ML and AI, Business Boosters for the Industry

Let me share a quick observation I did over the past few years about machine learning (ML) and artificial intelligence (AI). There are a lot of posts and news on the net presenting the shiny future of those technologies, thanks to self-driving car or AlphaGo projects that were in the headline news all over the world few months ago. And indeed the recent improvements are just extraordinary. More modestly I observed in several projects, a trend that may be much more revolutionary. This is a transfer of company and employee knowledge into algorithms.

Employee knowledge moving to algorithms

Let me illustrate what I mean with three examples.

• Condition monitoring: In the industry, until now the development of a solution for detecting wear and failures in parts using sensor data was a long process of may be several years where engineers were first understanding the physics of failures in the lab. After that, this knowledge was translated into programs using engineering techniques like signal processing. And tuning operations were a painful and long process. Now, we can reduce development time to few weeks using machine learning where classification and parameter tuning are automated and rely on solid mathematical concepts to guarantee performances.
• Visual inspection: Inspection of manufactured parts using images is now becoming more simple to develop as deep neural network or similar techniques can be used to learn what makes a difference between good and faulty parts. Here too, it is much less important for us to understand exactly what are the characteristics of faults because it is not necessary to develop the algorithm itself that will do the detection. ML or AI will do it automatically to some extend.
• Process optimization: In plants or warehouses with machines, employees, robots and conveyors, optimization is a very difficult task since systems are complex, partially unknown and could have a large variance in performances. Data mining, ML and more generally mathematical optimization can help process optimization. Algorithms can for example optimize picking in a warehouse without having engineers or managers to fully understand all the parameters and the logic behind the organization of the system. Same thing goes for a system corresponding to a physical/manufacturing process.

What we can observe from those examples and others is that it is no more necessary to understand all the detailed mechanics of the process we want to develop, monitor or control. Employees and engineers can focus on higher level tasks and let algorithms and AI to "understand" systems and tune the desired function that is been developed. This is a new equilibrium between human/employees and machine/algorithms: more know-how is transferred to algorithms while employees can focus on more productive tasks.

And this is not only true in production with robots. This trend is probably valid in many fields like:

• Engineers will spend more time on defining products and businesses rather than on tuning and developing them. Shift in development.
• Plant managers will spend more time in expending capabilities and adapting production to demand rather than controlling and tuning their manufacturing lines. Shift in operations.
• Insurance employees verifying customer invoices can spend time on cases with real and complex problems with the introduction of ML automated screening (fraud detection and process mining solutions) rather than checking cases with no fraud for sure. Shift in administration.

This means that development time can be much reduced and employees will focus on tasks that are much more profitable. We can spend more time on higher level tasks such as:

1. Focus on the problem definition and on the objectives
2. Make sure that algorithms and models fit the reality
3. Make sure we record as much data as possible

Record as much data as possible

Yes, to enable ML and AI techniques it is necessary to record and store as much as data as possible. Algorithms will learn from historical data. But the good thing is that it is now possible and cost effective to do it with the increase of the number of sensors (IoT) and the speed of networks, the introduction of costs effective data centers on the cloud and the development of scalable ML algorithms techniques.

Everything is ready to make this technology move happen

• Numerous and cheap sensors
• Network throughput
• Cost effective data centers
• Scalable ML and AI algorithms

At the end, ML and AI are fabulous business boosters

At the end of the story, ML and AI are really business boosters and specifically in the industry 4.0. Indeed, solutions where ML and AI can be used may have a combination of the following benefits:

• Shorter development time and lower costs
• Higher performance
• Higher level of innovation
• Better capability of adaptation
• Consolidation of know-how into algorithms (less impact of employee turnover)

Most probably more than ever it is now important for engineers, managers and directors, to monitor the time spent on low and high level tasks and have a look at ML or AI solutions because there are cases where it can save a lot of time and improve efficiency. ML solutions are extremely good at processing high dimensional spaces and large datasets, there is no need to do it for them.

Employees and ML/AI working together

To conclude on a more personal opinion, I feel very happy and satisfied as an engineer, that we are finally moving towards an era where it is possible to spend much less time on implementing ideas.

With a bit of exaggeration, processors and programming have been available since almost 50 years, and until recently we were still fighting with bits, bytes, simple arithmetic operations and low level libraries. Most of the development time were consumed for implementing ideas. But ML, AI and the explosion of Open Source community finally bring us technologies where we can implement solutions at much higher level and focus on objectives, constraints and functionality. This means more time for creating and testing ideas. Is it not what most of us are looking for?

Didier Nicoulaz

Data scientist services: www.adilem.com