Machine Learning in Manufacturing: a step-by-step guide


AllCloud Blog:
Cloud Insights and Innovation

In part 1 of our Machine Learning in Manufacturing series, we spoke about the value of ML and the cloud obstacles you may encounter, and how to overcome them. In part 2, we discuss how the cloud improves your overall manufacturing performance and processes for a more effective and error-free experience.

Cloud technology is clearly doing a lot to herald an era of incredible growth for companies, although It’s not as obvious how much the manufacturing and automotive industries are benefiting from the public cloud. 

Yet that is exactly where products, processes and machines are being developed to create the smart factory, with the focus on data. Data can be used within the machine learning (ML) context for machine vision or predictive maintenance, but also to help companies make strategic business decisions and answer previously unanswerable questions, such as “Which machines deliver the best performance?”, “Where should I invest more resources?” or “How can I tackle machine idle time?”. And, for the first time, the technology can also answer open questions that are very vague, including: “How can I improve this process?”

Using the cloud instead of planning and setting up your own infrastructure

The ability to access previous unknown and unused data and leverage it with machine learning is one of the drivers for Industry 4.0. Fortunately, companies no longer need their own infrastructure in order to store and analyze data. They can collect it using the cloud computing provider AWS and use it there directly, helping them benefit from the huge potential of machine learning with no need for their own expensive data center. 

More and more companies are already creating data lakes with AWS in which they aggregate all of their data, whether structured or unstructured. Even if they have no particular use for this data today, it will be ready and actionable for the ML-based innovations of tomorrow. 

Manufacturers interested in data lakes and ML should be prepared having already completed these two steps in their digitalization journey: the cloud migration of their IT, and the networking of their machines. Once these milestones have been achieved and the data lake starts to take shape, the next thing to do is ensure the quality of the data.

The basis for success: well-curated data 

ML needs to be based on clean data to work properly. The data must be available in digital format and be complete, as well as error-free. After all, algorithms can only reliably recognize and predict what was taught to them beforehand. 

It is still the case that nobody is spared the task of curating and preparing the data. If a company’s data has not been prepared properly, its data scientists will have a lot of work to do to curate it. It is not uncommon for 80 percent of the ML project effort to be invested in data preparation. Unfortunately, algorithms can’t help here as their work depends on data quality and correctly annotated data. Against this backdrop, it is particularly important to define responsibilities for the data, so there needs to be a data owner with a view to data governance.

Data usage made easy

ML solutions can be integrated once the cloud migration is complete, all the machines are networked, and the data lake has all clean, high-quality data ready. However, many companies don’t have the data scientists to do this. That’s why they should take a look at end-to-end solutions. 

AWS, being customer-centric, has recognized this need and provides ML solutions in the form of “low code/no code” approaches, aligned with a range of industrial use cases. This allows companies to implement ML solutions quickly and they have less of a need for data scientists. AWS now has services in its portfolio that integrate practical experience from multiple projects, such as object recognition. 

Quality management with machine vision

Amazon Lookout for Vision is a no/low-code solution for anomaly detection based on computer vision in image and video data. The ML service supports the recognition and prevention of defects and product manufacturing problems. This significantly accelerates quality management processes in particular and enables the automation of previously manual inspection procedures. Aside from defects in materials or surfaces, the technology also detects things like missing components in modules.

The highlight: Lookout for Vision is extremely user-friendly and can even be used by staff who have no training as data scientists. Companies start by designing a GUI or API for this and uploading the images. With some of the Lookout for Vision services, customers can start with as few as 30 images. Users can then indicate which of the pictures features good or bad quality. The model is then trained automatically based on this input. The service takes over the time-consuming decision as to which parameters and algorithms are best suited to the task. It also includes the integration of data pipelines and data transformation and although this costs a little more, it often enables companies to do without a data scientist for this work, so that their expert is free to focus on business growth tasks.  

Once Lookout for Vision goes live, the anomalies it detects are shown in an easy-to-read dashboard and the production manager can then react quickly. The service is also fully managed, so maintenance and upgrades are provided automatically. And when new versions are released, they are integrated seamlessly. 

Step by step to your goal

The path to utilizing ML solutions will be shorter or longer depending on the manufacturer’s level of digitalization. Many companies suspend or scrap the idea, as is underscored by the countless abandoned ML and data projects along with PoCs. Companies need to have achieved several milestones before they start: cloud migration, networking of the machines, data lakes containing good-quality data, and only then can they integrate ML services.

Where and how should companies start? There’s an African proverb that says: if you want to go fast, go alone; if you want to go far, go together. That’s the case here too. Companies can significantly speed up their digitalization journey by working with reliable partners; they can even gain ground on competitors that might have started earlier but have lost momentum due to the lack of external expertise.   

When choosing a partner, customers should not only look at the size of the company but also focus on its level of expertise. AllCloud’s Data Practice has many years of experience in strategic planning and implementation for manufacturing companies that want to leverage digital transformation in the cloud.

Contact us today to learn how our AllCloud AWS data experts can help you improve your manufacturing processes and use the cloud and ML as a catalyst for organizational growth.

 

Carsten Riggelsen

Head of Data & ML/AI

Read more posts by Carsten Riggelsen