Machine learning and artificial intelligence have enjoyed lots of interest across industries, but adoption in manufacturing has been slower because of process complexity and the need for customization performed by systems integrators. A bunch of startups have sprouted up to tackle this challenge, including Oden Technologies, which earlier this month announced an end-to-end machine learning (ML) and AI framework designed specifically for manufacturing.

The company said its infrastructure spans both the cloud and edge, letting manufacturers “deploy mission-critical ML and AI applications to avoid machine failure, eliminate waste, and optimize production in real time.”

The production-ready AI framework integrates algorithms and data science tools with structured and unstructured data from machines, operator inputs, QA, work orders, environmental monitors, and product specifications, Oden said.

The hybrid infrastructure between cloud and edge computing “delivers the power of the cloud without compromising the requirements of mission-critical applications,” it said.

With the framework, manufacturers can monitor their processes to predict quality and machine health, detect abnormal behavior, and provide recommendations on process settings.

In addition, they can use the AI framework to rapidly prototype and test new ideas for monitoring and optimization.

“Today’s launch marks the first step to delivering intelligent industrial automation to our customers,” said Willem Sundblad, co-founder and CEO of Oden Technologies. “The future of manufacturing demands intelligent systems that can provide real answers to the production issues manufacturers have been asking for decades. Only then can they achieve perfect production, with zero waste”.

The company said its customers saw improvements such as a 20% increase in monthly output and 50% decrease in total scrap, resulting in savings and additional revenue.

Earlier this year, Oden closed a $10 million Series A investment from existing investors, including EQT Ventures and Inbox Capital. Niklas Zennström, the CEO of Atomic CEO and co-founder of Skype, joined the company’s board.

Robotics Business Review recently spoke with Sundblad about the new AI framework and issues of adding ML and AI to manufacturing processes.

The difficulties of adding AI to factories

Q: Why has it been so difficult for manufacturers to adopt machine learning technologies at their businesses?

Sundblad: There are many reasons, but we’ll list a few. Their current systems and processes are not set up to handle this type of data processing and analysis. The current staff rarely has experience working with machine learning.

Also, there are two sides to the digital transformation journey that have to happen at the same time in order for it to be successful. One is upgrading the systems to be able to handle the data and do this type of analysis.

The other one is training the people on what these systems are and how they should use them. That’s why looking at it as a journey, with clear goals and outcomes of each phase, allows manufacturers to start transforming while getting incremental ROI on each step of the journey from data acquisition and analytics to ML and AI.

Q: What kind of machinery and systems does your platform work with? Lots of systems are proprietary, so how can companies get data out of their systems into your platform without a lot of customization or integration effort?

Sundblad: Oden’s IoT devices are compatible with almost every machine sensor available on the market. Our goal is to get the base data into a single source of truth – our software platform – before starting to add intelligence to the system. The data is fundamentally the intellectual property of the customer, not the machine builder or the system.

Even though some systems or machines are more difficult to integrate with than others, there’s always a way to achieve the desired outcome. We always look for the path of least resistance while ensuring the validity and trust in the data.

Q: Is it easier to work with a specific vendor’s systems versus others?

Sundblad: Absolutely. Certain systems and protocols may provide better documented APIs on the software side or standard protocols like Modbus or OPC communication.

Q: Beyond predictive maintenance and monitoring, are there other areas that benefit from your AI framework? What are the applications that clients are asking for? Or are they just looking to make sure their machines don’t break down?

Sundblad: Beyond predictive maintenance, our primary use cases have been improving production process and recipe optimization. Our platform helps manufacturers hone in on how the product is made, predict offline quality results, and recommend better settings for the operators.

Q: Is there a particular market segment or vertical that you’re working with initially? Does this work better in some industries other than others?

Sundblad: Our technology is fundamentally horizontal and can ultimately be applied across a variety of manufacturing verticals.

The first market we’ve focused on is the plastics manufacturing industry, and we’re seeing great results there. The manufacturers that experience the highest returns are the ones really driving optimizations and actively working to improve their processes.

When applied by manufacturers who focus on producing high quality products, our technology really shines, because these manufacturers really care about optimizing their processes to serve their customers better.