singapore, India, Pakistan, UAE and Saudi Arabia.
There must have been a nearly a dozen or so paper and pulp mills each with their own set of customers in each of these locations and order backlog when the ship hit the mill to take away the next lot of orders.
On the other hand each of customers ordered from a number of mills to optimise their mix of product range, prices, quantities and supplier relationships.
The task was to make sure that the cargo for each discharge post was ready and exposed for discharge when the ship hit the discharge port. At the same time, the ship had to be stable and not capsize because too much cargo was loaded in forward section or aft, or starboard, or port side of the ship.
Also, all the working hooks had to be available at all times in each of the ports to optimise cargo loading and unloading process, to expedite ship movement.
On top of it the cargo of paper and pulp is very weather sensitive and had to protected vigorously in a very rainy loading and discharging climate on both ends.
This ships hatch covers were hydraulically operated for ease of opening and closing readily, but because it was a very old ship, a lot of hydraulic piping was already corroded and would give way without any notice. Improvisation was the order of the day, especially on a rainy day - which was almost every day in most of these ports.
There were many other smaller practical considerations which I have not mentioned in this blog, but I wanted to give the theoreticians a flavour for a real life route optimization problem so that they get away from the belief that a mobile phone app will one day run their routes.
The case above is not even the most complex cases of route optimisation. That honour would have to go to the problem of milk runs made by a truck carrying detonators and sticks of explosives to quarries and construction sites around the mid-west for one of our large explosive manufacturing clients. I will not go into all the practical considerations related to the safety and peculiarities of the trade and region because I do not want to divulge confidential information.
Similarly, routes for container ships that load and unload from each port they touch, and have to have stacks ready for each while maximising the workability of each cargo hold and optimising the ship strength and stability is far more complex that any simple route optimisation problem. Yet, shipping companies have built sophisticated loading programs that achieve these outcomes - albeit on a more or less fixed route basis.
Now I come to real purpose of writing this blog. This is very important and must be read by everyone who is a big advocate of use of AI and algorithms in SCM.
Every algorithm and AI has a built in set of limitations and assumptions. That is similar to every radar having its blind spots and resolution errors. If you do not know these limitation and assumptions for the algorithm you are using - you have no business using that algorithm. That would be like driving a car once you know how to use the accelerator but without know the function of a brake.