The Power of Operations Research in Manufacturing - Q&A with Axel Parmentier: part 2
To understand how the manufacturing industry has been aided by Operations Research, as well as look ahead to the trends shaping the future, we spoke to Axel Parmentier, Researcher at the CERMICS, Ecole Nationale des Ponts et Chaussées. At Pelico we are very lucky to have Axel Parmentier as our scientific advisor. He has been helping us shape the Pelico solution. One of our Business Developers, Stanislas Thomann sat down with Axel to discuss why now is a great time for supply chain and manufacturing organisations to start unlocking the value of operations research.
- 1. How did you start working with Pelico, and what research problems do you typically work on?
This goes all the way back to my university days, where I got to know Mamoun Alaoui, Pelico’s co-founder. Then, ten years after we graduated, I received an email from him informing me he had started a new venture, called Pelico, and that they were looking for a researcher in mathematics and, more specifically, operations research. We had lunch, and by the time we had finished I was super interested in the concept behind Pelico and decided to jump on board.
A large part of the literature is focused on exact algorithms and heuristics, in a context where you have a lot of time to solve the problem. But this was definitely not the context we had at Pelico. Our first project together was to develop a scheduling algorithm that could answer almost immediately, so that if an operator using Pelico’s tool asked for a new schedule, a few seconds later he would receive it.
Secondly, we needed something flexible, because each of Pelico’s clients had their own specificities and constraints that would evolve with time. The challenge became: how do we design a flexible and extremely fast heuristics that gives very good results on a wide range of scheduling problems?
These are typically the kinds of problems I’ve been working on with Pelico for three years now.
- 2. What type of companies use Operations Research techniques?
There are basically three categories of companies that use operations research today.
The first are the large, traditional companies that have had operations research departments for a long time—say 60 to 70 years. This is any big company with a large-scale network, such as transportation, energy, or telecommunications companies. It is estimated that one third of the CAC40 companies in France have operations research departments.
These departments were mainly focused on very specific applications. For example, at Saint-Gobain, they were developing differently-shaped windshields from large sheets of glass. So you have to find a way to fit as many different shapes and sizes into this standard sheet of glass as possible. To do that, they apply operations research.
The second category are consulting firms dedicated to proposing operations research services to companies that do not have the skills in-house, or whose operations research capacities are too small and need extra firepower to scale to bigger problems. Some of them develop their own solvers, which are basically digital libraries that allow their clients to solve a very wide range of problems.
The third category that emerged in the last 10-20 years are startups. Most of the students that graduate from the operations research master programme that I teach in Paris end up joining startups. With the arrival of data, the scope of the problems that can be solved using operations research has increased dramatically. These startups will try to disrupt many markets where the established companies don’t have the required operations research skills in-house.
- 3. What is DDMS (digital design and manufacturing services)?
Simulation has been an important part of operations research for a long time. Whenever you want to make decisions based on algorithms, you need to run a mathematical model. The prerequisite before applying this kind of technique is the ability to model the processes of the company. When you have good models for your company, you have a good digital twin of many things.
- 4. What is a success story from your career?
One of the huge successes of the Operations research team that I was a part of at Air France was on machine learning and predictive maintenance. We had 10 years worth of data recorded by sensors on each flight which, for once, wasn’t stored with the aircraft company but with the airline itself. Because of our algorithm, they became extremely good at sending the aircraft for maintenance at the right moment, even in between flights right on the tarmac, maximizing airtime and minimizing downtime.
- 5. What problems can be solved with Operations Research nowadays?
Today we have a very wide range of applications that can be solved with operations research tools. Perhaps the most well-known technology in operations research is mixed integer linear programming (MILP). MILP looks very simple: you are minimizing a linear function under linear constraints and the additional constraints that some variables can be integers or not. The integrality constraints are needed for applications: if you are scheduling a task on a machine, either the operator is doing the task or he isn’t. But he cannot half do it. Nowadays, MILP solver technology is really impressive. It allows us to routinely solve problems with a million variables and a million constraints in just a few minutes.
What this means is that in practice, we are able to optimize almost any industrial process for which the data is available and for which the process is mature enough to be formalized with mathematics. So for instance: supply chains, manufacturing, financial services, communications networks, product scheduling, computer processes scheduling, designing microchips, and so forth. It helps us answer questions such as: how do you fill a plane, train or truck with parcels? Operations research can be applied to any industrial processes where you have to make decisions and manage risks.
- 6. Who are the typical champions of Operations Research at the organizations you work with?
In many companies, I would say you typically have a small operations research department with a few experts able to provide help on specific topics. And then, typically, there is a chief product officer from a different department that leads that product, but doesn’t necessarily know operations research that well. This person is going to design the specs and interact with the experts.
That’s the traditional model, but I am not certain it's the best one. The reason our project at Air France was such a huge success was because we had operations research engineers who were product chiefs and interacted directly with the operations team to understand their needs—and not only on the research and development part. If you want to truly innovate, it’s best if the people working on the model understand both the operations and the mathematics.
- 7. What are the key success factors to leverage applied mathematics in manufacturing operations? And who are the main champions?
There are two key elements of success. The first is that you must be able to solve the right problem. This means you need constant interaction between the operations and the mathematicians. And since operations often don’t know (and don't need to know) how the algorithm works, they just need a good user interface that shows the solutions corresponding to the right problems.
Secondly, you need good algorithms—not necessarily state of the art ones, but the ones that apply a technology that is understood by the tech team. Because if the developer doesn’t master the algorithmic technology, he or she will never come out of the debugging tunnel.
As for champions, it really depends on the market. For instance, the former secretary of the board of Air France was previously director of the operations research department. So in airlines, many people on the board understand these topics very well. In the energy sector some managers are at the crossroads of micro economics and operations research. Of course, that is quite rare.
What you also see is that MBA programmes at HEC and INSEAD have included awareness lectures on operations research. So many of the applications that come are pushed forward by chief financial officers or chief operations officers who maybe aren’t experts on the topic, but who were educated on the benefits of operations research at college and want to leverage them for their business.
- 8. Do you see a rise or decline in demand for operations research?
As I said before, the two prerequisites to apply operations research successfully are the maturity of the processes, and the availability of data. Of course, we also have much more powerful algorithms than before. What blocked progress before were really data and process maturity. The fact that you have much more data available today is the big game changer.
Overall, the need for operations research in the industry is really exploding. That’s why nowadays if you have a PhD in operations research, startups are going to come knocking on your door asking you to disrupt this or that market. Finding good talent is already a challenge, and that’s going to continue.
- 9. What is the most mature market for operations research today?
The kings of Operations Research are airlines. Why? Because an airline is basically a single huge process. Many of the most advanced operations research techniques have been developed for airlines. More generally, supply chains and logistics have been areas of huge success for operations researchers. It's always a question of networks, for instance energy suppliers who have to manage dozens, if not hundreds, of plants to address demand. These types of companies are the traditional sponsors of academic operations research teams, because that's critical to their survival. Without operations research, they are out of the market. Coincidentally, that's also the reason why, interestingly, airlines have been pioneers on revenue management issues, because they were very good at operations research. Now you have revenue management or yield management algorithms in many other industries.
- 10. Do you have an example of a company that predicted or responded quickly to Covid because of an advanced algorithm?
Building resilient operations has been a super fancy academic issue for over fifteen years. Everyone has been speaking about robustness, stochastic optimization, resilience and things like that. But you can imagine that now with the disruption caused by Covid, resilience in operations has become a hot topic—almost priority number one.
So what can we mathematicians do? It’s difficult, because you cannot predict Covid two years in advance. You can build resilience against traditional risk, but when there is a major disruption you can have the best algorithm in the world, but since the disruption is not in the data, it won’t be in the probability distribution.
So the second thing you need when the situation turns out completely different than anticipated, is algorithms that help you react extremely fast in real time. For example, even if their business suffered a big blow due to the pandemic, Air France was not so bad in adapting to the crisis because they have excellent operations algorithms that can simulate different scenarios. They managed to leverage decades of quality algorithms for simulating and optimizing their processes.
- 11. What would an advanced supply chain organization look like in 20 or 30 years?
Firstly, we will stop fire fighting. Secondly, we will try to automate all the non-strategic tasks with optimization algorithms to enable people to focus on the truly important strategic decisions that cannot be automated. Then, another thing that will be critical is resilience in real time. The way I see it, within 20 or 30 years, because of optimal processes and turbocharged mathematics, we will have a much clearer picture of the macro impact of each strategic decision made, online, in real-time, and on a much larger time horizon.
- 12. How far away are we from that scenario?
Academia has a role to play here. That's typically the kind of thing I'm working on with my postdoc students. Then we need firms like Pelico to spread the gospel of operations research. But even before that, we need to build the data infrastructure and the processes needed for algorithms to be useful. Depending on the market, it's already there or going to be there in the next three to five years. It's more a question on the maturity of the market from the process point of view. But I am absolutely sure that things are going to change in many markets very quickly.