11 June 2019, 13:41
The world is rapidly moving toward Industry 4.0 or the Fourth Industrial Revolution, where artificial intelligence (AI) and machine-learning based systems are not only changing the ways we interact with information but also revolutionizing the manufacturing sector. Industry 4.0 is happening at a rapid pace, which makes it hard for manufacturers to change their processes and keep up with the technological development. A lot of people think that AI is the future, but the truth is that AI is happening now!
This blogpost is part of a series of blog posts on manufacturing where we cover topics such as ‘Boosting your manufacturing process with Artificial Intelligence with minimum investment’, ‘Smart real-time dashboarding with IoT data’ and ‘5 lessons I learned about large scale IoT’.
In this first blogpost, we will illustrate how you can enable AI in your manufacturing facility using minimum investments. First some major concerns in the industry are addressed, after which guidelines are given on how to select a successful AI project. Finally, based on these guidelines, some sample use cases are given.
Why is AI not fully integrated in manufacturing systems?
Historically the manufacturing industry is a very conservative one. Historical trends show that manufacturers focus mainly on increasing precision, flexibility, complexity and speed while trying to decrease costs and use of resources. When talking about Industry 4.0 a lot of people think about the factory of the future where everything is interconnected, orchestrated and self-learning.
The truth is that most existing manufacturers don’t have the infrastructure to allow this to happen. Therefore, most companies fear high capital investment and process redesigns. There is desire to innovate but risks do exist, similar to other aspects of business.
According to an AI report from Infosys, most companies want to automate manufacturing to increase productivity (66%), minimize manual errors (61%), reduce costs (59%) and refocus people’s efforts on non-repetitive tasks that benefit from human intervention (50%).
So how can we tackle the problem we have in a conservative market?
Although every company is unique and there is no single path to drive AI transformation, there are some guidelines to allow AI to contribute to your manufacturing processes. The manufacturing world can learn a lot from other industries where AI has been successfully applied. It might seem odd but search engines, predicting prices in e-commerce, AI beating players in a game or the software of autonomous cars are very closely related to most manufacturing processes. Although the application is totally different, the mathematical problem which needs to be solved is very similar. This allows us to use algorithms which are used in other sectors and apply it to the manufacturing process.
Where you can apply AI with minimum upfront investments while obtaining fast results? Three rules of thumb.
It is likely that your company is saving data for many years and you can use these logs to obtain a huge amount of data without redesigning the processes.
You can train an AI algorithm more easily when you know what you want to optimize and you measured the desired outcome. Most AI techniques require a labelled dataset, which means that next to your input data you need to know the output.
Uncertainty means room for improvement. Uncertainty in manufacturing can be due to several reasons including, a.o.:
So where can we apply these techniques?
There are several applications where you can apply AI in the manufacturing industry, e.g. quality control, predictive maintenance, predicting end of lifetime, supply chain and demand optimization, inventory optimization and energy reduction. As are further described here and here. But I would like to demonstrate the three rules of thumb by two use cases.
One of the first steps in the steel, food, pharmaceutical or rubber industry is mixing certain ingredients to obtain a final product with the right material properties (similar problem statement as our previous blogpost here). These properties depend highly on the quality of the input material, the parameter settings and process variations, the mixture of components etc. This is a typical process in which there is a lot of internal and external uncertainty that can influence the final material properties. The process complexity results in a suboptimal solution and the manufacturer introduces safety factors to ensure a stable product. In most cases, the process is well established, repetitive and monitored by sensors for temperature, pressure, speeds etc. Therefore, it is likely that there is a lot of structured data available. Finally, the material quality is measured by inline sensors, sample lab tests and/or by visual inspections. It is important, and often forgotten, that the quality of the input material is also measured and well documented because it is likely that this will have a high correlation to the output material quality.
AI can be directly applied to predict e.g. the material properties using regression models or visual defects using deep learning algorithms without the need of capital investments or process redesigns. Next AI can give recommendations to the operator on how to change the settings of the machine. Moreover, AI can give a designer of a R&D department more insight into the process and can enable higher quality materials. Using real-time IoT data in combination with self-learning algorithms results in parameter prediction for optimal output while meeting the quality requirements and reducing resources and emissions.
Countless parameters in a manufacturing process can be mapped and optimized. Doing this manually is a tedious job and basically impossible due to the amount of possibilities and varying conditions. These uncertainties make predicting the desired output a difficult task. In most cases, these processes are well monitored by sensors and the desired outcomes are measured. Sample use cases are the improvement of a powertrain on an electrical bike and reducing the energy consumption of a datacenter.
To avoid high capital investments it is advised to first start by predicting the output, e.g. battery life, energy consumption. Next, you can create a so-called recommendation engine to make recommendations of process parameters. Finally, you can implement a full autonomous parameter optimizer allowing a self-learning system to find the optimal solution in every circumstance.
. . .
AI can make a significant contribution to manufacturing by faster action and decision making, innovative product, services and material designs, improved efficiency, higher precision and lower costs. It is time for a change. You can enable AI without the need for high capital investments or major process redesigns. You just need to look for the right opportunities to be more productive and sustainable. It is time to take your company to the next level, it is time for AI!